From Space To Time: Temporal Adverbials In The World'S Languages

Introduction

and neither can be reduced to the other. But space and time seem to show a peculiar relatedness that is perhaps not evident to a naive philosophical observer: Human languages again and again express temporal and spatial notions in a similar way, as for instance in E1-3.

This view is again confirmed in the present work. It would be quite impossible to give a purely notional definition of the expressions which fall in the scope of the investigation. Notionally we are concerned with temporal qualifications of situations (answering 'when?' questions), in particular expressions that serve to locate situations in time, expressions that measure the temporal extension of situations (answering 'how long?' questions), and expressions that indicate the regular recurrence of situations (answering 'how often?' questions). But it would hardly make sense to include all expressions in these notional domains in the cross-linguistic investigation, because the phenomena would be formally quite heterogeneous. Consider the boldfaced portions in E9a-f, a small selection of expressions serving to locate situations in time.

E9. a. I visited my uncle in the spring.

b. While the government prepared the attack on Jaffna, the Tamil Tigers deported the population from the town.

c. Much later he realized that she had been right all along.

d. Coming home, he immediately began to play with the kids.

e. World War II was followed by a 45-year period of "Cold War".

f. The simultaneity of these two conferences makes it impossible for her to attend both.

As these examples show, temporal location may not only be expressed by NPbased adverbials (in the spring), but also by tense (E9a), temporal adverbial clauses (E9b and E9d), adverbials based on adverbs or adjectives (E9c), by verbs (E9e), and by nouns (E9f).

A typological study of such a diverse set of phenomena would hardly be fruitful, and has in fact never been undertaken, even by those linguists who claim that typological studies must be based on purely functional definitions.

Thus, COMRIE's (1985:9) definition of tense as "grammaticalized expression of location in time" does not correspond to the much narrower set of phenomena that he goes on to treat in his book, although he adds the non-notional, formal condition "grammaticalized". Under all reasonable definitions of "grammaticalized", the preposition in in in the spring, or at the very least the Latin Ablative case in vere 'in the spring', would qualify as grammaticalized, but COMRIE does not discuss such expressions in his book on tense. Thus, the definition must be made more specific, e.g. "grammaticalized expression of location in time on verbs", or perhaps, if we want to include the nominal tense that has occasionally been reported, "grammaticalized expression of temporal location of a situation, marked on the main word expressing this situation".

Whatever the more precise formulation, it must contain another crucial formal condition.

The formal condition on the temporal expressions that are in the scope of this study is that they must be adverbials based on noun phrases. The qualification "based on noun phrases" excludes adverbial clauses and adverbials based on adverbs. Adverbial clauses are of course often related to NP-based adverbials, and in many languages adverbial subordinators and adverbial adpositions overlap to a large extent. On the basis of English words doing double duty (e.g. after her arrival/after she arrived), JESPERSEN (1924:89) goes so far as to deny the theoretical distinction between prepositions and adverbial conjunctions. But not all languages show the same degree of overlap as English, and in any case including adverbial conjunctions would have extended the scope of this work dramatically. In addition, if there is a close parallel between adverbial adpositions and adverbial subordinators, the subordinators are generally derived from the adpositions (cf. KORTMANN 1997: §5.2.2), so that it is easier to exclude subordinators in a study of adpositions than vice versa.

On the other hand, we must exclude adverbials based on adverbs. Adverbs like English now, then, when?, yesterday, tomorrow, afterwards, Albanian vjet 'last year', Modern Greek apópse 'this evening', German heuer 'this year' are common in all languages, and they are probably more frequent in texts than more complex NP-based adverbials such as three weeks ago or on a Sunday morning. But since they are essentially indivisible lexical items, they largely fall outside the scope of grammatical typology. 3 Temporal adverbials based on adjectives (earlier, previously) must also be excluded, but they are not common anyway.

This leaves us with noun phrases serving as time adverbials, such as every morning, last Friday, and adpositional phrases, such as in the winter and three hours ago. The reason why I throw these together into the single category of "NP-based" adverbials (instead of using the conjunction "adverbial NPs and PPs") is that there is a continuum from adverbial NPs to adverbial PPs. This continuum is not visible in a morphologically impoverished language like English, but many languages have adverbial noun phrases marked by various oblique cases, e.g. Hungarian kedd-en 'on Tuesday', január-ban 'in January', kilenc óra-kor 'at nine o'clock'; Korean yelum-ey 'in the summer', welyoil-kkaci 'until Monday', cinan cwu-puthe 'since last week'. Adpositions and case markers 3 Of course, a lexical-typological study of expressions for 'yesterday/today/tomorrow', 'last year/this year/next year' etc. would be very interesting and might yield implicational universals such as "If a language has a special adverb for 'last year', it also has a special adverb for 'last day', i.e. 'yesterday' ". But in this book my topic is grammatical typology.

are on the same synchronic grammaticalization path, and due to the gradience of grammaticalization there are bound to be unclear cases. The notion of "NPbased adverbials" has the advantages of being non-disjunctive and simultaneously avoiding the artificial issue of distinguishing between oblique NPs and PPs.

After this preliminary discussion, it is now time to give a first list of the major semantic sub-types of NP-based temporal adverbials, shown in Table 1.

Each of these semantic functions, as I will call them, is discussed in greater detail from a semantic point of view in chapter 2.

At first glance, simultaneous temporal location looks semantically quite simple:

The located situation is simultaneous with the reference time, which is either a point in time (e.g. five o'clock, our arrival at the summit) or a time span (e.g. the summer, the federal budget negotiations). Spatial models for simultaneous location are readily available: For reference time points, the one-dimensional spatial meaning 'at (a point in space)' is available, and for reference time spans, the two-dimensional spatial 'on (a surface)' or the three-dimensional spatial 'in (a container)' can be used. And indeed, this is the option that languages overwhelmingly choose: They transfer their simplest spatial markers to temporal noun phrases to denote simultaneous temporal location. Thus, the English expressions at five o'clock, on Thursday, in January are very typical.

There are only three languages in my sample, Lithuanian, Swahili and Abkhaz, which do not use spatial markers in any of the simultaneous functions. But even these languages make only limited use of non-spatial markers: Swahili and Abkhaz mostly show zero marking, and Lithuanian mostly has accusative case marking. Thus, they are fairly weak exceptions to the general rule.

Languages commonly restrict the application of their spatial markers to NPs headed by specialized temporal nouns, i.e. nouns denoting canonical time periods (plus a few others, such as 'time', 'beginning', 'end'). Thus, we can say in the winter, in the morning, at Easter, but not *in the federal budget negotiations, *in the soccer game (at least not in the temporal sense). 1 I have not made a systematic study of this restriction, but it appears that most languages require auxiliary nouns like 'time' when expressing simultaneity with an event denoted by a normal action noun, e.g. in Russian vo vremja peregovorov 'at the.time of.the.negotiations' (contrasting with *v peregovory 'in the negotiations').

Several European languages have a special preposition for this purpose 1 These expressions are possible in a non-temporal sense, where involvement in the action is implied. The contexts provided in (i)-(ii) show that only during is possible in English when a purely temporal sense is intended. (English during, French pendant, Italian durante, German während), but I have not found a similar special marker in most other languages, so I did not distinguish a separate semantic function for them.

Thus, the only simultaneous temporal adverbials investigated in this study are those that are based on the canonical time periods, i.e. time units (hours, days, months, years), calendar unit names (April, Thursday), and the qualitative periods of the day (morning, evening, etc.) and the year (spring, summer, etc.), as listed in Table 3 The enormous variation shown by these four languages and English is sufficient to make one skeptical of the approach advocated in WIERZBICKA (1993). She argues that a common meaning can be found for at in at five o'clock, at noon, at the beginning, at night, at the beginning, and for on in on Thursday, on the first night. This kind of in-depth semantic analysis of a single language can certainly be insightful, but it does not throw much light on the cross-linguistic variation.

In particular, it would lead us to expect that the English pattern of simultaneous markers should repeat itself in other languages (which is not the case). Of course, it is possible that different languages conceive of different time periods in different ways, but it is equally possible that a lot of these markings are fairly arbitrary conventions.

The goal of this chapter will be to discover a few cross-linguistic regularities in these diverse patterns. I have looked systematically at the expression of simultaneous location with reference time nouns of the seven types shown in the examples above.

E1. a. (orientation)

The priest stood before the altar.

b. (sequence) St. Michael's day is before Christmas.

E2. a. (movement)

Pepito is going to the village to help his granny.

b. (future)

The rain is going to help the farmer.

E3. a. (extreme part) We are still far from the end of the queue.

b. (last moments) You will be tired at the end of the day.

This phenomenon is so widespread in different languages across the world, and in different parts of the vocabulary, that we have to conclude that space and time are linked to each other in human thinking as well. One common way of conceiving of this relationship is by saying that temporal expressions are based on spatial ones, and that the transfer is a kind of conceptual metaphor (e.g. LAKOFF & JOHNSON 1980, CLAUDI & HEINE 1986).

That more abstract domains of language (and cognition) may be modeled on the spatial domain is an old insight, which goes back at least 150 years (see HJELMSLEV (1935) on the localists of the 19th century), and probably much further. But it is only fairly recently that linguists have begun the systematic study of the world's languages in order to verify whether the transfer from space to time is limited to languages of a particular cultural sphere (Europe) or a language family (Indo-European), or whether it is a widespread, perhaps universal phenomenon, found across the globe in languages of diverse families.

Such systematic typological investigations became more urgent when the old assumption of universality was challenged. In particular, Benjamin Lee Whorf, the famous student of Hopi and author of the "linguistic relativity hypothesis", claimed that Hopi (a Uto-Aztecan language of Arizona and New Mexico) does not show the metaphor from space to time: "The absence of such metaphor from Hopi speech is striking. Use of space terms when there is no space involved is NOT THERE -as if on it had been laid the taboo teetotal!" (WHORF 1956:146).

For the grammatical marking of time on verbs, i.e. the domain of tense and aspect, there is now a sizable body of cross-linguistic research which shows, among other things, that the use of spatial periphrastic expressions is by no means geographically, genetically or typologically limited (cf. ANDERSON 1973, TRAUGOTT 1974, 1975, 1978, DAHL 1985, BYBEE et al. 1994). Of course, not all temporal and aspectual expressions are based on spatial ones, but on the basis of the large-scale cross-linguistic surveys we now have a much clearer picture of their distribution.

But the use of spatial expressions for temporal notions is even more salient in temporal adverbials which relate a situation to a reference time expressed by a noun phrase. In E4-7, some examples from different languages are shown. In all these cases, a preposition is used both in a spatial and in a temporal sense. 'We'll leave for Pavia in ten months.'

Such prepositional temporal adverbials have not been investigated systematically across languages yet. This book is devoted to their study. I examine the most important grammatical markers expressing such adverbials (i.e.

adpositions and cases) in 53 languages from around the world, hoping to contribute in this way to the larger problem of the conceptualization of time through language.

The data confirm the universalist's expectation that spatial expression of temporal notions is extremely widespread in the world's languages, being limited neither genetically (e.g. to Indo-European), nor geographically (e.g. to Europe), nor typologically (e.g to languages with SVO word order). In this sense, the transfer from space to time can be said to be universal.

NP-based time adverbials

Not all temporal adverbials based on noun phrases are straightforward metaphorical extensions from spatial adverbials. This is clear from examples like E8 from English, where markers are used that have no corresponding use in spatial expressions: the prepositions after and for, the postposition ago, and the use of a bare NP (indicated by "Ø" in E8d).

E8. a. After the wedding, the couple went to the Baltic Sea coast for their honeymoon. b. Peace was concluded finally three weeks ago. c. Jacob served his father-in-law Laban for fourteen years.

d. Most trees bear fruit Ø every year.

Nevertheless, such non-spatial markers have been included in this study, because it is only by way of contrast with non-spatial markers that we can appreciate the role of spatial markers for temporal adverbials. Similarly, tense and aspect forms that are not based on spatial metaphor were included in cross-linguistic studies, because we need those other forms as a background.

Thus, the present book is intended as a study in the tradition of partial typology, where one limited area of grammar is studied in a large number of languages with the goal of discovering cross-linguistic generalizations. The main goal of this work is to assemble cross-linguistic evidence for the hypothesis that temporal notions are conceptualized in terms of spatial notions, but in addition I discuss a fair number of additional points that arise in connection with the data.

As far as I can tell, this book is the first typological study of NP-based time adverbials. 1 As I mentioned above, typological investigations of tense and aspect have already been undertaken (DAHL 1985, BYBEE et al. 1994), but so far nobody has looked in detail at grammatical marking of time through noun phrases. It is perhaps natural that linguists should have focused on the marking of time on verbs first, because tense and aspect are generally obligatorily expressed in every sentence and are therefore much more salient than temporal relations expressed by adverbials.

Furthermore, because of their generally primary nature with respect to temporal adverbials, spatial adverbials are more salient, and spatial markers have already been the subject of a systematic typological study (SVOROU 1994).

And finally, another area of grammar that is adjacent to my topic are temporal adverbial clauses. These have also received considerable attention from linguists, perhaps because of their greater complexity when compared to NP-based time adverbials. However, I know of no systematic cross-linguistic study of temporal clauses, although there are typological treatments of adverbial clauses in general, including temporal clauses (cf. THOMPSON & LONGACRE 1985, KORTMANN 1997, HENGEVELD 1997). Thus, NP-based time adverbials have so far been upstaged by tense and aspect, spatial adverbials and temporal adverbial clauses, but the program of partial typology will remain incomplete until all areas of grammar are illuminated by the crosslinguistic point of view.

The fact that this study is the very first attempt at a typological investigation of time adverbials expressed by noun phrases also determines some of the features of the work. First, my goal is a broad survey of the phenomena, not a detailed examination of particular problems. Thus, I study a fairly wide range of temporal relations that can be expressed by NP-based adverbials, rather than focusing on a few select ones. Second, I did not attempt to construct a biasfree world-wide sample of languages as the data base of my investigation.

1 KUČERA & TRNKA (1975) present a very thorough study of time adverbials in three languages (Czech, Russian, English), but their main concern is with cooccurrence restrictions rather than with the typology of form-meaning pairings.

While representative samples are certainly desirable in principle, they are not yet a realistic goal for many areas of language structure because there is simply not enough information available. This is certainly true of NP-based temporal adverbials: While probably all grammars have something to say on tense, aspect and spatial adverbial markers, many grammars are very incomplete with respect to NP-based temporal adverbials. Thus, my generalizations are based on a sample of fifty-three languages in which all continents are represented, but which is heavily biased toward European languages. It simply did not seem reasonable to me to exclude languages about which relevant information is readily available only in order to have a balanced sample, which would then have to be much smaller. But of course we have to keep in mind that from this sample we do not get a picture that faithfully reflects the situation in the languages of the world. Thus, the present work must be seen as a first approximation to the typological study of NP-based time adverbials, which should be followed by a more balanced study that truly reflects the current linguistic diversity on our planet.

In the remainder of this introductory chapter I will first give a definition of the subject matter of this study ( §1.3), followed by an overview of the secondary goals that I hope to reach ( §1.4). Section 1.5 discusses the criteria for determining the main semantic sub-types of time adverbials around which the presentation will be organized, and §1.6 deals with the sources of my data, especially the sample of fifty-three languages. In §1.7, I discuss a number of views on the relation between space and time in language that are found in the literature, and I conclude this chapter with some thoughts about the mapping of spatial structure onto temporal structure.

Definition of the domain of inquiry

The domain of inquiry of this book consists of adverbials based on noun phrases which serve as temporal qualifications of situations. 2

As I have argued elsewhere (HASPELMATH 1997), studies in partial typology must be based on mixed functional-formal definitions, i.e. the phenomena that are compared across languages are delimited by both functional (or semantic) and formal conditions.

Subsidiary goals of this book

The formal expression of each of the sixteen semantic functions of Table 1 has been investigated for the fifty-three languages of the sample, with the purpose of uncovering generalizations in the data that inform us about the way in which human language in general, and hence human cognition, structures the conceptual domain of time. As was made clear at the outset, the results strongly confirm the initial hypothesis that temporal relations are based on spatial relations in the large majority of cases. This result is of course what we expected from the beginning, but along the way quite a few other interesting observations are made, and a number of subsidiary goals are pursued.

First, we need to determine which temporal relations of NPs are expressed at all by grammatical means in languages, i.e. which conceptual distinctions are commonly made in the grammatical sub-system of language (cf. TALMY 1988 for the general research program). As elsewhere in the domain of grammatical semantics, the list of concepts expressed by grammar is quite limited. Those distinctions that recur reasonably often have been included in the list of semantic functions in Table 1, and some further subdivisions are discussed in later sections (cf. §1.5 for discussion of how these distinctions have been isolated). There are also some semantic distinctions that have not been included

in my cross-linguistic study but that need to be recognized from the point of view of universal grammatical semantics; they are mentioned briefly in §3.2.

My second goal is to provide some guidelines for the investigation of NPbased adverbials in individual languages, especially in newly described languages. Many fieldworkers have made the experience that a knowledge of the attested space of variation in other languages is of great help for charting the grammar and lexicon of a new language. And this is true not just for exotic languages -even in the study of well-known European languages, a look across the fence at what other languages do is often helpful for a deeper understanding of a phenomenon. In order to make such comparison possible, a first requirement is a terminological grid that can be applied independently of the language. A complete terminology for time relations of NPs is provided here for the first time in a theoretical work. 4And third, as in any cross-linguistic work, I have been looking not only for confirmation of an absolute universal (NP-based temporal adverbials are not limited genetically, areally or typologically), but also for possible typological connections, i.e. implicational universals. However, there is generally no connection between the expression of NP-based time adverbials and other parts of the grammar, 5 so there are few concrete results in this respect. Of course, in languages with grammatical and spatial prepositions, temporal NP markers will also tend to be prepositions, whereas languages with spatial postpositions will also have temporal postpositions; languages with rich morphology are more likely to have a terminative case (rather than an adposition) than isolating languages; and so on. But these are generalizations that pertain to the purely formal side of language structure, and a different kind of study would have to be undertaken to pursue these issues. In this study I will focus on those generalizations that relate to the temporal meaning of the markers and constructions in question. Of course, it is theoretically possible that semantic properties cluster in a way similar to morphosyntactic properties, so that we could distinguish, say, languages whose temporal relators are systematically based on spatial relators, wheres this is not the case in another class of languages (as is implied by WHORF's (1956) hypothesis that Hopi differs from Standard Average European in this respect, cf. §1.1). However, I have found no evidence for such a hypothetical semantic typology. Different semantic functions show a greater or lesser tendency to be based on spatial terms, but different languages do not. 6

Criteria for isolating the semantic functions

The list in Table 1 contains the major semantic functions of temporal NPs that recur in languages, and it will be taken as a point of departure for the organization of this work. Thus, it becomes important to give the criteria which have guided me in including conceivable semantic distinctions in the domain of temporal adverbial qualification in this list.

A semantic function has been isolated when there is a significant number of languages which clearly distinguish this type from related ones in their means of expression. Thus, the main criterion is a typological one, not a semantic one.

It would be very difficult to base such a list on semantic criteria alone, because then there would be no way of constraining the possible proliferation of 5 An interesting exception is discussed in §6.1 (E81-87), where word order typology appears to have an effect on the expression of temporal distance markers. 6 Whorf's claims regarding Hopi time expressions have been refuted in a comprehensive manner by MALOTKI (1983). 'Before his death Mitterrand asked to be buried in Jarnac.' c. Rebecca wurde vor Konradin geboren.

'Rebecca was born before Konradin.' d. Thomas ist vor einem Jahr nach Cambridge gegangen.

'Thomas went to Cambridge a year ago.'

A priori, these four specific uses could be assigned to one, two, three or four different semantic functions. In fact, I have set up two different semantic functions for these uses, anterior (E10a-c) and distance-past (E10d). Why are E10a-c collapsed in one semantic function? From a purely semantic point of view, E10a and E10b could easily be distinguished: In E10a, the time that has elapsed between the main event and the ice age is much longer than in E10b, so one might distinguish a remote anterior from a recent anterior (much like in the literature on tense, cf. e.g. COMRIE (1985:Ch. 4) on remote past tenses). But to all appearances, a distinction along these lines is made very rarely in the world's languages (an example is the Russian distinction between do and pered, see §4.4). An even more obvious semantic distinction is that between E10a-b and E10c: In the former, the NP governed by the preposition vor denotes an event, whereas in the latter, it denotes a person. Thus, vor clearly has a very different semantic interpretation in E10c, which is best described by a clausal paraphrase (Rebecca wurde geboren, bevor Konradin geboren wurde). A priori, we might expect that some languages would have different expressions depending on this semantic distinction. However, I have not found a good example of such a distinction. Thus, there is still no reason to set up a separate semantic function for E10c.

Matters are different with E10d. Again, the NP governed by vor denotes not an event, but a time span, and the semantic contrast between E10d and E10a-c is readily felt. Like E10c, E10d can be paraphrased in a way that makes its semantic structure transparent: Thomas ist ein Jahr vor dem Sprechzeitpunkt nach Cambridge gegangen 'Thomas went to Cambridge a year before the moment of speech'. But in contrast to E10c, E10d is expressed by a different marker in many languages, e.g. in English (ago in E10d, before in E10a-c). There are also many languages that express E10d in the same way as E10a-c, so German is not at all exceptional in this regard. Still, the fact that many other languages are like English means that this semantic distinction is highly relevant for a typological study and is therefore given the status of a separate semantic function.

The foregoing discussion naturally leads to the question of how to describe the various uses of this preposition, in terms of homonymy, polysemy or vagueness. Given the paraphrase relation between the 'before' and 'ago' senses of vor, we can probably exclude the first option, homonymy -there is little doubt that 'ago' is semantically related to 'before' (see the discussion in §6.1 below). Polysemy and vagueness are more difficult to distinguish (see, e.g., GEERAERTS 1993), but in the light of the discussion above one might be tempted to propose that the typological criterion of cross-linguistic distinguishability is also a strong argument for polysemy rather than vagueness at the level of an individual language. This, however, is clearly not the case, as has been extensively shown for other grammatical domains. Thus, nobody would say that the Russian past tense is polysemous rather than vague between the simple past and the perfect reading just because the perfect meaning is a semantic function that is widely distinguished in the world's languages. I will give just two examples from the domain of NP-based time adverbials.

Consider the German sentence in E11, which has two readings, corresponding to the English sentences in E12a-b (cf. KÖNIG 1974:554 for some discussion).

E11. German

Die Kneipe wird bis zehn Uhr offen sein.

E12. English a. The pub will be open until ten o'clock.

b. The pub will be open by ten o'clock.

As will be discussed below in §5.5, there are quite a few languages that make the English distinction between 'until' and 'by', so by the cross-linguistic criterion there should be two separate semantic functions for these two meanings.

Nevertheless, there are very good reasons for saying that E11 is simply vague with respect to this distinction, because a semantic analysis that covers both interpretations is possible. An even clearer example is the following contrast between English and Finnish:

E13. Finnish The interpretation of in is so different in these two cases that it is not obvious that there is a common element of meaning. But German is paralleled by other languages, including unrelated ones, in showing the same marker in these two cases (cf. §6.2.2 below), so the use of in in these two different meanings cannot be due to an accident. The cross-linguistic facts force us to look harder for a possible analysis in terms of polysemy. However, it must be admitted that cross-linguistic replicability is still not a hard and fast proof of polysemy. It could still be that synchronically speakers no longer perceive a relation that has existed earlier, and that the cross-linguistic similarities are due to common diachronic paths. Typological studies can exclude accidents and can demonstrate the existence of facts that need to be explained, but these facts are not necessarily synchronic. Given this bias of the sample, it is clear that quantitative statements should be treated with great caution -a sample that is as little representative of the world's languages as this one simply does not allow such extrapolations. But on the other hand, the sample contains languages from most major regions of the world (an exception being Australia), and within each continent the genetic spread is considerable. For instance, the fifteen Asian languages represent eleven unrelated families, and none of the five New World languages is genetically or areally related to any other sample language. Thus, the data used for this study do give us a good first approximation to the linguistic diversity found in the world. I hope that the cross-linguistic data discussed in this study will inspire field workers and researchers of little-known languages to investigate this little-studied phenomenon in the language of their expertise.

The data assembled here (and presented in list form in the Appendix) come from three different types of sources: Native-speaker answers of a questionnaire, published reference material (grammars and dictionaries), and translations of the New Testament. The first source, native speakers, was consulted for the following languages (for acknowledgments see the preface):

German, Italian, Russian, Bulgarian, Estonian, Armenian, Georgian, Persian, Hausa, Chinese, Japanese, Korean.

The second source, reference materials, was used to a greater or lesser extent for all other languages. Descriptive grammars were an important (and often the only) source of data for those languages for which grammars of the

Theoretical prelude: The relation between space and time

That the expression of temporal relations is often similar to that of spatial relations has often been observed. I have not been able to trace the origin of this observation; it may be that this has been known to linguists and philosophers for many centuries, or that it is so evident that it has often been rediscovered independently. In this section I will examine a number of views from the literature, showing that claims of different degrees of strength have been made.

The following quotations, more or less randomly selected, are illustrative of the kinds of views that have been expressed on the relation between temporal and spatial expressions: In addition, the claims also differ with respect to quantitative strength. Again, different degrees can be distinguished in the quotations in E16:

E18. The claim in E17 is true a. for English and other related languages (Clark) b. (implicitly) more generally than just for the language that is immediately under discussion (Jackendoff, Langacker) c. for many languages (Lyons) d. possibly for all languages (Wierzbicka) e. for language in general (Meyer-Lübke, Gamillscheg)

The systematic cross-linguistic study of temporal adverbials allows us to evaluate the impressionistic claims in E18 quite directly, although due to the bias in my sample the answer will still not be conclusive. But what does it mean for temporal expressions to be "based on" spatial expressions, and how can this be observed? The mere identity of related temporal and spatial markers is not sufficient -this merely amounts to claim E17a. Of course, E17a follows from E17b: If time adverbials are based on space adverbials, they will often be formally identical with them, but the reverse is not true. Space adverbials could also be based on time adverbials, or it could be, as JACKENDOFF (1983:210) suggests, that the spatial conceptual machinery is not transferred to the notional domain of time, but that both spatial structure and temporal structure are instantiations of "an abstract organization that can be applied with suitable specialization to any field". Thus, the distinction between E17a and E17b is important. If the stronger E17b can be shown to be correct, then JACKENDOFF must be wrong and the hypothesis of a conceptual transfer from space to time (i.e. E17c) is justified.

One kind of formal indication of a "based on" relation would be that temporal markers are systematically characterized by an additional element that is absent in spatial markers. This kind of formal asymmetry can be observed, for instance, in indefinite and interrogative pronouns, as I show in HASPELMATH (1997). Indefinites and interrogatives may be identical (e.g.

German was

'what; something') or the indefinite may be characterized by an additional indefiniteness marker (German irgend-was 'something'), but not vice versa. However, this kind of asymmetrical relation is not found in our current domain -time expressions are never derived from space expressions by a special "time marker".

Instead, the "based on" relation must always be understood in terms of metaphor or conceptual shift: German vor 'before' is based on vor 'in front' in that the spatial sense of vor is chronologically primary. At some point speakers decided to use vor 'in front' also in the temporal sense 'before'. Synchronically there may be no indication of the diachronically asymmetric relation between the two, but if evidence can be found that temporal marker are often diachronically secondary with respect to spatial markers, and if that relation is never the reverse, this constitutes a strong argument for the hypothesis of conceptual transfer.

The diachronic primariness cannot of course be directly read off from the synchronic data. However, for quite a few of the languages etymological information is available, and this consistently points in the same direction. The details will be provided in the relevant sections.

Although it is widely assumed that humans conceive of time in a way analogous to space, other views have been expressed. Thus, WIERZBICKA (1973) proposed that temporal location should be analyzed semantically in terms of the primitive notions 'world', 'become' and 'part of'. 7 Thus, she proposed the following explications of simultaneous location, anterior and anterior-durative: E19. a. John played the piano on Monday.

= The world of which John playing the piano was a part was the world called 'Monday'. b. Buddha lived before Socrates.

7 In the meantime, the author has abandoned this analysis (cf. WIERZBICKA 1993:453, GODDARD & WIERZBICKA 1994:45-46). The reason why I discuss her proposal here is that it helps us see clearly what a possible alternative to the "spatialist" conception of time would be, even though this conception is apparently uncontroversial nowadays.

= The world of which the living Buddha was a part was a world that was becoming the world of which the living Buddha was a part.

c. Carmen played until six o' clock.

= The worlds of which the playing Carmen was a part were the worlds which were becoming the world of six o'clock.

WIERZBICKA contrasts this explication with one proposed by some philosophers according to which the world is four-dimensional, and a time span can be thought of as a part of this world. According to this view, things have both spatial and temporal parts, so that, for instance, a woman would be said to consist of a a baby, a girl, a young woman and an old woman. WIERZBICKA correctly observes that "this conception is alien to common intuition" (1973:618), but the same is true, in my view, of her conception of location in time. We do not say of particular points or time spans that they are worlds, and we do not speak of the changing world as a series of successive worlds. In our ordinary language (and hence conceptualization), the world changes, i.e. its properties become different, but the world remains the same. Indeed, the very notion of 'becoming', a semantic primitive in WIERZBICKA's theory, seems to be derived from the notion of movement, judging by the number of cases in which a 'become' verb is derived from a movement verb. 8 If WIERZBICKA's hypothesis were correct, we should expect in addition that at least in some languages the 'before' expression would be based on the expressions for 'world', 'become', and 'part'. However, I have not come across a single language in which this is the case.

Mapping the spatial axes onto the time line

If spatial notions or expressions are carried over to temporal ones, there are a priori three simple ways of doing this, because corresponding to the single time line of one-dimensional time there are three axes of three-dimensional space: the frontal axis (front-back), the vertical axis (up-down), and the lateral axis (right-left). It has often been observed that it is overwhelmingly the frontal axis that is used for this purpose. CLARK (1973:49) notes this property of English, but it is true in general of human languages, as this 8 E.g. English become (cf. come), Polish zostać (cf. stać 'stand'), German werden (cf. Latin vertere 'turn'). For additional cases, cf. MICHAELIS (1997). study shows. I know of no single example of the use of the lateral axis for temporal relations (such as 'to the left of Monday' or 'to the right of the discovery of America'), and the use of the vertical axis is very rare. A well-known example is the use of 'up' and 'down' in Chinese for 'last' and 'next' (e.g. shàng 'up', shàngnián 'last year', xià 'down; next'). However, I know of no language whose regular 'before' or 'after' expression is derived from 'above/on top' and 'below/under' (but see BICKEL (1994) on Belhare, a Tibeto-Burman language of Nepal, where something similar seems to occur). In European languages, the up-down axis is usually restricted to marginal uses, as in the French cases discussed by ANSCOMBRE (1993) The reason why speakers of human languages so consistently choose the frontal axis for expressing sequential location is of course that the passing of time is conceived of in the same way as movement through space. In this way an immediate link with the frontal axis is established, because this axis, too, is defined with respect to movement. By contrast, the vertical axis is determined only with respect to gravity on earth, and gravity is effective also if no movement takes place. Of course, gravity becomes visible especially when things move toward the earth, but crucially this movement is bounded (falling things cannot fall further than the ground), whereas the passing of time is unbounded. The lateral axis is clearly secondary with respect to the frontal axis, because only objects that have a front-back orientation can be said to have a right-hand side and a left-hand side. The frontal axis is often defined in terms of the direction of canonical movement through space, 9 so there is a close association between this axis and movement. Thus, given that the passing of time is assimilated conceptually to movement through space, the choice of the frontal axis is well-motivated and well-understood.

9 Cf., e.g., LYONS (1977:691): "[Man] has his principal organs of perception directed towards the region in front of him; he normally moves in the direction in which he is facing...". Note that Lyons's first criterion is not as general as the second one because it does not apply to things that move but do not have organs of perception (e.g. arrows, shooting stars). See also the definition in FILLMORE (1971). CLARK (1973) defines the frontal axis in terms of the "canonical encounter" situation, which also has a movement component.

Chapter 2 Semantic functions of time adverbials

In this chapter we look in some detail at the semantic properties of each of the various temporal qualifications that are the topic of this study. This discussion of the semantics is intended as a preparation for the presentation of the crosslinguistic patterns in later chapters. It does not claim to be an independent contribution to semantic theory. In order to make the chapter maximally accessible, the presentation will be fairly informal. 1 However, I would expect that the cross-linguistic patterns discovered in this work will eventually be useful for a deeper understanding of the semantic structures of time adverbials.

Before the individual temporal relations are discussed, I will say a few general words about the semantics of time.

The semantics of time

Time is semantically very simple. It can be thought of as a sequence of points which are located on an imaginary time line (or "time axis"). In contrast to three-dimensional space, time is one-dimensional and has nothing analogous to the vertical axis (up-down) or the lateral axis (left-right). In addition, time is unidirectional in that for two points on the time line that do not coincide (i.e. are not simultaneous), one is unambiguously earlier and the other is later.

Stretches of time (called time spans in this work) can be evaluated quantitatively, i.e. as shorter and longer, and they can therefore be measured. Finally, time is not bounded on either side. This description exhausts the properties of time itself that are relevant for a linguistic description of temporal notions. Schematically the properties of time can be represented as in Figure 1.

Of course, talking about time gets somewhat more complicated and more interesting because of the way in which positions on the time line are determined and temporal extent is measured. The purpose for which we need time in language is to characterize situations, i.e. entities thought of as variable in time, in terms of their temporal location or extent. Situations can be located only with respect to other situations, and the temporal extent of situations can be measured only by comparing it to the extent of other situations. 2 Modern technology has made it possible to measure time in abstract units of measurement, but even the most sophisticated of measuring methods ultimately relate situations to other situations (e.g. the regular swinging of a pendulum, or the oscillations of certain crystals). The main reason why temporal expressions are often complicated in languages is that the situations that speakers can conceive of are so diverse with respect to their temporal structure and can be related temporally to each other in multiple ways.

The most common characterization of situations is with respect to the speech situation: Many languages have obligatory grammatical markers in every sentence (i.e., tense) that characterize the situation as occurring in the past (earlier than the speech situation), in the present (coinciding or overlapping with the speech situation), or in the future (later than the speech situation).

Since there is such an enormous experiential difference between the past, which can be remembered in minute detail, and the future, of which only vague outlines are known to people on earth, one might think that the past and the future are treated in radically different ways in languages. To be sure, there are often asymmetries between past and future in linguistic expressions, but on the whole it is surprising to what a high degree past and future temporal expressions are symmetrical. The temporal relation of situations to the speech situation, or time deixis, is more relevant to the study of tense than to the study of time adverbials, but deixis will play a role at various places in this study. A general discussion of deictic properties of time adverbial markers can be found in §3.1.

Temporal characterization is also possible by relating a situation to another individual situation, e.g. The baby was born before her great-grandfather died, or 2 This is quite evident when temporal adverbial clauses are used, but in all other cases of temporal characterization reference must be made, however indirectly, to some other situation. This is not always recognized, cf. HERWEG (1990:16-17): "Während die "eigentlichen" Zeitadverbiale immer ein Element enthalten, das direkt, d.h. ohne den Umweg über ein Ereignis, einen Zustand oder einen Prozess, auf eine Zeitspanne oder einen Zeitpunkt referiert, bestimmen temporale und durative Nebensätze die zeitliche Einordnung bzw. Dauer der im Hauptsatz eingeführten Situation mit Bezug auf eine andere Situation..." But there is no "direct" way of locating or measuring time, because the only way of identifying times is through the situations that take place at them. I will be happy as long as you are with me. Individual situations are generally represented by clauses, so the linguistic expression for this kind of temporal characterization are usually temporal clauses, which fall outside the scope of this work. However, in most languages nouns can also be used to denote situations. In the most common case, deverbal action nouns fulfill this function, e.g. after my arrival, during the strike of Metro employees, before the birth of Jesus Christ, etc. There is no clearcut boundary between deverbal action nouns used in time adverbials and temporal adverbial clauses, because in many languages subordinate clauses are more or less nominalized, or action nouns have a number of clausal properties, or both. Thus, while expressions like after my arrival in a language like English clearly fall in the scope of this work, they are not prototypical cases.

Many languages also have nouns denoting specific situations that are not derived from verbs, e.g. war, festival, flood, lunch, ceremony, etc. In the familiar European languages there is nothing special about these nouns, and in these languages they are the most typical nouns occurring in NP-based time adverbials. However, it must be kept in mind that such nouns are in all likelihood not universal.

The most common type of situation-denoting nouns are undoubtedly nouns denoting what I call here canonical time periods. The major cyclic events of the human natural environment on earth have probably always served as the main means of locating and measuring other situations: in particular, the alternation of light and dark, changes in the shape of the moon, and changes in the path of the sun across the sky (accompanied by marked climatic differences). We do not usually think of days, months and years as events, because we are so used to these cyclic events that we mostly focus on their function as measuring units.

It appears that all languages have nouns denoting (at least a subset of) these units of time measurement, and if this is true, then all languages must have NPbased time adverbials. A true counterexample would be a language that consistently used expressions like 'when it has gotten light three times' (for 'in three days' time'), or 'the leaves have fallen seven times since' (for 'seven years ago'). It could also turn out that there are languages whose speakers do not use higher numbers and are satisfied with lexical adverbs like 'today ', 'yesterday', 'this-month', 'next-year', etc. Such a language would still make use of the canonical time units, but would not have NP-based adverbials falling under the definition of my study. I have not found such a language, but I will point out cases where a grammatical description indicates that an NP-based time adverbial in our European languages corresponds to something very different in the languages described (cf. §3.3).

In addition to the natural time units, there are culture-bound artificial time units. In the currently dominating culture, this is the week, as well as finer subdivisions of the day (hours, minutes, seconds) and larger groupings of years (decade, century, millennium). But more importantly, the cycles of the year and the day are naturally divided into qualitatively different periods, the parts of the day (morning, afternoon, evening, night, dawn, etc.) and the seasons (minimally summer and winter in regions distant from the equator, often rainy and dry seasons in regions near the equator). Probably all languages have expressions locating situations in a part of the day or in a season, but again this does not necessarily mean that all languages use noun phrases for this purpose.

Since there are very few of these qualitatively different periods and they are overwhelmingly used for temporal location 3 , adverbs may well be more suitable for expressing such qualitative periods than nouns (cf. §7.4). Finally, some time units located in a particular calendar position have special names, e.g. months within the yearly cycle (January, February,...), and days within the weekly cycle (Sunday, Monday,...). Other time units are merely numbered, e.g. days within the monthly cycle (March 1st,2nd,etc.), and hours within the daily cycle (one o'clock, two o'clock, etc.).

Table 3 on the next page lists the major cyclic time periods as used in our culture (see LEECH (1969: ch. 7), FILLMORE (1971: 28-37) for further discussion).

Three different sub-types of canonical time periods must be distinguished for the purposes of this study: 1971). The relevance of deixis for temporal location will be taken up again in §3.1.

Before turning to the individual temporal relations, let me briefly explain the conventions used here for representing these relations in diagrams. I will show the situation that is characterized by an NP-based time adverbial (the characterized situation) below the time line, and the situation or time unit with respect to which the main situation is characterized (the reference time, abbreviated RefT) is shown directly above the time line. This is exemplified in Figure 2, where the characterized situation is located with respect to the situation, so it is a located situation, a special case of a characterized situation. 4 Further notational conventions will be explained below.

4 The terms reference time and characterized/located situation are not widely used, but I hope that they are self-explanatory. My terms are completely analogous to those used in HERSKOVITS (1986) in a spatial context (reference entity, located entity). SNOOK (1988) uses locating term and located term, and the Prague Academy grammar of Russian (BARNETOVÁ et al. 1979) uses vremennyj orientir, a felicitous term whose English translation (roughly, 'temporal orientation mark') is unfortunately not nearly as elegant.

Simultaneous location

The label 'simultaneous location' refers to markers that locate a situation with respect to a reference time (i.e. another situation or canonical time period) which is simultaneous with the situation. In many languages the same markers are used for this function as for interior spatial location ('in'), but in contrast to the spatial interior function, the reference time need not properly include the located situation. This is perhaps the prototypical case, but it is by no means the only possibility. An example of inclusion is Carsten passed the exam last week, illustrated in Figure 5 (here and below, S represents the moment of speech): Here the located situation is punctual and is therefore included in the reference time, which is a time span. However, the relation may also be the reverse: The reference time may be a point in time, and the located situation may be durative and take up a longer time span. In this case, it would be more appropriate to say that the located situation includes the reference time. An example is shown in Figure 6. Things get even more complicated when both the located situation and the reference time are non-punctual. An example of this is given in Figure 7: This sentence is clearly true if Lea was sick for a period of less than a month which completely fell into last month (this reading corresponds to proper inclusion of Figure 5), but it is also true if Lea has been sick for three months and is still sick, i.e. if the located situation includes the reference time rather than vice versa. This is indicated by the dots at both ends of the horizontal line representing the located situation in Figure 7. Thus, the necessary and sufficient condition for the possibility of using expressions like last month, at 4.15 is that the located situation and the reference time should overlap, i.e. be (at least partially) simultaneous. Hence, the term "simultaneous location" seems to be the best term (cf. COMRIE & SMITH's (1977:32) term "general temporal location", which is less specific but would also be appropriate).

Simultaneous location is generally marked by the most grammaticalized locative markers, as is the case in English (in, on, at, Ø). In addition, some and habitual situations. Thus, durative is not the same as 'having a certain duration' (as in the terminology of QUIRK et al. (1985:201)). However, my use of durative seems to be close to its traditional sense.

This contrast is not only difficult to describe, it is also difficult to find an analog of during outside of Standard Average European (i.e. Germanic, Romance, Slavic). This is the reason why I have not distinguished a separate semantic function for during. Its special properties can be illuminated by the cross-linguistic point of view only to a limited extent.

Within the semantic category of simultaneous temporal location, I distinguished seven special cases, six of which correspond to the most important canonical time periods of Table 3: (1) hour, (2) part of day, (3) day, (4) month,

(5) season, (6) year. The seventh special case is (7) festival. These are not only the cases for which it was easiest to get data (in the Comrie-Smith grammars, there are sections for each of them, and all of them are represented in the New Testament), but they are also presumably the most frequent cases,foot_1 and are hence the most likely to show special behavior. Thus, it is unlikely that a language would use a special adposition or case marker for 'in the third millennium', different from the marker used in 'in the third century'. Among these categories, (2)-( 6) are unproblematic from a semantic point of view: They all denote cyclically recurring time spans. Many languages of course make finer distinctions, which will be discussed when the cross-linguistic data are presented (Ch. 7).

Somewhat special are the cases of the hour and the festival. The hour is also a cyclically recurring time span (cf. Table 3), but when it is used for temporal location, reference is usually made not to this time span, but to the completion of an hour, i.e. to a point in time. Thus, it is not surprising that location at an hour is treated in a special way in many languages ( §7.3). Festivals are special in that they typically recur cyclically within the annual cycle, like the seasons, but they usually occupy only one day or at most a few days.

Sequential and sequential-durative location

In this section, I will discuss four semantic functions in which the located situation is related to the reference time in that one is earlier and the other is later, i.e. that they occur in a sequence (hence my general term sequential, which was suggested to me by TRAUGOTT's (1978:379) sequencing). In the anterior and posterior functions, nothing more is conveyed: An anterior marker locates the situation earlier than the reference time, and a posterior marker locates the situation later than the reference time. As in the case of simultaneous location, the punctual/bounded or durative nature of the located situation is immaterial.

But there are two related semantic functions with more specific meanings which require that the located situation be durative: The anterior-durative ('until') and posterior-durative ('since') functions. In addition to specifying a 7 In KORTMANN (1997), the terms anteriority and posteriority are used in the opposite way:

After is said to be a conjunction of anteriority, before is a conjunction of posteriority. KORTMANN describes the meaning of these conjunctions as follows (1997: 84-85): Anteriority: 'after p, q: p simply precedes q in time'; Posteriority: 'before p, q: p simply follows q in time'. Thus, for KORTMANN the reference times are before ("anterior") and afterwards ("posterior"), whereas for me the located situations are before and afterwards. My usage is not only in conformity with the forms of the markers (in Latin, ante denotes anterior, post denotes posterior, in my terminology), but also accords with the intuition that markers of time adverbials serve to characterize situations in terms of reference times, not reference times in terms of situations. relation of sequence, these indicate that the duration of the located situation overlaps with the reference time. For instance, in E23a the located situation of Cameron's being in Glasgow is not only characterized as being later than the reference time February, but also as overlapping with February, i.e. the sentence would not be true if Cameron's being in Glasgow began only in March. By contrast, Cameron has been in Glasgow after February would also be true in this latter case. Analogously, E23b implies that Henriette worked not only before, but also in the summer. More precisely, it implies at least that Henriette worked at the beginning of the summer. Whether the reference time is included or not depends on the context. Thus, in E24a until is more likely to be inclusive, while in E24b it is more likely to be exclusive.

We camped there from June till September.

(inclusive or exlusive) (ii)

We camped there from June through September.

(inclusive) (iii) Ja budu rabotat' tam do sentjabrja.

(inclusive or exclusive) 'I'll work there till September.' (iv) Ja budu rabotat' tam po sentjabr'.

(inclusive) 'I'll work there through September.'

9 For more discussion of the semantics of sequential-durative (especially posterior-durative), see GIAUFRET-COLOMBANI (1989), LYSEBRAATE (1982), SNOOK (1988), MANZOTTI & RIGAMONTI (1983). The dual semantic condition of anteriority/posteriority and overlap (i.e. simultaneity) with the reference time means that the verb of the located situation cannot be punctual/bounded, but must be durative, because a situation that is thought of as having no temporal extension cannot both be simultaneous with and prior/subsequent to a reference time. 11

In the case of the posterior-durative function, many languages make an additional meaning distinction, which relates the located situation also to the moment of speech. For instance, English since is only appropriate if the located situation is in the past and overlaps with the moment of speech (or a different deictic center, as in narration), and thus it can be said to have two deictic meaning components (past and overlap with the present). If Cameron's stay in Glasgow extended from February to June and the moment of speech is on September 28th, one cannot say *Cameron has been/was in Glasgow since February. Instead, one must say Cameron was in Glasgow from February on.

Similarly, if the moment of speech precedes the located situation, since cannot be used because it is confined to past situations, independently of whether the located situation overlaps with the moment of speech. Thus, one cannot say *Cameron will be in Glasgow since February if the moment of speech is in January, 10 E.g. QUIRK et al. (1985:691) ("terminal point"), BENNETT (1975:119) ("the notion 'end'"), KORTMANN's (1997:84-85) terms terminus ad quem, terminus a quo; and the case name terminative (Russian predel'nyj), which is employed, e.g., in Udmurt and in Hungarian grammar. 11 In some languages, a secondary use of a posterior-durative marker is possible in which it comes close to the simple posterior use (cf. TEN CATE 1989), e.g.

(i) English:

Since your last encounter with her she has gotten married. (ii) German:

Seit seiner Heirat haben sie zwei Kinder bekommen.

In these examples, the usual posterior preposition (after/nach) could also have been used (requiring a change of the tense in English). The semantic distinction in these cases is quite subtle. I have no data on the extent to which this secondary use occurs cross-linguistically.

or *Cameron will be in Glasgow since now if the moment of speech is in February.

English must use from (..onward) in these situations.

To sum up, here are the generalized representations of the anterior, posterior, anterior-durative and posterior-durative functions: The more specific meaning of English since (which we may call posterior-presentperfect) can be represented as in Figure 17.

Temporal distance

In addition to locating situations by marking them as simultaneous with, prior to or subsequent to other situations, we can locate situations even more accurately by indicating their temporal distance from a prior or subsequent reference point. This of course presupposes that temporal distance can be measured. In almost all cultures counting is used, and we have already seen that the cyclic time units (day, month, year, etc.) can be used for quantifying temporal extent. When measuring temporal extent, these units are used somewhat differently compared to their use in locating situations. When year is used to locate a situation, it denotes a period with a fixed initial point (January 1st) and a fixed terminal point, i.e. it is used calendrically. The situation in E26 can be represented as in Figure 18: But there is a type of distance marking that is much more common and for which it is not difficult to obtain descriptions: marking of distance from the moment of speech. Depending on whether the located situation precedes or follows the moment of speech, I distinguish the two functions distance-past ('ago') and distance-future ('in'). They are illustrated in Figures 1920: In these examples, the distance is measured with respect to a reference point which is different from the moment of speech. I call these semantic functions distance-retrospective and distance-prospective, respectively. In English, the formal marking is different from that of the distance-past/future functions.

Other languages make no such distinction, as will be discussed in greater detail in §6.3 below.

To sum up, here are the generalized representations of the distance-past and distance-future functions: As in the case of the anterior-durative and the posterior-durative functions, there is no entailment that the situation began or ended at a particular time.

12 My terms atelic extent and telic extent are innovations. They have the disadvantage of suggesting that the extent is telic or atelic, whereas in fact it is the verbal situation that is telic or atelic. However, I know of no better pair of terms in English. I have encountered the following terms in the literature: The beginning and the end can be inferred by implicature, but this implicature can be canceled, e.g. The children watched television for five hours, and in fact all day.

Another kind of extent adverbial is the semantic function that I call telic extent, i.e. an adverbial that indicates the length of time that it takes for a telic situation to be completed. A typical example is Tony painted the picture in five hours. In contrast to atelic extent adverbials, telic extent adverbials always characterize a bounded, telic situation, and it makes no sense to say *The children watched television in three hours, or *I slept in twenty minutes. It is not easy to specify the difference in the meanings between the two types of extent adverbials, but the different combinatory possibilities are so salient that the distinction between telic and atelic extent adverbials has become one of the most important tests for telicness (VENDLER 1957, DOWTY 1979). The diagrammatic representation of an telic extent adverbial is thus very similar to that of atelic extent adverbials: Stephen lives in Hongkong since ago five years

d. (Persian) Stephen az panj sâl-e piš dar Hongkong zendegi

Stephen from five year-ATTR ago in Hongkong life mi-kon-ad.

IMPF-do-3SG

If we just look at English, it appears that nothing special is going on -this is just a regular atelic extent adverbial combined here with a verb in the present perfect tense, which yields the meaning that the five-year period in question extends into the present, i.e. that it began five years before the moment of speech. The diagrammatic representation is shown in Figure 25: Persian, where the overt marking is transparently composed of these two markers, as shown in E29c-d (desde 'since', hace 'ago'; az 'since', piš 'ago').

One might ask whether the counterpart of the distance-posterior type exists as well. A priori, we would certainly expect this, because precedence and subsequence, past and future are typically fairly symmetric in languages. A "distance-anterior" function would occur in a sentence like 'I will live in this cheap apartment until in two years'. I know of no language that has a special marker for this meaning. To express this meaning naturally, English would again use the simple atelic extent marker (cf. E30a), but so would German (cf.

E30b).

E30. a. (English) I will live in this cheap apartment for two years.

b. (German) Ich werde zwei Jahre lang in dieser billigen Wohnung leben.

The overlap with the moment of speech is not expressed in these sentences, but it would be possible to make it explicit by adding more in English (two more years) and noch in German (noch zwei Jahre lang), with a result that would be parallel to E29a-b for most purposes. However, I do not distinguish a special "distance-anterior" function because I am not aware that any language deviates from the pattern in E30a-b in an interesting way in expressing this meaning (but see §5.5, footnote 5, for two French prepositions with a related meaning).

The three functions atelic extent, telic extent and distance-posterior are shown in the generalized diagrams on the following page: 13

13 For more discussion of the semantics of extent adverbials, see TEN CATE (1984), DEHON (1993), LENARDUZZI (1993).

Chapter 3 General issues

In this chapter I will discuss a few further general points: First, I will discuss the extent to which markers of NP-based time adverbials are deictic ( §3.1); second, I will list a few semantic functions of time adverbials that have not been included in the cross-linguistic study but that should at least be mentioned ( §3.2); finally, I will discuss a number of languages that do not employ NPbased time adverbials for some of the semantic functions, resorting to alternative constructions instead ( §3.3).

Deixis in temporal adverbials

Temporal adverbials often include a deictic meaning component. This is evident with adverbs like today, yesterday (contrasting with non-deictic or anaphoric on that day, the day before), but there are non-lexical markers which signal a deictic distinction, too. In English, the presence or absence of the definite article may be significant, as pointed out by ALLEN & HILL (1979) (e.g. Two weeks ago Frank promised to come the/Ø next Monday).

However, in this section I will concentrate on deictic meaning expressed by the adverbial marker itself. In chapter 2 we saw that among the various semantic functions of NP-based time adverbials, it is particularly the distance functions that commonly incorporate a deictic meaning component, i.e. contain a reference to the moment of speech (cf. ch. 6 for more details). The anterior and posterior functions are never combined with deictic meaning, as far as I can determine -i.e. no language has different expressions for 'after the storm' depending on whether the storm precedes, follows or coincides with the moment of speech. But we also saw that deixis is often present in the posteriordurative function, where e.g. English has a contrast between since and from (...onward), and in the distance-posterior function (cf. §2.3 and §5.4).

The function to be highlighted in this section is the simultaneous location function and its deictic properties. As a rule, NP-based simultaneous adverbials are non-deictic, i.e. the forms of 'at five o'clock', 'at Christmas', 'in the morning' do not depend on the relation between the reference times and the moment of speech. This contrasts with time adverbs, which are often deictic, e.g. 'yesterday', 'tomorrow', 'now', 'soon'. However, there are interesting exceptions to this generalization.

Four languages of my sample have clear examples of simultaneous location

markers containing a deictic meaning component: Maori, Tagalog, Swedish and Greenlandic Eskimo. Maori and Tagalog are both Austronesian, so conceivably these two cases are related (although they are fairly distant from each other, both genetically and geographically). Maori has different prepositions for past and future location: i or noo express past simultaneous location, while a or hei express future simultaneous location. The marker kei is used for both present and future location. This is illustrated in E31-32. In view of the restriction of this tense variation to non-verbal sentences, one might suspect that the prepositions in E33 are really locational verbs (i 'was at', kei 'is at', kei/hei/ko 'will be at'). 1 The sentences in E31-32 could then perhaps be regarded as biclausal (E31a: 'It was on Monday, they went to Rotorua').

However, some temporal location markers lack corresponding spatial tensemarked prepositions (e.g. noo, a). I must leave the deeper analysis of these fascinating facts to the specialists of Maori.

In Tagalog, the usual locative marker, the preposition sa, is often restricted to future time reference. These four languages are sufficient to show that temporal location markers may be associated with a deictic meaning component, although this is not very common. More cases from different families must be examined before any cross-linguistic generalizations can be ventured.

Some additional semantic functions

The semantic functions that were discussed in chapter 2 are the major semantic distinctions made in NP-based time adverbials, but there are a number of additional semantic functions that I have not investigated systematically and that will not play a prominent role in this work. For the sake of completeness, I mention here some of the distinctions that I have encountered in several languages and briefly discuss their properties.

Medial

Most languages seem to have a way of locating a situation between two ref- The spatial concept 'between' is a readily available and unproblematic model for expressing temporal intervals. I have not come across a language that does not use its spatial 'between' expression in the temporal sense, so I did not consider this semantic function sufficiently interesting to warrant a detailed study.

Approximative

Some languages have a special adposition marking approximate simultaneous location in time. Particularly with times and periods of the day it is often useful to have such a marker which corresponds to 'near' in the spatial domain.

However, the spatial 'near' is not commonly extended to a temporal sense.

Rather, the spatial concept 'toward, against' is often used in Europe (e.g.

Perdurative

Another meaning that is sometimes represented by a special adposition is that of English throughout, i.e. 'during the whole duration of' (throughout the summer). In English, the preposition throughout is clearly based on the spatial throughout (The epidemic spread throughout the country). In Dutch, the preposition gedurende has this function, but its source is analogous to that of English during, Italian durante, etc. (Dutch duren 'to last'). According to VERKUYL (1973:584), its meaning must be described by means of a universal quantifier, like English throughout: Gedurende de vergadering zat hij te lezen 'Throughout the conference he sat reading'. Thus, both a spatial and a non-spatial model can give rise to the perdurative meaning.

Purposive extent

An example of the function that I call purposive extent is They went to Vilnius for two years. Here the preposition combines with a noun phrase denoting a time span, but its meaning is not entirely temporal. The sentence can be paraphrased by 'They went to Vilnius in order to stay there for two years'. Thus, the meaning of purpose is also a semantic component of for in this function.

Purposive extent adverbials very often occur with verbs of movement, but it is not possible to say that for x time units generally means 'in order to stay for x time units'. For instance, E43a-b are unacceptable. It appears to be typical of purposive extent markers that they are based on purposive markers. This is the case in English, but also in the languages in E44.

E44. a. German

Irgendwann fahren wir für drei Tage nach Hiddensee.

'Some day we'll go to Hiddensee for three days. Sandra traveled away for some days 'Sandra left for a couple of days.' g. Latvian (NICOLE NAU, p.c.)

Es uz pāris stund-ām aiz-ie-šu uz bibliotēk-u.

I for couple hour-PL.DAT away-go-FUT.1SG to library-ACC 'I'm going to the library for a couple of hours.'

Regular recurrence

Many languages have a special adposition or case for the canonical time periods when they qualify a regularly recurring event, e.g. 'every day', 'every Sunday', In principle, it should be possible to distinguish between a universal-distributive meaning, e.g. 'every Sunday', and a meaning that merely involves regular recurrence but does not require universality, e.g. 'on Sundays'. However, in practice it is very difficult to keep these two strictly apart. 4 The option of using a universal-distributive quantifier seems to be available in most languages, so I will not illustrate it further here (see also §7.6, E146, for further examples).

Sometimes the marker used for regular recurrence is used more generally in the distributive function, e.g. Russian po, Udmurt byde (see HASPELMATH (1995) for the connection between distributive adpositions and universal quantifiers).

In other cases, the plural of the time unit is sufficient to denote regular recurrence (of course, this is possible only with calendar unit names and qualitative periods, not with time units): 'on Saturdays'

Alternatives to NP-based time adverbials

There is no denying that there is an ethnocentric bias in the choice of the semantic functions for this study. Many non-industrialized cultures are not as obsessed with time as we are, and many languages outside of Europe have traditionally lacked the rich nomenclature for calendar units and temporal relations that we find in European languages. Due to the European bias in my sample, this is not sufficiently reflected in my data, and to make up for this shortcoming I will use this section to highlight some languages spoken far away from the centers of Western technology and capitalist economy. In most cases, the available descriptions of such languages are simply silent about the more complex temporal units and relations, but some of the Comrie-Smith grammars explicitly say that certain constructions are not possible. In the present final subsection of this chapter I will discuss languages in which some of the core semantic functions cannot be expressed by NP-based time adverbials, so that they have to resort to alternatives in order to render the same ideas.

I begin with simultaneous temporal location. This is probably the least problematic temporal relation, and I know of no language that completely lacks NP-based simultaneous adverbials. However, not all of the canonical time periods occurring in expressions of simultaneous location are universal.

According to DERBYSHIRE (1979:120-23), Hixkaryana has traditionally lacked words for 'hour', days of the week, months of the year, and expressions for particular years, and the concept of specific festivals on certain days or occasions is a new one. Similarly, speakers of Kobon (a language of Papua New Guinea, DAVIES 1981:140-45) have not traditionally known hours and weeks, and the lunar cycle has not been related to the solar cycle, so that months could not be identified by their position within the year. Not surprisingly, the words for 'hour' and 'week' are often the youngest words in a language, and were often borrowed from a dominant culture with more concern for time (e.g.

Chechen saħt, Nkore-Kiga eshaaha, Swahili saa from Arabic saaʕ0at; English hour, German Uhr from Latin hora, etc.). However, all of these considerations are of less concern to linguistics than to anthropology, and there is a rich anthropological literature on time measuring in non-Western cultures (e.g. HALLOWELL 1939, FETTWEIS 1958, MÜLLER 1962, ALVERSON 1994) 'He came at the time when women return from the farm.' (ca. 17 h) Thus, speakers may be fairly specific about the time of the day even if no clock technology is available to them. However, since all these expressions are adverbial clauses rather than NP-based adverbials, they fall outside the scope of this study.

In Hixkaryana, even the borrowed expressions of time of the day take the form of an adverbial clause: 'We were here before the tenth of October.' (lit. 'We were here before, October tenth came afterwards.') b. Made lug-nɨg g-ab hon aui mɨd-ei-nab-un.

[Monday fall-PURP do-PRES.3SG] we here be-DUR-FUT-1PL 'We will be here after Monday.'

A number of languages also employ clausal adverbials for the distance-past and distance-future functions. For Hixkaryana, DERBYSHIRE (1979)

Chapter 4 Sequential location

The first notional domain that will be discussed from a cross-linguistic perspective is that of the anterior and posterior functions, i.e. sequential location.

These two functions are formally parallel in many ways.

Anterior/posterior based on spatial front/back

The space-to-time hypothesis can be tested quite easily on the expressions used for the temporal anterior and posterior functions. These two notions are among the simplest temporal relations because neither the nature of the reference time nor the nature of the located situation is relevant for them. The cross-linguistic evidence overwhelmingly confirms the view that time is conceptualized in terms of space, more particularly in terms of the frontal axis.

A large number of languages from a wide variety of families show this association either synchronically or diachronically. In almost all cases, the front is associated with 'before' and the back is associated with 'after'. Table 4 on the next page lists those languages in which at least one of the 'before' and 'after' expressions are synchronically identical to the corresponding 'in front of' and 'behind' expressions.

In addition, there are a number of cases in which 'before' and 'after' adpositions are originally derived from spatial 'front' and 'back' expressions, but synchronically the two adpositions are not identical for one of a number of reasons. These cases are discussed below in §4.3. Altogether there are thirtythree languages in my sample for which a current or earlier identity (or near identity) of spatial and temporal sequential markers can be established, contrasting with just seven languages in which at least one of the markers is clearly not based on a spatial expression. (In the other languages, the original meaning of the anterior/posterior markers is spatial but not anterior/posterior, or it is not known.) These seven languages and their markers are discussed in §4.4 below.

Udmurt aźyn

The data of my study thus largely confirm HILL's (1978:524) claim that "in most languages the lexical resources used for representing orientation along the front/back axis in horizontal space are also used for temporal orientation".

But while the cross-linguistic data do show a clear preference for the spatial option, it is also clear that it is not the only option. We thus cannot simply say that humans think of sequential relations in terms of the spatial frontal axiswe have to say that there is a strong tendency for them to do so.

The front/back orientation of the time line

In §1.8 I provided an explanation for why the time line is usually modeled on the frontal axis of space, rather than the lateral or the vertical axis. A further question concerns the orientation of the time line, i.e. why the temporal notion 'before' is systematically associated with 'in front', and 'after' is associated with 'behind'.

Before attempting an answer to this question, we have to characterize the meaning of 'in front of' and 'behind'. When the reference object is an animate being, this is straightforward: 'In front of' means 'near the side of the primary organs of perception' and/or 'near the side which arrives earlier at places when the animate being moves ' (cf. FILLMORE 1971). This side is called the front side, and the opposite side is the back side. With animates, 'in front of' simply means 'near the front side of'. In addition to animates, some inanimate objects which are closely associated with human individuals (e.g. shoes, chairs, bicycles, houses) also have front and back sides, defined analogously. However, these criteria cannot be applied to inanimate objects which do not move and are not connected closely to human individuals, e.g. trees, stones and tables. Yet it is possible to say 'in front of the tree', 'behind the stone', 'in front of the table' in many (perhaps most) languages. The reason is that in addition to the objectbased use of the front/back concepts, there is also a subject-based use, 1 in which objects are treated as if they were a mirror image facing a conscious subject. Thus, in Figure 29 person A would say that the soccer ball is in front of the tree, whereas person B would say that the soccer ball is behind the tree, although the tree does not have an inherent front or back side.

Figure 29.

When the reference object has an inherent front/back orientation, both the object-based and the subject-based descriptions are possible. Thus, in Figure 30 person A could say that the ball is in front of the bike (object-based), or that the ball is behind the bike (subject-based).

1 In these terms, "subject" and "object" have their non-technical senses. In order to predict how the meaning of the front/back axis is transferred to the time line, we have to determine whether reference times (reference situations) can be said to have inherent fronts and backs. Reference times of course are not animates, but they are commonly thought of as moving.

It has often been observed that there are two ways in which time can be conceptualized in terms of movement: Either time is stationary, and the observer in the world moves through it, or the observer in the world is stationary and time moves past him or her. These two models of conceptualization are called moving-ego and moving-time here, following CLARK (1973:50). Both models are attested elsewhere in the language, for instance:

E56. moving-ego

As we go through the years...

As we go further into the 1990s... We're aproaching the end of the year.

In the weeks ahead of us... This coming Tuesday...

Bygone events

E57. moving-time

The time will come when...

The time for action has arrived.

Noon crept up on us.

Time flew by.

Thursday rushed by.

If time is thought of as moving, then points in time or time spans can also be thought of as having an inherent front-back orientation (CLARK 1973:50). Since time moves in the direction of the observer (or to the observer's now), earlier times are "in front" of later times. Thus, the two sentences in E58 are quite parallel:

E58. a. The Christmas season is approaching, and Thanksgiving is before it.

b. The king's car is approaching, and the bodyguards' cars are in front of it.

Given the moving-time model, the consistent relatedness of 'before'/'front' and 'after'/'back' is explained.

But now let us consider the other model of conceptualizing time, movingego. In this model, the observer moves from earlier moments to later moments and thus faces the future. In this situation, times do not have an inherent front-back orientation, so this axis can only be used in subject-based descriptions. Like objects in space, times and situations would be treated as occupying a mirror-image position. With regard to future situations, this would give the same result: If the observer "looks ahead" to a future event, say, his death, then situations that are earlier than his death are "in front" of his death.

Thus, the use of a spatial 'in front' adposition in a sentence like She wants to see her granddaughter before her death can be explained both by the moving-time and by the moving-ego models. However, with regard to past situations, the moving-ego model makes a different prediction: If the observer "looks back" to a past event, say, his birth, then situations that are earlier than his birth are "behind" his birth. But recall from §3.1 that anterior/posterior adpositions are never sensitive to the deictic past/future distinction -there are no languages that invert their 'before' and 'after' adpositions in past situations. We can conclude that it is the moving-time model that is generally responsible for the use of spatial front/back terms as anterior and posterior markers.

Another interesting question is whether the transfer from the spatial domain to the temporal domain occurs equally often with anterior and posterior markers. Table 4 suggests that there is an asymmetry. 'Before' is identical to 'in front' in seventeen languages, but only eight languages show identity of 'after' and 'behind'. In eleven languages, only 'before' shows this identity, while there are only two languages where only 'behind' shows it. The data are not sufficient to prove conclusively that there is a significant asymmetry here, but such an asymmetry may well be motivated: According to VANDELOISE's (1991) semantic analysis, 'behind' is not defined purely by its topological position on the frontal axis, but additionally contains the functional meaning component 'hidden'. In the temporal domain, this meaning component has no placeearlier times or situations do not "hide" later times or situations. Thus, if VANDELOISE is right in his analysis of 'behind', this would provide an explanation for the slight asymmetry observed in Table 4: Due to its meaning component 'hidden', the spatial marker 'behind' is less suitable for transfer to the temporal domain than its counterpart 'in front'.

Diachrony and grammaticalization of sequential markers

In a number of languages, there is diachronic evidence showing that the 'before' and/or 'after' expressions were originally identical to the 'in front'/'behind' expressions, though this is no longer synchronically the case. In other instances the temporal expressions are only formally related to the spatial expressions, and it is not certain that the original meaning was spatial. But in each case the available evidence (synchronic or diachronic) is consistent with the hypothesis that temporal expressions are based on spatial expressions, never vice versa. The cases are shown in Table 5. When a new word denoting the concept 'in front' came into the language (in English, in front through complete lexical renewal; in French, devant through reinforcement by the prefix de-: *de-avant > devant), they came to denote the temporal sense exclusively. This is not what one would expect under a

Gesamtbedeutung approach (BENNETT 1975). Under such an approach the expectation would be that the new word would immediately have both the spatial and the temporal sense. The same point can be made with the Standard Arabic and Maltese posterior prepositions. In the more conservative Arabic, there is a contrast between baʕda 'after' and waraaʔa 'behind'. In the more advanced Maltese, wara (< waraaʔa) means both 'behind' and 'after' (baʕda has been lost from the language). This is again an example of a secondary temporal sense of a preposition.

A further observation relating to the diachronic dimension is that there is apparently a general tendency for temporal markers to be older, or more grammaticalized, than spatial markers. The cases of my sample are listed in

Further sources of anterior and posterior markers

In addition to the anterior and posterior markers which are based on spatial anterior and posterior markers, several other sources give rise to temporal sequential adpositions. These other sources are less symmetrical than the main source, spatial anterior and posterior adpositions.

In a number of languages the temporal anterior adposition is based on the ordinal number 'first' (or perhaps 'former'): Italian prima di (based on the adverb prima 'at first, earlier', from primo 'first'); Punjabi páílãã (< *prathila-'first', a suffix variant of Old Indic prathama-'first'); Latvian pirms 'before; earlier' (< pirmis, an adverbial form based on pirmais 'first'); Kannada modalu 'before; first'.

I have not come across an analogous use of 'last' (or 'later') for 'after'.

Another source are markers of the anterior-durative function 'until'. Russian do is both 'before' and 'until', and the priority of the 'until' meaning is clear from the original spatial meaning 'to' (cf. §5.1). This polysemy is not surprising 2 However, there are three counterexamples to this trend in my sample: Russian posterior posle/za, Latvian posterior pe̅ c/aiz and anterior pirms/priekša. The first two are clearly areally related.

3 "It is precisely because of the predominance of the spatial conceptualization that more expressive forms can be used for these than for the corresponding temporal conceptualization."

because the anterior and anterior-durative meanings are fairly similar; 4 what is perhaps surprising is that it is so rare (cf. §5.1 for further discussion). Russian do is also interesting in that it coexists with another anterior preposition, pered 'before', derived from the spatial anterior preposition. There is a clear semantic contrast between do and pered: Do has the more general meaning, pered refers to the location in time immediately before the reference time, e.g. do vojny

'before the war', pered vojnoj 'right before the war, on the eve of the war'. The only other language that shows this source is Abkhaz, which has the suffix -nja 'until', and the combination -nja-g'ә 'before' (-g'ә is the focus particle 'even'). It is not clear to me how 'before' results from 'even until'. In this respect, Abkhaz is quite symmetrical: 'after' is -štax'-g'ә lit. 'even since' (but -štax' also means 'behind', cf. A few languages have a posterior marker based on 'end, finish': Turkish sonra (cf. son 'end'); Nanay xoǯiočiania/xoǯipia (cf. xoǯi-'finish, end'); Indonesian sesudah/setelah/sehabis 'after' (cf. sudah/telah/habis 'finished'). And finally, the verb 'pass' may also yield a posterior marker, cf. Lithuanian praėjus (converb of praėti 'pass'), French passé (e.g. passé une heure du matin 'after one o'clock in the morning'), English past 'after' (cf. five minutes past twelve).

Chapter 5 Sequential-durative

As we saw in §2.3, the anterior/posterior-durative functions are semantically closely related to the anterior/posterior functions, in particular as far as the meaning component of location is concerned. From this point of view, one might expect that this similarity would be reflected in similar markers of these functions. However, formal similarity of these two types of functions is very rare. Sequential-durative markers have their own characteristics, so a separate chapter is devoted to them.

Allative and ablative sources

The most important sources of anterior-durative markers ('until') and posterior-durative markers ('since, from') are spatial allative and ablative markers, i.e. goal and source markers. Again the priority of space over time is confirmed by the sources of temporal markers. The cases attested in my sample are listed in Tables 78. there is a straightforward explanation for the ablative and allative sources: the concept of "abstract motion" (cf. LANGACKER 1991). Even in the spatial domain, directional expressions are commonly used to denote location along a line which is scanned sequentially by the mind and is thereby assimilated to a directed path: There are wheat fields from the lake to the forest; The highway goes all the way to the Arctic Sea; There are apple trees along many country roads. In these spatial examples, the located entities occupy a large area of space, and it would not be possible to characterize compact objects in this way (??There is a bike along the bike-lane). Thus, ablative and allative sources are particularly well suited for expressing the location of situations with a large extension, i.e.

durative situations.

It is also not difficult to explain why allative markers are associated with the anterior-durative function, whereas ablative markers are used for the posterior-durative function, rather than vice versa. We have here the model of the observer moving along a temporal path from earlier to later, and when situations are thought of as occupying a path in time, speakers scan them sequentially from their earliest part to their latest part (i.e. from their beginning to their end). 2 The alternative model of time moving past the stationary observer is not applicable here, because in this model it is only fixed reference times that are thought of as moving from the future to the past, not located 1 In addition, a number of complex anterior-durative markers contain an allative component:

French jusqu'à, Italian fino a, Irish go dtí, Polish aż do.

2 With many physical objects that can be thought of as having a beginning and an end, it is not predetermined which extremity is the beginning and which is the end (e.g. strings, roads, peninsulas, baguettes). Due to the unidirectionality of time, this ambiguity does not arise with entities that are located in time.

situations. The situations that are characterized by time adverbials can only be thought of as moving in the same direction as the observer. Thus, in the case of the anterior-durative and posterior-durative functions the question of which model is chosen can again be answered unambiguously.

There is a slight asymmetry in Tables 7 and8: It is apparently significant that more languages use an ablative marker for posterior-durative than an allative marker for anterior-durative. This latter function tends to have a separate marker which just means 'until'. I have no explanation for this asymmetry.

Before leaving the allative/ablative sources of sequential-durative markers, I would like to point out a further cross-linguistic generalization. The tendency for posterior-durative and anterior-durative markers to be similar to simple source and goal markers is even stronger when both markers are present together, i.e. in expressions specifying both a beginning and an end of a situation, e.g. 'from morning till evening', 'from five to eight'. In these expressions (henceforth, beginning-to-end constructions), the order is obligatorily iconic, i.e. the posterior-durative expression precedes the anteriordurative expression, and the two are preferably adjacent. Similarly, German usually has seit/von..an and bis, but in beginning-to-end constructions von alone can be used (however, German does not allow the simple allative zu instead of bis here). The most frequent case is illustrated by the (a) sentences of E59-60. The (b) sentences show that the simpler markers from...to and von are not possible if only the beginning or the end of the duration are given. The difference in acceptibility between the (c) and (d) sentences is due to the iconic order requirement, and the difference between (e) and (f) is due to the adjacency requirement.

Before venturing an explanation of these facts, let us look at the cross-linguistic data. Of course, I do not have data of the same level of detail for other languages, but since quite a few other languages show the same tendency for a less explicit form of begining-to-end constructions, it may well be that restrictions similar to those in E59-60 also apply in other languages. In all cases of Table 9, at least one of the markers in the beginning-to-end construction is simpler than the regular sequential-duration marker and is identical to the usual (spatial) source or goal expression in the language. The explanation clearly lies in economic motivation: In beginning-to-end constructions, the lexical content of the expression, together with the iconic order and the adjacency (cf. E59-60), are by themselves so informative that simple source and goal expressions are often sufficient where otherwise a more explicit marker would be used. This economic aspect is particularly evident in cases where a complex sequential-durative marker consists of a simple source or goal marker plus an additional component (e.g. Italian fino a), and where this additional component is omitted in beginning-to-end constructions. Such cases are French (jusqu') à, Italian (fino) a, Latin (usque) ad, Irish go (dtí), Turkish -den (beri/itibaren), Hungarian -tól (fogva).

'Beginning' and 'end' as sources

In a number of languages, the words 'beginning' (or 'to begin') and 'end' (or 'finish') are the sources of the sequential-durative markers. The relevant data are summarized in But it is more likely that in the course of grammaticalization the original entailment got lost and persists nowadays as an implicature.

A source of 'until' that is related to 'end, finish' is the verb 'arrive, reach', which is employed in the anterior-durative function in a number of languages, which are also listed in Table 10b. Again, the notion of 'arrive' is transferred from the spatial domain to the temporal domain, and the semantic correspondence is quite transparent: If someone travels along a path and arrives at a place, this means that the path (which is analogous to the duration of a situation) overlaps with this place. The meaning of 'arrive' is also similar to that of 'until' in that the notion of someone's arrival at a place does not exclude a continuation of the trip -one can arrive at intermediate goals, whereas one cannot continue a trip after finishing it.

'Since' from 'later'

In a number of Germanic languages, the 'since' word is based on an original comparative form of an adjective meaning 'later', i.e. 'since X' comes from 'later than X': à partir de 8 heures. 'Last year I was at the office every day from eight o'clock.' Thus, seit, since and depuis have the more specific function called "posteriorpresent-perfect" (cf. §2.3), covering only part of the domain of the posteriordurative function. Markers with this deictic meaning component are also attested in quite a few other languages. I do not have the same detail of information for other languages, but I can make a few generalizations and suggest some hypotheses. My data are not generally sufficient to determine whether a posterior-durative marker is analogous to German seit ("present perfect"), von..an ("past" or "future") or ab (only "future"), and in most cases all I can make is a past-future distinction. Monday-from-TOP this-in be-FUT.1

'I'll be here from Monday on.'

3 In the COMRIE-SMITH grammars, sections 2.1.1.6.8 and 2.1.1.6.9 distinguish between "posterior-durative-past" and "posterior-durative-future". Note that the present seems to pattern with the future in general (*seit jetzt/von jetzt an; *since now/from now on; *depuis maintenant/à partir de maintenant).

Since there are quite a few languages for which my information is insufficient (there is no relevant future context in the New Testament), I take the situation illustrated in E70-72 as the default, and the distinction between "posteriorpresent-perfect" markers and "posterior-durative-future" markers as the special case. The languages for which I have found a distinction between past and future and their markers are listed in Table 11. In some languages the past marker is clearly the more specific one because the other marker is not restricted to future contexts. This is the case in Japanese, for instance.

E73. Japanese (KAORU HORIE, p.c.)

a. Kare-ga sensyuu kara/irai Nairobi-ni iru. To some extent, this is also true of German von..an and English from..on, although these are in complementary distribution with seit and since. But as shown above in E66, they are not restricted to future contexts.

However, there is also evidence that the past context is the more typical context for posterior-durative markers, because the future marker is sometimes transparently derived from the past marker. Thus, a number of languages have a structure like 'from X forward' in future contexts: Italian (in poi, lit. 'in then', i.e. 'later'), Modern Greek (kj épita 'and then'), Irish (amach 'away, out'), Chinese (qǐ 'up, rise'). Of course, the English and German pattern is similar to these, although on/an is not added to the past marker since/seit, but to the general source marker from/von.

But the reverse pattern also occurs, with the past marker being a more specific version of the general posterior-durative marker. This is the case, in particular, in languages that incorporate the element 'hither' in their past posterior-durative marker. Due to the presence of this element, we can be confident that these expressions are semantically much like since, i. Alates eelmise-st nädala-st on ta ol-nud Nairobi-s.

beginning last-ELAT week-ELAT is he be-PTCP Nairobi-INESS 'Since last week he has been in Nairobi.'

In all other languages, the present tense is used in these contexts. This may seem surprising, given that many other languages have a perfect tense-aspect form that is not unlike the English, Swedish and Baltic Finnic perfect in other respects. We must conclude that this use of the perfect in these languages is highly marked.

Finally, let us ask whether a contrast similar to the one between 'since' and 'from..on' could be found in the anterior-durative function. A priori, there is no reason why it should not be possible. There could be a language that distinguishes between 'until1' for a situation like Figure 31a, and 'until2' for a situation as depicted in Figure 31b. This distinction would be completely analogous to the distinction between 'since' and 'from...on', see Figure 16-17 in §2.3. However, such a distinction is unattested, and there are good reasons to think that it does not exist. Not only has no such distinction been noted for any of the well-documented European languages; even the Comrie-Smith grammars, which do include two separate sections for "anterior-duration-past" (2.1.1.6.6) and "anterior-duration-future"

(2.1.1.6.7), do not record such a contrast, which was clearly anticipated by the authors of the questionnaire.

Anterior-durative and anterior-limitative

In this final subsection of this chapter I will introduce a semantic function that is The anterior-durative and anterior-limitative functions are similar in that they say something both about the reference time and about the period before it.

While anterior-limitative has the paraphrase 'at RefT or earlier', anteriordurative can be paraphrased as 'at RefT and (at all points) earlier'; thus, the difference boils down to the conjunction/disjunction in the paraphrase (see KÖNIG (1974) for a logical-semantic analysis of this distinction in terms of the duality of existential and universal quantification). These are not enough cases for definitive generalizations, but it seems that there is a tendency for anterior-limitative markers to be based on directional spatial markers with meanings like 'toward, against, onto'. In this respect, too, the anterior-limitative function resembles the anterior-durative function. 5 5 A related semantic function which I have only encountered in French so far is illustrated in (i) (cf. the discussion in BERTHONNEAU 1993a).

(i) Il me faut une réponse avant deux jours. sous 'I need an answer within the next two days.'

Here the anterior-limitative function is combined with the distance-future function. Literally avant deux jours/sous deux jours could be translated as 'by in two days' time'. This combination is analogous to the one in the distance-posterior function (which combines posterior-durative and distance-past, cf. §2.5 and §8.3).

Chapter 6 Temporal distance

Distance markers based on sequential markers

The two semantic functions of temporal distance, distance-past and distancefuture, are appropriately discussed right after the semantic functions of sequential location because they also involve temporal location in which the located situation and the reference time are in a sequential relationship. This semantic affinity is reflected in formal identity or similarity between the distance-past and the anterior functions, and between the distance-future and the posterior functions. Indeed, this similarity is much greater than that between the anterior/posterior functions and their durative counterparts, which a priori would seem to be more closely related semantically.

As in the other cases, there are a small number of different semantic sources for distance-past and distance-future markers. However, the majority type is clearly the formal identity between distance-past markers and anterior markers on the one hand, and distance-future and posterior markers, on the other hand.

The twenty-five languages of the sample in which this occurs and their markers are listed in Table 13 on the next page.

What is the explanation for the frequent formal identity of the sequential and distance functions? First of all, we have to distinguish carefully between two sub-types of markers in Table 13. The first sub-type, exemplified by German vor (vor Dezember, vor drei Minuten), shows both morphological and syntactic identity. In the second sub-type, only the form, but not the syntax, of the two markers is identical.

Let us examine the case of Turkish önce 'before; ago', which belongs to the second type. In the sense 'before', this is a postposition governing the Ablative case (-dAn), as shown in E81a. However, in the distance-past function, önce does not govern the Ablative, but the Nominative case, as seen in E81b. In E88b, the distance phrase is in the Instrumental case and the argument of the postposition is spelled jointly with the postposition -this is analogous to Turkish E82. The other construction in E88c shows the reference time (három hét 'three weeks') in the Nominative case, just like the reference time in the sequential construction E88a -it is thus analogous to German vor.

Thus, a different explanation has to be found for the formal identity be- the reference time must be taken as a period one of whose end points coincides with the moment of speech. Thus, here again a deictic interpretation is imposed but remains implicit. We can thus translate Bulgarian sled dva dni literally as 'after a two-day period beginning now', and Hebrew lifney šloša yamim as 'before a three-day period ending now'. The possible alternative interpretations 'after a two-day period ending now' and 'before a three-day period beginning now' are presumably excluded pragmatically because they are not informative enough, coinciding more or less with 'after now'/'before now'. But the interpretation of these expressions is enriched even beyond the introduction of the moment of speech as an implicit end point: Bulgarian E89a in fact means 'immediately after a two-day period beginning now', and Hebrew E89b means 'immediately before a two-day period ending now'. Again, these pragmatic strengthenings are entirely expected from a Gricean point of view. As KÖNIG (1974: 560-61) observes, "[w]henever we place a certain event on a time scale we try to do this as precisely as possible. Thus, a sentence of the form 'S after t'

will normally be taken to imply that S occurred in the environment of t ... [O]n hearing a sentence of the type 'S after t 1 ' we will assume -unless we are explicitly told not to do so -that the speaker has been maximally informative, that he has placed the event 'S' as precisely as possible on the time continuum and that the event occurred near t 1 ." KÖNIG only tries to explain why a politician who promises 'We shall start working on this problem after Easter' will be justly criticized if nothing happens until December, but the same principle also explains why E89a means 'two days after now', not just 'anytime after a two-day period beginning now'. But note that in the case of 'before' and 'after', the interpretative enrichment can be canceled and thus has the status of an implicature.

Other sources of distance markers

In the distance markers discussed in the previous section, we saw that distancepast and distance-future markers behave in a relatively symmetrical way, except that there are more distance-past markers based on 'before' than distance-future markers based on 'after' (possibly this is related to the analogous asymmetry observed in §4.2 with regard to Table 4, but it is not clear to me how). When we now look at the other sources of distance markers, we notice that there is mostly a surprising lack of symmetry. Very different kinds of source notions are made use of for expressing the semantic functions 'distance-past' and 'distance-future'.

'Pass'/'exist'

In a number of languages, the source of a distance expression is the verb 'pass', or the more general verb 'exist, be'. The simplest way in which this can serve to express distance is by occurring in a temporal adverbial clause with the distance phrase as its subject, e.g. 'When five years have passed/When it's five years' for 'in five years' time'. This strategy is used, for instance, in Lithuanian and

Hungarian. In these languages, the converbal form of the verb 'pass' is used.

Since I do not have much evidence that these verb forms have become grammaticalized along the grammaticalization path from converbs to adpositions (cf. KORTMANN & KÖNIG 1992 At the same time, it is clear that French il y a introduces a dependent phrase, and is not an independent juxtaposed clause, because il y a phrases may be 3 This is also true of Spanish hace (past hacía), which makes it very unlikely that ELERICK's (1989) etymology of Spanish hace is correct. He derives hace not from the 3rd person singular present form of hacer 'make', but from Latin *abhince 'hence'. Or at least, if this is the historical orgin, it is clear that at some stage hace was identified with the form of hacer.

focused (cf. E100a) and be the object of the preposition jusque (cf. E100b) (HENRY 1966(HENRY :211-12, 1968; see also GROSS 1986, BERTHONNEAU 1993b 'Fourteen years ago, this man...' Thus, in these cases the distance-past meaning 'two hours ago' results from a source which means 'at the time of the past two hours', i.e. 'at the beginning of the past two hours'. A period of time is used here to refer to an extreme point of this period. This is not so unusual, as we will see in the next subsection.

'Within'

There are quite a few languages that are like English in that they use a marker which also does duty for the spatial inclusion These examples from languages in three continents are sufficient to demonstrate beyond doubt that the use of in in English and German is not an accident. This is underlined additionally by the fact that a number of languages use their emphatic, less grammaticalized variant of 'in', i.e. an adposition meaning 'inside, within' (Spanish dentro de, rather than en; French dans rather than en;

Punjabi andar in addition to vicc; in Finnish kuluessa 'in the course of' is an alternative to the Inessive case). We thus need to establish a semantic connection between 'inside, within, in' and the notion of distance-future.

My proposed explanation takes the following form. I start from the observation that within in English (and analogous adpositions in other languages, e.g. innerhalb in German) can be used to denote the telic-extent meaning when used with accomplishment predicates. In contrast to in, the more emphatic within/innerhalb highlights the fact that the completion of an event does not exceed certain temporal boundaries.

E104. English/German

They tore down the house within five hours.

Sie haben das Haus innerhalb von fünf Stunden abgerissen.

When used with a future tense, within/innerhalb is already fairly similar semantically to the distance-future function:

E105. English/German

Bob will make 77 cookies within one hour.

Bob wird innerhalb einer Stunde 77 Plätzchen backen.

Under the assumption that Bob will start making the cookies now (again, the moment of speech can be left implicit as a default), the event will be completed one hour from now. The next step is the crucial one: the extension to verbs with different aspectual properties that cannot normally cooccur with telicextent time adverbials, e.g. momentary verbs ('explode', 'criticize') and durative atelic verbs ('dance', 'work'). These do not combine happily with within/innerhalb, but if they cooccur with within/innerhalb, the resulting interpretation is even more similar to the distance-future meaning.

English/German E106. The bomb will explode within an hour.

Die Bombe wird innerhalb einer Stunde explodieren.

E107. We will work within a month. Wir werden innerhalb eines Monats arbeiten.

E106 is compatible with a distance-future reading, although it is also felicitous if the event takes place at any time that is less than an hour away from the moment of speech. The same is true for E107, which is felicitous only if the activity predicate is used with a perfective, i.e. inceptive interpretation. The final step in the semantic development has been taken once the location is understood more specifically as the end of the time period, and the beginning of the time period is fixed as the moment of speech. Now that I have proposed a possible scenario for the semantic development from 'within' to distance-future, we can also explain the asymmetry with respect to the distance-past function, i.e. why E108 is a common development, while E109 is unattested. The oddity of the development in E109 is derived from the oddity of sentence E109a itself. 'Within' would be used with past situations only if multiple events are involved, e.g. 'I received sixty-nine letters within a month'. With single past events, the precise location is usually known, so that the specification 'within X' makes the sentence less precise than necessary. 4

'Back'

In a number of languages, the spatial directional adverb 'back' is used for the The most interesting point about this use of 'back' is that it contrasts strikingly with the use of 'before' or 'in front' for expressing the same semantic function. Clearly, in this case the image of the observer moving forward in stationary time is predominant, whereas in cases like Turkish önce 'before, ago' the image of time moving toward the observer is predominant. It is perhaps not an accident that the instances of 'back' in E99-102 all seem fairly young, whereas the cases of 'front' for the analogous function in Table 13 above are in part much older. In general, the image of moving time appears to be expressed by less surprising, less figurative, more abstract language, whereas the image of the moving ego is expressed by more figurative language.

'Over'

In a number of languages, a spatial marker that means 'over, across' also serves as a temporal marker of the distance-future function. This usage, too, can be understood on the basis of the movement metaphor for time. But in contrast to the use of the verb 'pass', where it is time that moves, the use of 'over' for distance-future is based on the moving ego. 'Over two hours' is short for 'when we have passed the mark of two hours', just like 'over the bridge' is short for 'when the subject has passed the bridge'.

'Yet'

In two languages I have found a marker of the distance-future function that has no relation to space: the additive phasal time adverb 'yet, still' (German noch)

(see VAN DER AUWERA (1997b) for a thorough cross-linguistic study of this and related adverbs). The two languages are Hebrew and Indonesian:

5 Some varieties of German have an anlogous use of über, cf. (i) from a well-known folk song:

(i) Übers Jahr, übers Jahr, wenn mer Träubele schneidt, so soll die Hochzeit sein. 'In a year, in a year, when we reap the grapes, then will be the wedding.' The converse of 'yet, noch' is 'already', so we might expect that some languages use their 'already' word for the distance-past function. I have not found an example of this, but this may be an accidental gap in my data, because 'already' is widely attested in the distance-posterior function (cf. §8.3.2), which is closely related to the distance-past function and is commonly expressed by similar formal means.

Deictic and non-deictic distance expressions

So far we have been exclusively concerned with deictic distance expressions, i.e.

temporal adverbials locating a situation at a certain temporal distance from the moment of speech. But we saw already in §2.4 that distance expressions can locate situations also with respect to other time locations. First, let me say a few words about distance expressions like a year after the Kobe earthquake, which measure the distance between the located situation and an explicitly given can also be used when the distance is measured not from an explicitly given reference point, but from a reference point that is implicit in the context (hence, KUČERA & TRNKA (1975:38) and KLEIN (1994:156) call such expressions 'anaphoric'). In many languages, a different marker must be used when the are easy to find. In 13 of these 27 languages, the distance-prospective marker is identical to the distance-future marker. Thus, the deictic/nondeictic distinction is made by a substantial proportion of the languages, as in English, but there are also many languages in which this distinction is lacking. The data are summarized in Table 14 on the next page. In addition to the distance-future and distance-prospective markers of the 27 languages for which data are available, I have included the posterior markers in Table 14, because these are also often formally identical or related (marked by "=", "(=)", or "≈" between the distanceprospective and posterior columns).

A number of observations can be made on the basis of Table 14. First, within Europe, there is a clear areal patterning in the distribution of the deictic/nondeictic distinction: Western and northern European languages tend to make this distinction, whereas eastern European languages tend to lack it. Interestingly, Latin and Ancient Greek (which is not in the sample but is included in Table 14) pattern with the eastern European languages, not with Romance and Modern Greek. This is of course a familiar pattern in the areal typology of Europe, and it suggests that the existence of the deictic/non-deictic distinction should be included in the list of features identifying Standard Average European (cf. VAN DER AUWERA (1997a) for some recent discussion of this Sprachbund).

Second, the data in Table 14 suggest the following implicational generalization: If in a language the posterior and the distance-future markers are identical, then the distance-prospective marker also takes the same form. This gives us some additional insight into the identity of these two markers, which was already discussed in §6.1 above. It is clear that in languages where all three markers are formally identical (e.g. Ancient Greek, Bulgarian, Arabic, Japanese), the distance markers are derived from the posterior marker, and the distance-future function is treated as just a special case of the distanceprospective function.

Among the languages that distinguish the distance-future from the distanceprospective function, those that show the 'within' type of distance-future marker (discussed in §6.2.2) are particularly prominent. 'Within' markers never express both distance-future and distance-prospective, they are always purely deictic. It is not clear to me why this should be so.

Another observation is that when the posterior and distance-prospective markers are not identical, then the distance marker tends to be an adverb which combines with a (mostly preceding) distance phrase. The adverb can be a comparative of the adjective 'late' (as in English, German, Swedish, French, Modern Greek, Finnish, Indonesian), or based on a spatial adverbial ('behind, in back, afterward'; Spanish, Maltese, Estonian).

Before concluding this chapter, I would like to discuss briefly expressions such as E126a-b (cf. also E99b from French, which is similar). These show an apparent non-deictic use of the otherwise exclusively deictic adpositions ago and in, being synonymous with the unidiomatic phrases three years before today, and eine Woche nach morgen, respectively. In their discussion of ago, KORTMANN & KÖNIG (1992:678) Thus, the only possible explicit reference points are heute, morgen and gestern, plus compounds derived from the two latter words. 9 While I have no explanation for the nature of this restriction, I feel that these cases are not sufficient to question the basically deictic meaning of ago, in, etc. These expressions contradict my analysis somewhat, but they should be analyzed as idiomatic, analogous to completely idiosyncratic and non-compositional phrases like tomorrow week 'a week from tomorrow'.

8 KORTMANN & KÖNIG also claim that ago, like before (and like from in their distanceprospective example three years from now), is a preposition which in three years ago today takes today as its argument, often allows an unexpressed argument (resulting in a deictic interpretation), and requires an obligatory "specifier" (i.e. preposed distance phrase). This analysis, though ingenious, is implausible because of the marginal status of the "argument" following ago (see E127). Its main virtue is that it identifies the somewhat unclear syntactic status of today with the well-understood status of being an argument of a preposition. But this analysis of English ago does not extend to German E126b, where morgen intuitively has the same function as English today in E126a. (To be consistent, KORTMANN & KÖNIG would have to analyze German in as a postposition followed by a Dative-marked distance phrase, which I doubt they would be prepared to do.) 9 Incidentally, I find überüberüberübermorgen vor einem Jahr much better than the synonymous *in fünf Tagen vor einem Jahr 'a year ago in five days', although the former is almost unprocessable. This shows that the constraint is lexical rather than semantic.

Chapter 7 Simultaneous temporal location

An implicational map for simultaneous location markers

In order to capture some regularities in the distribution of various markers (adpositions and cases) over different types of time unit nouns, I propose to make use of the methodology of implicational maps that has been pioneered by L. ANDERSON (1982) and used and further developed in KEMMER (1993), HASPELMATH (1997: Ch. 4) and KORTMANN (1997: Ch. 7). An implicational map (also called "semantic map", cognitive map") shows a geometric arrangement of several (semantically or otherwise) distinguishable functions or uses that a grammatical element may have. A grammatical marker in a given language may express a range of different functions, but these functions must always be adjacent to each other on the map, i.e. a marker must cover a contiguous area.

Thus, an implicational map makes the prediction that only a subset of the logically possible patterns of polysemy actually occur in languages, i.e. it expresses the universal constraints on polysemy or polyfunctionality of grammatical markers. An implicational map can be seen as an abbreviatory statement of a number of implicational universals (HASPELMATH 1997). At the same time it shows the patterns of relatedness among different functions, because adjacency on the map is always due to functional closeness.

The fewer connections exist among the functions on an implicational map, the greater the constraints on polyfunctionality. In the case of time period nouns functioning in simultaneous adverbials, the implicational map that I propose shows nine connections among six functions, which is not optimal (the most restrictive map would show just five connections), but it does capture some restrictions (the most permissive map, which allows any combination of functions in a marker, would have fifteen connections). The map I propose for the six time period types hour, day part, day, month, season, year is shown in Unfortunately, however, this implicational map is not without exceptions. The clearest violation of the map is found in Nkore-Kiga, which according to TAYLOR (1985) shows the preposition aha for the functions (1), ( 3) and ( 5), and the preposition omu for the functions ( 2), ( 4), and (6).

Nkore-Kiga Perhaps omu can also be used with days (cf. TAYLOR 1985:118), which would remove the anomaly in the distribution of this preposition, but aha still violates the implicational map. In Finnish, the Essive case (-na) is used with days and years (vuon-na 1990 'in 1990'), but not normally with any of the other time units (however, it is used with other time units when these have a preceding modifier, cf. E144 below). Finally, in Italian the preposition a is only used with hours (alle due 'at two o'clock'), day parts (alla mattina 'in the morning'), and months (a maggio 'in may').

Thus, the implicational map does not represent an absolute universal, but only a tendency. Nevertheless, as such it retains its usefulness. In the following sections, I will try to explain why the time unit types are arranged on the map the way they are, and I will make further comments on the individual markers, to the extent that generalizations emerge from the data.

Location in hours

As I mentioned above, the hour is the only time unit that is often (indeed, usually) employed to indicate temporal location at a point rather than in a time span. We can say 'in the seventh hour', but we are much more likely to say 'after six o'clock' or 'between six and seven', using non-simultaneous markers. This is apparently true for most languages. The hours are used with simultaneous location markers mainly when a point in time, or at least a more specific temporal location is intended, e.g. 'at seven o'clock'.

On the basis of this semantic difference, one might think that hours tend to be marked with one-dimensional spatial markers, whereas the other time units take other spatial markers (cf. WIERZBICKA 1993). However, there is not much evidence bearing out such an expectation. Besides English (at), only the Romance languages could be cited, which use their preposition a/à, contrasting with en/in for most other time units (cf. MEYER-LÜBKE 1899:492).

In about half of the languages of my 53-language sample, the marker used for simultaneous location of hours is the same as that used for several other or the majority of time unit types. In about ten languages, a marker that is more specific than the other markers is used for hours ; më 1912 'in 1912'). This special relationship between hours and years may be due to the fact that both hours and years are usually named by numbers (days of the week and months are also numbered in some languages, but this is much rarer).

Location in day parts and seasons

The parts or periods of the day, called "day parts" here for short, are of course more heterogeneous than the hours. While there is a tendency for the different day parts to behave in a parallel fashion (in the morning, in the afternoon, in the evening; Russian Instrumental case in utr-om 'in the morning', dn-em 'during the day', večer-om 'in the evening'), there are numerous instances in which different day parts are marked in different ways. Thus, Irish has ar maidin 'in the morning', san oíche 'at night', um tráthnóna 'in the evening'; Italian has all'alba 'at dawn', di mattina 'in the morning', nel pomeriggio 'in the afternoon'; Lezgian has ekünaqʰ 'in the morning (Superessive case), jifiz 'at night' (Dative case), näniz/näniqʰ 'in the evening', nisiniz/nisiniqʰ 'at noon (Dative or Superessive case). It is not possible here to mention all these differences, and it would be very difficult to generalize over them.

There are two reasons for the variation that we observe in day part nouns:

First, day parts are generally very frequent in discourse, and due to their high frequency, their combinations with grammatical markers tend to be lexicalized.

Thus, there is not as much analogical pressure on day part adverbials as there is on hour adverbials. And second, different day parts may have different semantic properties (cf. WIERZBICKA 1993): While the morning and the evening are semantically quite parallel, the night is a much longer period and has a very different function for people. And dawn, dusk and noon are probably thought of as points in time rather than time spans.

In particular this latter hypothesis is confirmed in quite a few languages:

Words for 'noon ', 'midnight', 'dawn', etc. tend 'at midnight' 'at ten o'clock' 'in the morning'

The formal marking of day parts in simultaneous location adverbs is in several ways similar to that of the seasons, which is not too surprising in view of their semantic similarity. Both day parts and seasons are not successive time units, but rather periods that are qualitatively different from each other. Names of months and of days of the week are not more than names for time units defined by their position in the calendar, whereas day parts and seasons are defined by their qualitative characteristics much more than by their calendar position.

The first shared formal property is that they tend to be lexicalized in their adverbial form. Both Udmurt and Chechen have a fairly rich case system, and they could easily use a locative case for the simultaneous location in day parts or seasons. But the adverbial forms in E136-137 are quite irregular. And even in Russian, whose Instrumental case of day parts (utr-om 'in the morning') and seasons (let-om 'in the summer') is quite regular from a formal point of view, a widespread view of these forms is that they are fixed adverbials rather than case forms (cf.

KUČERA 1966 for some discussion).

Hungarian nicely shows how an oblique case form of such a noun may become fixed as the base form of the noun. In this language, reggel means 'morning' or 'in the morning', and éjjel means 'night' or 'at night'. These forms are evidently old Instrumental case forms (-Al plus gemination of the preceding consonant) of the roots reg-and éj-, 2 which have been reinterpreted as base

2 The Instrumental case is also used with the seasons, e.g. összel 'in the fall', tavasszal 'in the spring'. The old form éj is preserved in compounds such as éj-szaka ('night period') and in fixed phrases such as jó éjt 'good night'.

forms because of their high frequency. 3 The result is that the adverbial form of these nouns is now zero-marked. A similar process may also account for zero marking in Nanay, which is otherwise a highly inflecting language:

E138. Nanay ). Thus, the link between day parts and seasons on the implicational map is well-motivated.

To conclude this sub-section, let me say a few words on the seasons. These examples seem to show that there is a tendency for winter and summer to pattern together, contrasting with spring and fall. Moreover, Hungarian, Polish and Georgian are surprisingly similar in that they seem to treat summer and winter more like containers (perhaps because they are the prototypical seasons), whereas spring and fall are treated differently (like surfaces?)

(perhaps because they are conceived of as transitional between summer and winter).

Location in days, months, years, and festivals

Days, months and years are the prototypical time units, and it is here that we find the most typical simultaneous markers. One general observation is that as a rule simultaneous location in time periods is expressed by fairly abstract spatial markers, often the most grammaticalized spatial markers. Even in languages with only a moderately rich case system, it is often case markers rather than adpositions that are used for simultaneous location. Not uncommonly, even abstract non-spatial cases are used for this function, e.g. the accusative in Lithuanian, Modern Greek and Imbabura Quechua, the Dative in Lezgian, the Essive in Finnish, the Ablative in Latin. But in many other cases a 4 However, d'autunno is also possible. (The French contrast between en hiver, en été, en automne, but au printemps can be explained diachronically: en + le was originally contracted to au, just like à + le (MEYER-LÜBKE 1899:493). Modern French has preserved quite a few cases of au alternating with en in this way.) more specific spatial marker is used, and this generally points in the same direction: interior spatial markers ('in'), as used for three-dimensional inclusion.

Since the time line is one-dimensional, the a priori expectation would be that temporal location tends to be expressed by means of an 'on' marker. This is indeed occasionally the case (English on (Friday), Hungarian Superessive case, Swedish på), but 'in' markers clearly predominate.

The position of the time unit 'day' on the implicational map is not surprising -it is adjacent to the next smaller time unit (hour), to the next larger time unit (month), and on the other hand, to the qualitative periods of the day. Similarly, the time unit 'month' is located, as expected, between the day and the year on the map. Otherwise there is not much more that can be said about location in days and months. Complications may arise from the different treatment of different kinds of day/month expressions. Thus, in Italian names of the month can take the prepositions in or a (in maggio/a maggio 'in May'), whereas the word mese 'month' only takes in (nel/*al mese di maggio). This variation is even more widespread with days. Irish has zero marking with names of days of the week (tiocfaidh sé Ø Dé Luain 'he is coming on Monday'), the preposition ar with dates (ar an dara lá déag de Lúnasa 'on the twelfth of August'), and the preposition i with plural day words (sna laethanta sin 'in those days'). When there is variation, I have generally chosen the marking of the days of the week for the data set listed in the Appendix.

While the expression of temporal location in years exhibits no peculiarities worth commenting on further, the marking of festivals is special in many languages. This function has not been included in the implicational map of

Zero or minimal marking with certain modifiers

Zero marking of temporal location is not uncommon across languages. We saw individual examples of this in previous sections, and there are languages that have zero marking in all or almost all simultaneous functions, e.g. Swahili and

Kobon. Particularly day parts, days of the week and seasons tend to be zeromarked in my data.

However, in addition to these individual cases, there is a class of expressions that systematically exhibit zero marking in a substantial number of languages. Thus, the phenomenon is extremely widespread and is by no means restricted to European languages. How can we account for these contrasts? MCCAWLEY (1988) discusses the English data of E140 and proposes that they should be analyzed as prepositional phrases with a zero preposition. 6 6 MCCAWLEY's article is a reply to LARSON's (1985) article, which proposes that nouns such as those in E140 have lexical entries containing a feature that allows them to assign Case to NPs of which they are the head. As MCCAWLEY notes, this does not permit one to account for contrasts such as those in E140, where morning, Friday, summer etc. occur with or without a preposition, depending on its modifier. MCCAWLEY gives two empirical arguments for his "zero preposition" proposal which are supposed to show that these phrases behave like adverbials which are prepositional phrases, not like adverbials in general (cf., e.g., the word order possibilities in Smith may have *that day/*on a subsequent day/subsequently withdrawn his lawsuit). I find it more straightforward to say that both that day and on a subsequent day are NPbased time adverbials, contrasting with the adjective-based time adverbial subsequently. But whatever conceptual choice one makes in describing the alternations noted in this section, the real challenge is to explain why certain modifiers should allow time unit nouns to dispense with an adposition or case that they normally require when used in simultaneous location adverbials. To my knowledge so far nobody has even asked this question, let alone proposed an answer.

However, it can be shown that this proposal fails to account for languages in which the contrast does not reduce to the presence or absence of a preposition.

First, note that the Indo-European and Semitic languages with a well-developed case system and a nominative-accusative contrast tend to show the accusative here (in the above examples, German, Russian, Croatian and Arabic, as well as Georgian), even if the preposition in the non-modified form does not govern the accusative case. Thus, the contrast is not just due to the omission of the preposition. Second, in some other languages with a rich case system which includes a locative or other concrete case (e.g. Japanese -ni, Quechua -pi, Finnish -na), this case suffix may be dropped and the NP may occur in its base form.

Again, while it is obvious that the contrast is completely analogous to the one in English, it cannot be described as a contrast between noun phrases and prepositional phrases.

The direction in which I would speculate toward an explanation of this phenomenon is that it is due to some kind of economy, at present ill-understood. When certain frequent modifiers are present in the temporal NP, this appears to give the NP more weight and allows it to be marked less explicitly.

In some languages, this means that an adposition is dropped or an oblique form is replaced by the base form. But economy is also served if the case marking is minimal, e.g. accusative in Indo-European, Arabic and Georgian.

Apparently the nominative case is not suitable for this purpose in these languages, perhaps because it is not morphologically less marked than the accusative case in these languages. The accusative is as unmarked (and hence as economical) as the nominative, and as it is the case that is in general more versatile semantically, it is chosen over the nominative. This tentative explanation would perhaps also cover the cases of Finnish and Georgian. In Finnish, the Nominative case is used with joka 'every' (cf. E143d), but with the other modifiers, the Adessive case is replaced by the Essive case:

E144. Finnish The dependence of the case marking of the NP on the modifier seems quite puzzling, but perhaps it can be argued that the Essive case is less marked than the Adessive case: Formally it is shorter in that it lacks gemination, and semantically it is at least as general in that it occurs with years, days and festivals, whereas the Adessive occurs only with day parts and seasons.

Similarly, the Georgian Accusative/Dative in -s is less marked formally and semantically than the Locative in -ši.

I cannot claim that I have explained the phenomenon described here, but in any case the cross-linguistic data cited here for the first time permit us to recognize the surprising generality of the phenomenon, which must ultimately be explained in universal term. 'Can't you stay awake together with me even for an hour?'

In languages that have a well-developed case system, including grammatical cases such as nominative and accusative, atelic extent adverbials tend to be marked by the accusative case, as illustrated in E145. In languages lacking a case system or at least a nominative-accusative opposition, atelic extent adverbials are in the basic form, i.e. they show zero expression. However, in this case it is much less clear than in the case of temporal location that a conceptual transfer from the spatial to the temporal domain has taken place. In contrast to spatial markers like 'within', 'in front', 'in back', which are often ultimately based on body-part terms or spatial landmarks (SVOROU 1994), the accusative case or zero has nothing inherently spatial about it.

Furthermore, spatial extent adverbials such as the examples in E147 are not particularly frequent in discourse, probably less frequent than temporal-extent adverbials. Thus, the hypothesis of a transfer from the spatial domain does not explain much.

One might speculate that zero marking is chosen in these languages simply as a kind of default -NPs with a noun denoting a time unit inherently denote a temporal extent, and no additional marking is necessary. All other time adverbials based on time unit expressions are more specific, e.g. distance expressions ('two days ago', 'four weeks from now') and telic extent adverbials (e.g. '(finish a job) in three hours'). This would account for the widespread zero expression of atelic extent adverbials, and the accusative marking in languages with a clear nominative-accusative apposition might be due to the fact that the nominative is reserved for the subject, whereas the accusative is the minimal, i.e. least specific non-subject case that is available for adverbial use. 1 The default-case hypothesis receives additional support from the fact that the accusative case is also widely used in a certain class of simultaneous locational adverbial NPs, as we saw in §7.6. Semantically, locational adverbials like diesen Monat (this.ACC month) and extent adverbials like einen Monat (one.ACC

1 Quite generally, the direct object differs from the subject semantically in that it allows a much wider range of semantic roles. One might object that in Greenlandic Eskimo, an ergative language, the Absolutive case is used, which is a subject case in intransitive clauses. However, intransitive subjects generally share with direct objects the property of allowing arguments of a wide range of semantic roles.

month) have little or nothing in common, but in both cases it can be argued that the temporal relation is relatively predictable, so a default marking is sufficient.

A second possibility is that the accusative case is motivated in a positive rather than in a purely negative way. One might propose that the atelic-extent function is often expressed by the accusative case because speakers assimilate it in some way to the direct object of their language. This hypothesis (the "directobject hypothesis") would not only be compatible with the data from those languages in Table 15 that show accusative case marking, but also with almost all those languages that show zero-marking -these languages also have zeromarked direct objects. 'I have taken this pill' 'I have taken one pill.'

The same contrast is found with atelic-extent adverbials. The most common case is that shown in E149a, where the adverbial is indefinite, but in those infrequent cases where it is definite, as in E149b, the adverbial has the case marker -râ. 'I worked in the garden for the last two hours.'

2 A problem for the direct-object hypothesis is Japanese, which consistently marks direct objects with the postposed particle o, but zero-marks atelic-extent adverbials. However, the direct-object marker o may be omitted in certain styles.

The direct-object hypothesis receives further support from the fact that atelic-extent adverbials sometimes behave like a direct object in other respects besides case marking. Thus, in Mandarin Chinese atelic-extent adverbials can only come in the position after the verb, not in the pre-subject position or in the position between the subject and the verb. But why should extent adverbials be modeled on direct objects? Here we should note that atelic-extent adverbials and spatial-extent adverbials as in E147 are not the only cases of direct-object-like extent phrases. The "direct objects" of verbs like 'weigh', 'cost', 'last' are also extent expressions from a semantic point of view. The common semantic denominator of extent phrases and direct objects can perhaps be found in their bounding function. In some languages, e.g. in Russian, the boundedness of a verbal situation can easily be tested because only bounded situations may occur in the perfective aspect. Thus, the atelic verbs in E158 do not have perfective counterparts.

E158. Russian a. Kolja paxal.

(*Kolja vs-paxal.)

Kolya plowed (Kolya PFV-plowed)

b. Olja rabotala. (*Olja pro-rabotala.)

Olya worked (Olya PFV-worked)

However, when the same verb occurs in a syntactic-semantic context in which the situation is bounded, e.g. when there is a direct object or an atelic-extent phrase, the verb may also be perfective, e.g. 'for two years' resulting from an event as in E165a. Verb forms denoting an event may undergo semantic change to denote the state resulting from this event (e.g. perfect to resultative, cf. English they are gone, from 'they have gone'), so a change from purposive-extent 'for' to atelic-extent 'for' is also well-motivated.

Other sources of atelic-extent markers

The other sources each occur in a handful of languages, and they will be mentioned only briefly. Five languages use markers that also occur in temporal locational functions and in spatial functions. These cases are Swedish (preposition i), Latin (Ablative case), Lezgian (Inessive case), Basque (Modal or

Locative case), Tagalog (preposition nang). This expression type is not so surprising, given that time units can always be conceptualized as time spans "within" which a situation is located.

Two languages use a preposition that also means 'through, over': Polish (przez) and Lithuanian (per). Again, this is not surprising because the atelicextent function is similar to the perdurative function of §3.3.3 ('throughout'), and this function is also sometimes marked by 'through'. Two other sources of atelic-extent markers are close to the perdurative function: German lang (lit. 'for many days'

The explanation for this polysemy must evidently be sought in the extension of the posterior-durative marker to the distance-posterior sense. To be strictly compositional, the above expressions would have to contain the distance-past marker (e.g. English *since...ago, German *seit vor, French *depuis il y a, Polish *od przed), but this is omitted because there is no risk of misunderstanding. A combination such as "since three years" could not be interpreted literally with the posterior-durative sense, because the reference time ('three years') is not a location in time, but a time span. The nearest non-literal, extended sense is the distance-posterior interpretation, so this is a possible reading in several languages.

Distance-posterior is modeled on atelic extent

Another way of rendering the distance-posterior meaning is by assimilating it to the atelic-extent function. This is probably the most common expression type, and it is of course exemplified by English, where for fulfills both functions.

A few examples from other languages are cited in E175-178, where the (a)

sentence shows the atelic-extent function, and the (b) sentence shows the distance-posterior function.

E175. Swedish a. De reser i en timme. 'I've been here for six months.'

Other languages which simply use their atelic-extent marking are Latvian, Finnish, Estonian, Basque, Hausa, and Japanese. The use of atelic-extent marking for the distance-posterior function is easy to explain: Particularly in those languages that use the present perfect tense (English, Swedish (cf. E175b), Finnish, Estonian, i.e. the same set of languages that use it with posteriordurative adverbials, cf. §5.4, E77-78), the distance-posterior meaning arises automatically: In the sentence He has lived there for a year, the use of the present perfect ensures that the situation is understood as beginning in the past and continuing into the present, so that the beginning of the measured time span must be a year before the moment of speech. In languages that do not use the present perfect, but the present tense in this situation, the distance-posterior reading apparently results from a conversational implicature. Thus, E178b from Tamil is literally 'I am here for six months', and 'for six months' is interpreted as referring to the period immediately preceding and including the moment of speech. The implicature here probably arises from the fact that the present tense refers to the moment of speech, which is just a point in time and cannot easily be modified by an atelic-extent adverbial. It would be odd to say I am here for six months if the speaker arrived three months earlier and will stay for another three months. 4

In order to make explicit the present perfect sense of the present tense, i.e.

the fact that the time span measured by the extent adverbial ends in the present, several languages employ aspectual adverbs such as 'already' or 'now'.

Some examples are given in E179 (the Greek adverb tóra in 176b fulfills a similar function). The grammatical status of these adverbials is somewhat ambiguous.

On the one hand, they are more or less obligatory for co-signaling the distanceposterior sense, but on the other hand, they should probably not be considered According to SUTCLIFFE (1936), the construction "il-X Y" goes back to "ħin l-X Y", lit. 'the time to X is Y', i.e. 'X has the time of Y'. In this construction ħin l-was transformed phonologically into ħill-, ħil-, hil-and finally il-. Thus, while being quite unusual as a synchronic marker of distance-posterior, Maltese il-has a diachronic origin that fits well into the group discussed in this sub-section.

A related phenomenon is the use of a presentative particle ('behold', French voilà) which introduces the noun phrase denoting the time span. This construction is found in several of the languages of my sample, e.g. Russian (vot), Lithuanian (štai), French (voilà, voici), Hebrew (hine), and also in New

Testament Greek (idoú). The construction can either be overtly biclausal, as in E186a, E187-188, or monoclausal.

Chapter 9 Conclusions

It is now time to step back and recall the major discoveries of this book. After all the detail of the individual languages, what can we say in general about the marking of noun phrases as time adverbials in the world's languages, with particular regard to the transfer from space to time?

The metaphor from space to time

We saw in almost all of the semantic functions investigated here that the large majority of languages employ an originally spatial adposition (or case) to signal a temporal relationship. The systematic cross-linguistic study has thus confirmed earlier impressionistic statements concerning the ubiquity of conceptual transfer from space to time. There are no languages that depart from this general trend, and in this sense it is truly universal. However, in each of the individual semantic functions, there are a few languages that have a nonspatial source for their marker of the function. In this sense the space-to-time transfer is not universal, but only a strong tendency.

But is the ubiquitous close relationship between temporal and spatial markers really due to metaphor? TRAUGOTT (1978:371) explicitly denies this: "spatial expressions for time are not metaphorical ... at least not those of the sort discussed here; there are metaphors of time, such as 'going round a corner in time', but they are distinguishable from basic spatio-temporal expressions ..." However, Traugott's view is clearly based on a narrow understanding of metaphor. If we adopt a broader view of metaphor, where metaphor is defined as conceptualization of a target domain in terms of a source domain, keeping the profile constant (LAKOFF & JOHNSON 1980, CROFT 1993), there is no reason not to regard the shift from space to time as metaphorical. For instance, when we want to express the concept 'before August', we may conceptualize the temporal domain in terms of the spatial domain, transferring the spatial concept 'in front of' to another domain without changing its designation (or profile, in LANGACKER's (1987-91) terminology), thus giving expressions like German vor August (cf. vor dem Haus 'before the house').

But independently of the terminology that one wishes to employ in this area, the crucial aspect is the secondary nature of temporal markers with respect to spatial markers. Whether one describes this as metaphor, as "image-schematic transformation", or as "imaginative extension" (cf. KUTEVA & SINHA 1994), it is clear that spatial meaning is primary and temporal meaning is secondary. This fact is not explained on JACKENDOFF's theory (cited already in §1.7) that both spatial structure and temporal structure are instantiations of "an abstract organization that can be applied with suitable specialization to any field". If this were the case, then we would expect that transfer from time to space should be as common as transfer from space to time. But in fact the former is virtually unattested.

Of course, there is one type of transfer from time to space that is not at all uncommon, illustrated in E192a-b. E192. a. The road leads from Minsk to Smolensk.

b. The poplar is after the oak. This is based on the phenomenon of abstract motion (mentioned already in §5.1), the construal of spatial configurations in terms of movement (cf. LANGACKER 1991). Spatial configurations may be scanned sequentially and thereby assimilated mentally to sequences of events in time. In E192a, the observer mentally travels the road and is thus "led" by it to the destination, although there is no physical movement. In §5.1 we saw that this explains why directional adpositions may be used for temporal notions. In E192b (cf. VANDELOISE 1991, BERTHONNEAU 1993a), abstract motion explains how a temporal adposition seemingly comes to have a spatial use. But in fact, after in E192b is not really spatial, because the sentence is only possible if the observer encounters the poplar later than the oak on a mental path (for instance, if both stand beside a road that would be taken to locate the poplar). Thus, it is rather misleading to call after/before 'spatial adpositions (cf. VANDELOISE's (1991) book title). In a different terminology, one could say that E192b metonymically stands for 'The encounter with the poplar is after the encounter with the oak'. It seems that this special quasi-spatial use of 'after' and 'before' is never conventionalized and turned into a really spatial use.

So what would be a real counterexample to the claim that the transfer from space to time is unidirectional? HEINE et al. (1991:51) point out a possible case from Solomons Pijin, as described by KEESING (1991:335): Solomons Pijin uses the temporal adverb fastaem 'first' (from English first time) as a temporal preposition (much like the languages cited in §4.4), e.g. fastaem long faet 'before the fight'. This is then extended to the spatial meaning 'in front', e.g. fastaem long haos 'in front of the house'. However, this extension was modeled closely on the substrate languages, cf., e.g., Kwaio na'o-na omea 'before the mortuary feast', na'o-na 'ifi 'in front of the house'. Thus, the Solomons Pijin example is not a real counterexample at all, because it arose in a situation of intensive language contact, perhaps better described as relexification.

A better example of transfer from time to space is French depuis, which originally must have meant 'after' (cf. Spanish después), but now means 'since'.

In addition, depuis has now acquired a spatial sense, as in depuis la fenêtre 'from the window'. However, such examples are apparently extremely rare.

Types of temporal expressions

In this book my main concern has been with grammatical markers of temporal NP relations, and I found that these are largely based on spatial markers. But what about other kinds of temporal expressions?

Let us first consider nouns that directly mean 'time'. ANSTATT (1996) has studied 'time' words in a fair number of (mostly Slavic) languages, concluding that these are not as a rule based on a spatial metaphor. This contrast between temporal relation markers and time nouns can be explained on the basis of CROFT's (1993) observations on metaphor ("domain mapping") and metonymy ("domain highlighting"). CROFT points out that as a rule domain mapping (metaphor) is induced in relational expressions by autonomous expressions, as in E193a-c, where the verb, the preposition and the noun are relational and are interpreted metaphorically because they are combined with autonomous expressions from a different cognitive domain that require domain mapping for the sentence to make sense. Here the nouns and the verb are autonomous and are interpreted metonymically because they are combined with relational expressions that require domain highlighting. For instance, in E194a the noun La Repubblica (a newspaper) can only be interpreted in combination with 'arrive at the press conference' if its designation in the domain of newspaper authors (i.e. journalists) is highlighted. Since 'time' nouns are autonomous in this sense, not relational like mouth in E193c, we would not expect them to show the effects of metaphorical domain mapping.

Next, let us look at the grammatical expressions for temporal relations on verbs, i.e. tenses. Again, we find much less evidence for spatial expressions in this area than in temporal NP markers. This is probably simply due to the fact that tense markers are usually strongly grammaticalized elements that show few synchronic traces of their origins. However, tense markers typically go back to aspectual constructions (BYBEE et al. 1994), and these are very often based on space (e.g. She is going to sell her house, German Er ist am Kochen 'He is (lit. at) cooking', French Elle vient de publier un article important 'She has just (lit. comes from) published an important article'). In this way, spatial markers can find their way into the tense-aspect system of a language. Impressionistically, however, there are many more non-spatial sources for verbal and temporal categories than for adverbial time markers (e.g. expressions of volition and obligation for future tense, participial periphrases for perfect tenses, etc.). As a result, again we find few spatial-temporal metaphors in tense-aspect expressions.

A summary of the spatial sources of temporal markers

As we saw in §4.2, there are two different ways in which the time line can be mapped onto the front-back axis: the moving-time model, where earlier events are in front and later events are behind, and the moving-ego model, where earlier events are behind and later events are in front.

The moving-time models is the basis for the majority pattern of anterior and posterior markers ( §4.1-2): 'before' is modeled on 'in front', 'after' is modeled on 'behind' (this is also true for the cases of 'after' from 'trace, track', cf. §4.4).

Furthermore, the majority type of distance marker is based on anterior and posterior markers ( §6.1), and thus often indirectly on spatial markers.

The moving-ego model is the basis for the majority pattern of anteriordurative and posterior-durative markers ( §5.1), for 'arrive, reach' as an anterior-durative marker ( §5.2), and for 'back' and 'over' as distance markers ( §6.2.3-4).

Spatial location is also overwhelmingly the model for simultaneous location and one type of temporal extent. In simultaneous location, it is mostly spatial interior markers denoting inclusion in three-dimensional space that are used, but two-dimensional spatial markers such as 'on' and 'at' are also found (ch. 7).

Among the extent functions, only telic extent is largely expressed by a spatial source, 'within' (ch. 8), whereas atelic extent markers are quite rarely spatial.

Grammaticalization in temporal markers

Markers of different semantic functions are systematically correlated with different formal properties, i.e. they show different degrees of (synchronic) formal grammaticalization (see LEHMANN (1995) for the approach to grammaticalization assumed here). The most grammaticalized functions are the atelic-extent function, which is very often expressed by a grammatical case or zero, and the seven sub-types of simultanous location. These are also commonly expressed by zero or grammatical cases, and otherwise by semantic cases (in languages with rich inflectional morphology) or monosyllabic adpositions.

The next highest degree of formal grammaticalization is found in anteriordurative and posterior-durative markers. There are a number of languages that have case markers for these functions (terminative case, ablative case), and monosyllabic adpositions are still fairly widespread. In this respect the telicextent function is quite similar.

Finally, the lowest degree of formal grammaticalization is shown by anterior and posterior markers and by distance markers. These are almost never expressed by case inflection, but typically by fairly bulky, often disyllabic adpositions.

According to the principles of grammaticalization theory, the degree of formal grammaticalization should correlate with the degree of semantic grammaticalization, i.e. semantic generality. The semantic parameter is more difficult to evaluate independently, but it is probably not controversial that the simultaneous function is more general than the distance functions, so on the whole the predictions of grammaticalization theory are borne out once again. However, it is not clear to me in what sense we could say that anterior-durative is more general semantically than anterior (and analogously for posteriordurative and posterior). If anything, the a priori expectation would be that the relation is the reverse (this is also reflected in my choice of terms, which make the sequential-durative function more specific). This is an interesting theoretical issue that should be addressed by future research.

Finally, I have found some limited evidence that temporal markers tend to be more strongly grammaticalized formally than spatial ones, cf. §4.3. This observed asymmetry is not very strong, but the data bear out the general predictions of grammaticalization theory.

Universals of time in language

This study has thus documented the massive cross-linguistic regularities in the expression of NP-based time adverbials. As I remarked in the introductory chapter, there are not very many typological correlations to be observed in this area: I have found no way to predict, for instance, whether a language will model its expression of the distance-posterior function on the posteriordurative marker, on the atelic-extent marker, or base it on 'exist' or similar source constructions. This seems to be typical of semantically-based typologies like the one investigated here: Language typology seems to constrain the forms of grammar more than the semantic sources of grammatical markers.

Typology predicts, for instance, whether a language uses a case inflection, a preposition or a postposition for expressing the simultaneous function with seasons, but the language is "free" to choose an interior, adessive, or instrumental marking. Since many of the questions I asked in this study concerned the semantics of the sources of temporal markers, I was bound to find more universals than typological divisions.

This study also has a bearing on the question to what extent the conceptualization of time is universal or culture-bound. For a long time, linguists and especially anthropologists have emphasized the relativity of time concepts in different cultures (cf. ALVERSON 1994:1-7). This study shows exactly the opposite: The expression of time in one important domain of grammar, at least, is amazingly uniform across languages. Thus, my results fully agree with ALVERSON's (1994:6) thesis that "the human experience called "time" (or alternatively, "temporality", "duration"), like most of human experience in general, is built upon and arises from a panhuman Bauplan... [A]ll linguistic/cultural manifestations of temporal experience exhibit clearly the properties and effects of an underlying universal structure of embodied, enculturated mental experience".

Appendix: The data

9

Table 1 : The major semantic functions of NP-based time adverbials

1. Simultaneous location (ch. 7) (a) Hour at five o'clock (b) Day part in the morning, at night (c) Day on Tuesday, on the first day (d) Month in February, Ø next month (e) Season in the summer, Ø last fall (f) Year in 1962, Ø this year (g) Festival at Christmas, at Easter, at Passover

I. Location in time

Table 2 : The languages of the sample of fifty-three languages, by genetic grouping

monographic treatment of time expressions in Hopi. For a number of languages, dictionaries were an important source of data, especially for Hungarian, Udmurt, Indonesian, Basque, Nanay, and Chechen. the original questionnaire). Perhaps the trickiest problem with older translations is that they tend to be very literal, and one may suspect that the text not always reflects the naturally occurring language. However, this problem is probably not greater than in the case of native speakers who are asked to translate a sentence into their language. And it is mostly restricted to older translations of European languages, for which generally other sources of data are available as well. In modern translations into languages such as Indonesian and Haitian Creole, I found enough examples of very free translations to make me confident that the less free examples are not unnatural in the language.

for Tagalog, and of course

Comrie-Smith series are available (for bibliographical references, see the Appendix): Romanian, Modern Greek, Punjabi, Abkhaz, Arabic, Babungo, Nkore-Kiga, Maori, Kannada, Greenlandic, Hixkaryana, Basque, Japanese,

Table 3 : The canonical time periods

'I went at seven o'clock.' (Lit. 'I went when it was seven o'clock.') 'I'll be here sometime after Monday.' (Lit. 'I'll be here when Monday passes.') E52. Tagalog (SCHACHTER & OTANES 1972:474-76) a. Magpasyal tayo bago mag-alauna. 5 take.walk we [before VERB-one.o'clock] 'Let's take a walk before one o'clock.' (Lit. '...before it's one o'clock.') b. Ang balak ni Herodes ay iharap siya sa bayan pagka-tapos TOP plan GEN Herodes PT lead he.TOP at people [CONV-pass ng pista (Acts 12.4) GEN festival] 'Herodes intended to bring him forth to the people after passover.' (Lit. '...when passover has gone by.') E53. Kobon (DAVIES 1981:146) a. Hon nöd aui mɨd-aj-un hainö Oktoba ten ar-öp. we before here be-DUR-PAST.1PL [after October tenth go-PERF.3SG]

Table 4 : Languages with identical spatial and temporal anterior/posterior markers

Table 5 : Languages with related spatial and temporal sequential adpositions

efter 'after' < 'behind' French avant < *ab-ante, e.g. Old French avant lui 'in front of him' (GAMILLSCHEG 1957:248) après < 'behind' Italian dopo < *de-post, Latin post 'behind; after' Bulgarian predi cf. pred 'in front' Turkish önce cf. ön 'front' (-ce adverbial sufix)

Table 6 : Temporal markers are older than spatial markers

3

: "Gerade wegen der Vordringlichkeit der Ortsvorstellung können für diese expressivere Ausdrucksformen eintreten als für die entsprechende Zeitvorstellung".

GAMILLSCHEG (1957:246)

. Space is more concrete and can therefore be renewed and reinforced more easily, an insight that goes back at least to

HEINE, CLAUDI & HÜNNEMEYER 1991:157)

French devant < de + avant, Spanish delante < de + el + ante(s), Swedish fram + för(e), QUALITY (cf. also

observed: The spatial forms are in almost all cases longer, often by a whole syllable. The greater length of course results from the fact that the spatial markers have been reinforced more recently than the temporal markers (e.g.

Table 5

(i) I cannot be back until noon. ≡ (ii) I cannot be back before noon.

4 In English, before and until are equivalent in negative contexts (EKKEHARD KÖNIG, p.c.):

Traces are phenomena that are found behind moving entities, so this source again seems to point to the moving-time model. Examples from my sample are Russian posle, Croatian poslije (< Common Slavic *poslědi 'afterwards', based on *slědŭ 'trace'), Bulgarian sled (directly from *slědŭ 'trace'), Finnish jälkeen (cf. jälki 'trace, track'), Estonian pärast (Elative of pära 'residue, rest, hind part'), Hungarian után (locative case of 3rd person agreement form of út 'path', so húsvét ut-á-n is literally 'Easter path-its-on', i.e. 'after Easter'), Latvian pēc (<pēdis, Instrumental plural of pēds 'trace', i.e. 'in the traces (of)'). As these examples show, this source forms an areal cluster in eastern Europe.

A fairly common source of posterior markers are nouns meaning 'track, trace' or similar notions. Again, these have no analog among anterior markers.

The posterior function is expressed by an adposition that originally means 'close to, near' in a number of languages: German nach (cf. nahe 'close', nächster 'closest; next'), French après (< AD PRESSUM, lit. 'at close'), Modern Greek metá (in Ancient Greek also 'with'), Basque ondoan (from ondo 'ground; vicinity; consequence'). By contrast, no anterior marker is based on a simple proximity word, as far as I know.

A further interesting source is a particle meaning 'not yet'. In Indonesian, the preposition sebelum 'before' is derived from belum 'not yet', and in a parallel fashion sesudah/setelah/sehabis 'after' is derived from sudah 'already, finished', telah 'finished, already', habis 'finished'.

).

Table 7 : Posterior-durative markers from ablative markers

Table 8 : Anterior-durative markers from allative markers 1

than a spatial location. At first, this is surprising, because the spatial notion of a direction does not have a clear analog in the conceptual domain of time. But

Table 9

exemplifies beginning-to-end constructions in nine languages, contrasting them with the regular posterior-durative and anterior-durative markers.

Table 9 : Simple source and goal markers in beginning-to-end constructions

Finnish aamu-sta ilta-an vs. lähtien -asti 'from morning (ELAT) till evening (ILL)'

Table 10a : Posterior-durative markers based on 'begin(ning)'

Table 10b : Anterior-durative markers based on 'end, finish'or 'arrive'

Fino alle otto, Gianni ha lavorato; dopo, non so. 'Until eight, Gianni worked; after that, I don't know.'

E62. Italian

'I have been working on this article since last December, and in fact I worked on it even earlier. I began working on it in October.'

Ja rabotaju nad ètoj stat'jej načinaja s dekabrja prošlogo goda, i na samom dele ja nad nej rabotal i ran'še. Ja načal nad nej rabotat' v oktjabre.

E61. Russian

4. Deixis in the posterior-durative function: 'since' vs. 'from(..onward)'

E64-66, and by analogous examples from French (depuis) in E67-69 (ROHRER 1981:159-163). If seit/depuis cannot be used, the markers von..an/à partir de are used. If the reference time is in the future, colloquial German also allows ab. *Seit nächste Woche Von nächster Woche an gilt der neue Tarif. Ab nächster Woche 'From next week the new price list will be in force.' E66. *Seit letzten Mittwoch Vom letzten Mittwoch an hatte ich Schluckauf, aber am *Ab letzten Mittwoch Sonnabend war es vorbei. 'I had a hiccup from last Wednesday, but it was over on Saturday.' E67. a. Il est 10 heures. Le conférencier parle déjà depuis 8 heures. *à partir de 8 heures. 'It is 10 o'clock. The speaker has been speaking since eight.' b. Jean était au bureau depuis 8 heures. *à partir de 8 heures.

5.in

(< Proto-Germanic *sīþiz, cf. Latin sētius 'later') Originally, these forms were only used as adverbs, i.e. without a reference time. This use is still current in English (I met her last May, but I haven't seen her since). The "transitive" use, which turned since/seit/sedan into a preposition, appeared only later. A form that is perhaps similar is Nanay tawaNki 'afterwards; since'. This class of forms are clear cases of temporal markers that are not based on spatial relations. However, such expressions do not seem to be widespread in the world's languages. Furthermore, I have not found a counterpart for 'until' based on 'earlier'.

Table 11 : Different markers for posterior-durative-past and posterior-durative-future

e. posteriorpresent-perfect. Examples are given in E74-76. E74. Lezgian (HASPELMATH 1993:100; 219) (iniq h 'hither') a. Ha i jiqa-laj xüpüq h wi-jr-i čpi-n xürünwida-l that day-SUPEREL Xüpüqian-PL-ERG selves-GEN covillager-SUPERESS c'iji t'war ecig-na. 'Since the creation of the world his invisible properties have been clearly perceivable in his actions.' E78. Estonian (URMAS SUTROP, p.c.)

Table 12 : Examples of special anterior-limitative markers

Table 13 : Languages with identical anterior/distance-past and/or posterior/distance-future markers.

My analysis of the frequent formal identity of 'before'/'ago' and 'after'/'in' thus involves interpretative enrichment or pragmatic strengthening plus the implicit restriction to time spans one of whose end points coincides with the moment of speech. This latter restriction is sometimes made explicit in a way not unlike the use of 'this' in E82-83, E85 and E88b. For instance, Latin and Greek use the proximal demonstratives hic/hoũtos in their distance expressions, me.ACC this.ABL 2day.period.ABL or 3day.period expect.IMP 'Expect me in two or three days.' (lit. '...in these two or three days') (Lit. 'Expect me in this two-day or three-day period.') b. Ergo his annis quadrigentis Romae rex erat? (Cic. Rp. 1.58)

Table 15 : Atelic-extent adverbials expressed by a minimal case or zero

von..an seit letzten

ǯuer časa-la 'at two o'clock' 2. Ø čimii 'in the morning' 3. DAT (-du)/LOC subbota-du 'on Saturday'/ LOC/ACC las siti-pi/-ta 'at seven' 3. LOC/ACC lunis-pi/-ta 'on Monday' 4. LOC/ACC iniru-pi/-ta 'in January' 5. LOC/ACC tamya timpu-pi/-ta 'in the rainy season' 6. LOC 1980-pi (*-ta) 'in 1980' 7. MLT sa: sal-lum 'until today' HAU sai/har: sai gòobe 'until tomorrow'/har lìtìnîn 'till Monday' har < Arabic BAB zì': zì' tɨ⁄ vәshíshwì têe 'until five o'clock' zì' 'until' a č'awalaj (iniq h ) 'from (since) that day' iniq h 'hither' CHE düjna: stoxka düjna 'since last year' ABK -štax': a-š˚ax'à à-štax' 'since Monday' -štax' 'behind' GEO -dan: janvri-dan 'since January, from January on' 'from' ARM ABL: šabat'van-ic' 'since Saturday' 'from' HEB meʔaz: meʔaz beriʔat ha-ʕolam 'since the world's creation' ARB munðu: munðu l-xalqi l-ʕaalami 'since the world's creation' MLT minn: minn żmien Gwanni 'since John's days' minn 'from' MAO mai i..raa: mai i te Kirihimete raa 'since Christmas' mai 'hither' ESK ABL (-miit) aasa-miit 'since the summer' HOP angqw: nalöstalat angqw 'from the 4th day on' trí lá agus trí oíche 'for three days and three nights' WEL am: am ddeng munud 'for ten minutes' am 'for; about' BSQ MOD/LOC: bi urt-ean (two year-LOC) 'for two years'/ bost egun-ez eta gau-ez 'for five days and nights' ni-nen-kan 'for two years'/ni-zi-kan 'for two hours' kan 'period' KOR tongan: twu sikan tongan 'for two hours' IND Ø/selama: selama lima bulan 'for five months', satu jam 'for 1 hour' TAG nang: nang dalawang oras 'for two hours' nag 'in' MAO moo: moo te rua tau 'for two years' moo 'for' KOB Ø: ñin möhau 'for three days' ESK ABS: minutsit pingasut 'for three minutes' QUE ACC: ishkay wata-ta 'for two years' A.7. Telic extent SCR za: za tri dana 'in three days' za 'behind' BLG za: za tri dni 'in three days' LIT per: per tris dienas 'in three days' per 'over' LTV LOC: trīs gados 'in three years' HAU (à) cikin: (à) cikin mintìi 'in ten minutes' cikin 'in(side)' PER dar: dar (arze) dah daqiqe 'in ten minutes' dar 'in' SWA katika: katika siku tatu 'in three days' katika 'in' CHI zhī nèi: liǎng-ge xiǎoshí (zhī) nèi 'in two hours' nèi 'within' JAP -kan de ni-zi-kan de 'in two hours' de 'in' KOR -(m)an-ey twu sikan-(m)an-ey 'in two hours' NAN DAT: dūin ajŋani-du 'in four years' IND dalam: dalam tiga hari 'in three days' dalam 'inside' TAG sa loob ng: sa loob ng isang linggo 'in one week' sa loob ng 'inside' ESK akunnir-: nalunaaquttap akunnir-i-ni marlun-ni 'in two hours' akunnir-'between' A.8. Distance-future ITA tra/fra: tra un anno ritorno 'I'll be back in a year' fra 'between, among' SPA dentro de: dentro de ocho días 'in eight days' dentro de 'inside' ROM peste: peste o lună 'in a month's time' peste 'over' q'we wacra-laj 'in two months' SUPEREL 'across' CHE dälča: ill minot jälča 'in ten minutes' ABK INSTR (-la): y˚ә-sàat-k' rә-la 'in two hours' GEO -ši: or saat-ši 'in two hours' -ši 'in' ARM ABL (+ heto) erku taru-c' (heto) 'in two years' time' heto 'after' JAP de/go-ni: ni-zi-kan de/ni-zi-kan go-ni 'in two hours' de 'in'; = posterior KOR hwu-ey/: twu sikan hwu-ey 'in two hours'/ hwu-ey 'after' twi-ey sam nyen twi-ey 'in three years' twi-ey 'behind' NAN bipie: ǯuer ajŋani-du bipie 'in two years' bi-'be' paucis his diebus 'a few days ago'/ ante annos 14 ante 'before' HAI fè: jodi-a fè kat jou 'four days ago today' fè 'makes' RUS nazad: dva goda tomu nazad 'two years ago' nazad 'back' LZG wilik: 250 jis idalaj wilik '250 years ago' (lit. 250 years before this) = anterior CHE ħalxa: pxi šo ħalxa 'five years ago' = anterior ABK -àpx'a: y˚ә-sàat-k' r-àpx'a 'two hours ago' = anterior GEO c'in: or saatis c'in 'two hours ago' MAO noa atu raa: rua haora noa atu raa 'two hours ago' ENG for (+PERF): Bill has been in Manchester for three years. GER seit: Ich wohne hier seit Jahren 'I've lived here for years' seit 'since' SWE i (+PERF): Han har bott där i ett år 'He has lived there for a year' i 'in, for' FRE il y a/ RUS ACC/uže: (uže) odnu minutu 'for one minute' uže 'already' POL od: od wielu lat 'for many years' od 'since' SCR već: već osam godina 'for eight years' već 'already' BLG ot: ot mnogo godini 'for many years' ot 'since' 'I have been living here for two years.' GEO 'be': ori tve-a Bambergši var 'I've been in B. for 2 months' (E183c) ARM arden: arden yot' tari 'already seven years' (E179e) 'already' HEB ze: ze šanim rabot ani ʕoved eclexa ze 'that, it (is)' 'I've been working for you for many years' ARB munðu: munðu sanawaatin ʕadiidatin 'for many years' MLT il-: ilni tliet snin 'for three years I...' HAU Ø: kwaanaa uku 'for three days' SWA tangu: tangu siku nyingi 'for many days' tangu 'since' PER 'be'/az..piš: noh sâl ast ke 'for nine years' (E183b)/az noh sâl-e piš (E29d)

(-la)

wieczor-em 'in the evening' 3. w (+acc) we wtorek 'on Tuesday' 4. w (+loc) w miesiącu szóstym 'in the LOC

INSTR

A terminological grid is also found in the questionnaire of the COMRIE-SMITH grammars ("Lingua Descriptive Studies"/"Routledge Descriptive Grammars"), and my terms were in part inspired by it.

The figures fromFRANCIS & KUČERA (1982), a frequency dictionary of English, are as follows: year 1661, day 1077, week 425, month 327, hour 325, minute 242, decade 80, second 57, millennium 8.

There is probably also some relation to the system of tense-aspect markers. The particle i also occurs as a past tense marker.

In the Comrie-Smith grammars, section 2.1.1.6.2 is entitled "frequentative". The authors of the grammars give examples glossed sometimes by 'every', sometimes by English plurals.