Testing Morphological Productivity

zyxw zyxw zyxwvutsrqponmlkjihg zyxwv TESTING MORPHOLOGICAL PRODUCTIVITY Mark Aronoff Department of Linguistia State University of New York at Stony Brook Stony Brook, New York 11794 Roger Schvaneveldt Department of Psychology New Mexico State University L m Cruces, New Mexico 88001 Chapter 1 of Jespersen’s Philosophy of Grammar’ is called “Living Grammar.” Here, in a discussion of ‘ I . . . the psychological side of linguistic activity . . .” Jespersen sets out the important distinction between formulas and free expressions. A formula is a linguistic string in which everything is fiied, whereas free expressions “. . have to be created in each case anew by the speaker. . .” As examples of syntactic formulas, Jespersen given “How do you do?” and “Good morning!” Examples of free expressions are “John gave Mary the apple.” and “My uncle lent the joiner five shillings.” Modern linguistics has made Jespersen’s distinction familiar to us all. Indeed, the progress of syntax is due in great part to a realization that the central task of a grammar is accounting for a speaker’s ability to produce and understand novel utterances, the ability which Chomsky calls linguistic creativity, and which Jespersen more aptly called Living Grammar. But Jespersen said that living grammar is not confined to syntax: “The distinction between formulas and free expressions pervades all parts of grammar.” Here the modems have not heeded his word. This is especially true in morphology. It has been argued elsewhere*that no study of morphology can succeed which fails to distinguish between two distinct types of morphological phenomena: the actual words of a language and the possible words of a language (in Jespersen’s terms, f l e d formulas and free expressions). The actual words, the fixed formulas, constitute the lexicon of a language, and as such are inherently unpredictable (it has been recognized at least since Bloomfield3 that the basic criterion for a word’s being listed in a speaker’s lexicon is its unpredictability or irregularity). A theory that attempts to generate the actual words of a language with all their idiosyncracies is bound to face insurmountable difficulties. These difficulties can only be overcome by turning one’s attention away from the lexicon and concentrating instead on the possible words, the speaker’s potential for forming and understandig new words. By focusing on this potential one can arrive at a highly constrained and interesting theory of morphology in which the lexicon and its irregularities can also be accommodated, albeit in a (properly) derivative fashion. The elements of such a theory are presented in the above mentioned work. The theory is concerned with defining the notion possible word of a language. Since the set of possible words contains the set of existing words, the dictionary of existing words fulfills part of this definition. The other part, the definition of possible but non-occurring words, is accomplished by a set of Word Formation Rules (WFRs), which provide the patterns according to which new words may be formed in the language. These rules, which define the notion morphological structure, are the central concern of any theory of morphology. This approach thus distinguishes itself from others in its decision to study mor- . 106 0077-8923/78/0318-01~ $01.7512 @ 1978. NYAS zy z zyxwv zy Aronoff & Schvaneveldt: Morphological Productivity 107 zyxwvut zyx zyx phology in the same way as we now study syntax, devoting our attention to the language’s “rule-governed creativity.” Morphology differs from syntax, however, in one crucial regard with respect to this analogy: we make up and accept new words much more infrequently and reluctantly than we do new sentences. Why this should be is not altogether clear, but it does have practical consequences when we try to test morphological hypotheses. Another difference between morphology and syntax is that of productivity. Though speakers are generally reluctant to accept new words, some new words are more successful than others. In the case of a suffix like agentive -er, which is very productive, new words are constantly entering the language, almost unnoticed (programmer, synthesizer, lettuce dryer). The suffix -ous, by contrast, is much less successful. Only a few words ending in -ow have managed to enter the language since 1800, and most of these are obscure or facetious: magnetiferous, edacious, scrumptious. Finally, once productive -th has had no success since the coining of width in 1627, despite such valiant attempts as greenth, illth, and lowth. Once one accepts the approach to morphology that we have outlined, one must find methods of analysis that are compatible with it. Since we are modeling our enterprise after that of the syntactician, we should first look at how syntax is studied. The most common analytic tool of the modern syntactician is the grammaticality judgment. The investigator constructs sentences the grammaticality of which is predicted by a hypothesis. The value of the investigator’s hypothesis is determined by the extent to which its predictions agree with the judgment of native speakers. Though this method is not perfect, it has greatly expanded our syntactic horizons. On the analogy of this syntactic method, we should be able to make up words in accordance with a certain morphological hypothesis and submit them to speakers for judgment. Such a simple test, however, is blocked by speakers’ reluctance to deal with new words, even when they are well formed, and by the variation in productivity of patterns. The work cited above has depended on traditional techniques involving dictionaries and word-lists (Aronoff,’ chapters 3 and 6, contains examples of analyses within this tradition). These techniques, though, are somewhat deficient in that they deal in a very indirect fashion with those aspects of morphology which are our most central concern: creativity, and productivity. The major goal of our general project, then, is to develop other methods for dealing with morphological structure, particularly the notion “possible but nonoccurring word,” and to find more subtle ways of testing hypotheses which deal with such entities. Psychologists have long been interested in words (much more so than in sentences, traditionally) and the ways in which people use and process them. Recently, workers in cognitive psychology have developed reliable techniques for investigating the semantic, phonological, and orthographic structures of words, as well as the ways in which people process these structures. One of these te‘chniques is known as the lexical-decision task. In this task, people are required to judge whether various stimuli are instances of English words or not. Both the decision (yes or no) and the time taken to reach the decision provide data for testing hypotheses about linguistic structures and the psychological processes that represent and use these structures. Several recent experiments have used the lexical-decision task to investigate the role of lingusitic structure in recognizing words and nonwords. Rubenstein, Lewis, and Rubenstein‘ found that unpronounceable nonwords (e.g., BRAKV) are judged faster and more accurately than pronounceable nonwords (e.g., BEAN). The latter were judged faster than nonwords whose pronunciation is homophonic with an actual word (e.g., BRUDE, homophonic with BROOD). These findings suggest that the time required to classify nonwords in the lexical-decision task provides a scale of “wordness” according to phonological structure. The more zy 108 zyxwv zyxwvuts zy zyxwvu Annals New York Academy of Sciences a nonword embodies the phonological structure of words, the longer the time required for people to judge that it is a nonword. Other recent experiments have shown that phonological structure influences lexical-decision performance with actual words as well as with nonwords. 5 , 6 These investigators found that judgments about words were influenced by the orthographic and phonological similarity of the preceding word. For example, the word DIME is judged faster following TIME than it would be if it followed an unrelated word. In contrast, LEMON is more difficult to judge following DEMON than following a dissimilar word. Presumably the phonological structure is responsible for the differences in the two cases. The important point for our purposes is that linguistic structure on a phonological level can be investigated for both words and nonwords in the lexical-decision task. More directly related to our own work, a few studies have investigated morphological structure using the lexical-decision task. Mackay’ presented subjects with verbs (e.g., DECIDE) and asked them to produce a related noun (DECISION) as fast as possible by adding either -merit,-ence, or -ion. Reaction times and errors were related to morphological and morphophonological complexity. Taft and Forster’ found evidence to support the hypothesis that, in a lexicaldecision task, prefixed words are analyzed into their constituent morphemes before lexical access occurs. They found that nonwords which are the stems of prefixed words (e.g., JUVENATE) take longer to classify than nonwords which are not stems (e.g., PERTOIRE). They also found that prefixed nonwords took longer to classify when they contained a real stem (e.g., DEJWENATE) compared with control items which did not (e.g., DEPERTOIRE). Thus nonwords which are morphologically closer to real words lake longer to react to. It is of some note that despite Berko’s pioneering study9there have been very few attempts by linguists to study morphology experimentally. This situation can largely be traced to the lack of a sufficiently explicit theory of morphological structure. Within the small amount of literature that exists, it is not difficult to find work which is marred by elementary misunderstandings. For example, Steinberg’s studylo of the “reality” of the phonological system of Chomsky and Halle” is rendered useless by the fact that the test items used in it, which are supposed to be novel, morphologically complex words of English, consistently break well-attested rules of English word formation and hence are impossible. In general, Steinberg pays no attention to such crucial variables as the final morpheme of the word to which a suffix is attached. The suffix -iry, for example, seldom attaches to native or monomorphemic words, yet we find among the -iry forms in Steinberg’s study only the following, all of which are either native or monomorphemic or both: SNIDE + ity, EFFETE + ity, OVERGROWN + ity. One cannot expect a speaker of English, when faced with these bizarre “new words” to react in anything but a puzzled fashion. The mere possession of a coherent theory thus puts us on a much firmer base than was previously available. Furthermore, our ambitions are quite modest. We do not presuppose some broad and unsupported linguistic foundation and then build experimental work on that. Rather, our purpose is to use experiments to test the strength of the linguistic theory, before going on to further “psychological“ study. zyxwv zyx The Productivity Experiment A central construct of our theory is the possible but non-occurring word. Among such words, we can further establish a ranking of probability of occurrence, a ranking which is associated with the notion “productivity.” If a given Word Formation Rule (i.e., affii) is more productive than another such rule, then words formed by zyxwvu zy zyxwvut zyxwv zyxw zy zyxwvu Aronoff & Schvaneveldt: Morphological Productivity 109 the former are more likely to enter the language than those formed by the latter. Productivity and this particular way of viewing it are discussed in more detail in chapter 3 of Aronoff.’ A question that immediately comes to the mind of one who regards word formation as part of living grammar is whether productivity is a simple historical fact (some patterns are more successful than others in the long run) or whether it figures in the individual’s knowledge of the language. The latter possibility is much the more intriguing, and there is some anecdotal evidence that it may be true. If you present speakers of English with the two words obsessiveness and obsessivity, neither of which is defined in Webster’s 111, most of them will tell you that obsessiveness “sounds better.” This judgment parallels the fact that the -ness suffix is more productive than -ify with words ending in -be. Speakers thus seem to possess an ability analogous to the productivity metric. The purpose of this experiment is to verify the existence of such an ability and its consistency across individual words and across speakers. Material: Xive#ness and Xiv + ity The English nominal suffixes -ness and -ity are “rivals” in that they both often attach to the same morphological and semantic class of adjectives (CONCRUOUS/CONGRUOUSNESS/CONCRUITY; POROUS/POROUSNESS/PORO.SITY; IMMENSE/IM- The productivity of each of the suffixes varies with the morphology of the base: -ity is more productive with bases ending in ic (ELECTRIC/ELECTRICITY) and ile (SENILE/SENILITY), while -ness is more productive with om (DEVIOUS/DEVIOUSNESS) and ive (DECISIVE/DECISIVENESS). Note that though one suffix is more productive with a certain class of base, the other is not impossible: SPECIFICNESS, JUVENILENESS, UNCTUOSITY and DECEPTMTY are all attested. Productivity and its analogues can thus be studied in a very narrow range: the attachment of two rival affixes to bases of the same morphological class. The experiment deals with -ness and -ity attached to bases of the form Xive, where we know that -ness is far the more productive. This is easily demonstrated by the analytic techniques of AronofP; most obviously, there exist only 28 words of the form Xivity in Walker,I2 versus 140 of the form Xiveness. Similar studies can be done with -ness and -ity attached to other bases, as well as with other rival pairs of suffixes. MENSENES~IMMENSITY: SCARCE/SCARCENESS/SCARC~Y). Design In this experiment there are three different types of items, each consisting of an equal number of letter strings of the form Xiveness and Xivify. The three types are defined as follows: (1) words-actual words in the language (listed in Webster’s Collegiate Dictionary); (2) possible words-these items do not occur in the language but the form Xive does; (3) nonwords-these items do not occur in the language, nor does the form Xive, nor does the form X. Each subject judges 40 words, 100 possible words, and 40 nonwords. All subjects judge the same 40 words, but the possible words and the nonwords come from two different lists such that a particular Xive form will appear as Xivify in one list and as Xiveness in the other. This counter-balancing ensures that performance on the possible words and nonwords can be attributed to the ending (-ness and -ity) and not to any peculiar characteristics of the Xive items. Furthermore, each subject judges a particular Xive stem only once, precluding any effects of repeating the stems. An outline of the assignment of materials in the experiment is shown in TABLE1. TABLE 2 contains a complete list of the materials we use in the experiment. 110 zyxwv zyx zyxwvuts Annals New York Academy of Sciences TABLE1 Type of Stem Word Suffix -ness -it y Possible Word -ness -ity Nonword -ness -ity Number Per Subject 20 20 50 50 20 20 Examples from List 1 List 2 perceptiveness captivity perceptiveness captivity augmentiveness pr opulsivit y propulsiveness augmentivity depulsiveness remortivity remortiveness depulsivity The choice of a 2:5:2 ratio for the three item types was motivated by the hypothesis at issue. The subject should see some clear cases of words, some clear cases of nonwords, and many instances that could be judged either way. The selected ratio should encourage subjects to discriminate among the possible words, judging some to be words and some, nonwords. The hypothesis makes clear predictions about which possible words are more likely to be judged words. An additional variable in this experiment is the instructions given t o the subjects. One group is asked to judge whether the items are in their vocabulary. Another group judges whether the items are English words. A third group judges whether the items are meaningful words. This instructional manipulation should provide some useful information about the effect of altering the task criterion. We expect the proportions of affirmative judgments for possible words to change with instructions, but the predictions should still hold. Anticipated Results zyxwv zyxw zyxw If speakers can consistently distinguish productivity, we expect that nonexistent words of the form Xiveness will be judged to be actual words more often than nonexistent words of the form Xivity. We also expect results both within and across speakers, as well as within and across Xive stems. This may tell us something about the extent t o which productivity of Word Formation Rules is an individual or social phenomenon and the extent to which productivity holds for particular words. Procedure We asked 141 students at the State University of New York at Stony Brook to make judgments about the items listed in TABLE2. The students were divided into three groups of 47 persons each, and each group was given different instructions as described above. The suffixes -ity and -tress were counterbalanced with the stems for the possible words and the nonwords. The items were presented in six random orders on mimeographed sheets. Subject’s made yes or no judgments by circling Y or N in adjacent columns. Obviously, we were not able to collect response time with this procedure. This makes our study different from most experiments involving the Lexical Decision Task, where response time is considered the central variable. Our decision was motivated by several considerations. Firstly, the items that we are testing are different from most of those used in previous experiments of this type: We are interested in possible words, rather than simple words or nonwords. Secondly, by not zyxwvu zy zyxwv zyxwvutsrqp Aronoff & Schvaneveldt: Morphological Productivity 111 TABLE2 A. Words: 20 words of the form Xivity and 20 words of the form Xiveness proclivity negativity relativity activity objectivity decisiveness compulsiveness expansiveness offensiveness responsiveness productivity sensitivity capt ivity nativity festivity explosiveness massiveness aggressiveness expressiveness possessiveness creativity perceptivity positivity conductivity subjectivity permissiveness exclusiveness elusiveness obtrusiveness attractiveness retroactivity reactivity passivity receptivity selectivity effectiveness destructiveness primitiveness deceptiveness assertiveness B. Possible Words: the list contains 100 words of the form Xive. From each of these is formed a pair of words of the forms Xivity and Xiveness. effervescive abrasive propulsive ascensive ostensive implosive errosive asper sive contorsive recursive accessive assuasive egressive irrepressive suppressive obsessive omissive concussive tussive refusive extrusive contusive siccative indicative amplificative vellicative supplicative explicative domesticative inculcative reciprocative evocative gradative oxidative exudative permeative derogative arrogative conjugative mediative retaliative expiative initiative ablative dilative extrapolative legislative emulative stimulative granulative inflammative affirmative reformative rheumative emanative combina tive subordinative contaminative culminative illuminative agglutinative inchoative emancipative ex tirpative adumbrative reverberative enumerative vituperative asseverative elaborative invigorative pejorative impetrative administrative remonstrative eructative cantative mutative extenuative enervative relaxative tractive reflective inflective maledictive deductive structive inhibitive exploitive inceptive redemptive presumptive invertive assortive contrastive congestive insistive por tative deflective C. Nonwords: the list contains 40 nonwords of the form Xive. From each of these is formed a pair of words of the forms Xivity and Xiveness. remortive ditestive malipestive transemptive affentive mortentive amnective condictive rassive ollutive nebiative tulsive carmosive valiative incrative pulmerative argitive sebutive agrancive permulsive marbicative fulgurative ramitive lugative quentive pervictive aliomutive rubictive laspat ive prensive promutative exputitive mtusive redunsive florsive ancotive entractive hortentive plastive axiative 112 zyxwv zyx zyxwvutsr Annals New York Academy of Sciences TABLE3 Proportion of ‘Yes’ Instruction (Question) Suffix Words In your vocabulary? +it y .88 .84 .90 mess An English word? +ity mess +ity mess A meaningful word? .87 .92 .93 zy Possible Words No nword s .28 .34 .39 .46 .47 .52 .09 .10 .18 .20 .20 .19 measuring reaction time, we make the experiment much simpler to perform. It can be administered to large numbers of subjects in a very short time. Results zyxwvu zy The results are shown in TABLES 3 , 4 , and 5. As can be seen from TABLE 3, the expected results were obtained with the possible words. People responded affirmatively to the possible words with -ness suffixes more than they did with -ity suffixes, regardless of the instructions. This effect is statistically reliable. The instructions were effective in varying the proportion of positive judgments people made, with .42, .SO, and .54 of the total responses being affirmative with vocabulary, English word, and meaningful word instructions, respectively. The instructional effect shows that people can vary their criterion for what counts as a word, but more importantly for our purpose, such variations have little effect on the influence of morphological structure on their judgment. In other words, the possible words show a very similar influence of morphological structure for the different instructions. We take this to mean that the phenomenon is robust. The actual words and the nonwords showed less influence of morphological structure. This may reflect the relatively leisurely judgments people.were allowed to make in the experiment. With the words, different items are involved in the two suffix categories and since they were selected as foils, no effort was made to control for other factors. The addition of a reaction-time measure may show effects where the judgment proportions d o not. Particularly with the nonwords we used, the final TABLE4 zyxwvu zyxwv Entries Are Numbers of Items Possible Words Vocabulary Language Meaning More Yeses to iry form Equal Yeses to ity and ness forms More Yeses to ness form Sign test 33 5 62 p < .01* *Significantly different from chance (a = .05). 27 9 64 p < .01* 38 6 56 p < .05* 4 zzzzz z zyxwvutsrqponm zy z z5 0 2 fo, K Y TABLE5 Instruction More Yeses to i f y items Equal Yeses for ity and ness items More Yeses for ness items Sign test V 21 14 12 p = .25 Entries Are Numbers of Subjects Words Possible L M V L 20 13 14 p<.25 *Significantly different from chance (or = .05). 12 15 20 p = .25 12 3 32 p <.01* 11 2 34 p<.Ol* rn M 13 2 32 * p=.Ol V 12 22 13 p > .25 Nonwords L 12 10 25 * p = .05 2 sE M 19 ' E 10 .a 18 p > .25 0, 3 0 e 2a e e w 114 zyxw zyxwvuts zyxwvu zyxwvu Annals New York Academy of Sciences judgment may not be affected by the suffix, but the process of arriving at the judgment could be different for the two suffixes. In TABLES 4 and 5 the data are broken down according to individual items and individual subjects. The results of these analyses give added support to our conclu4, which is in terms of individual possible words, we sions. Looking just at TABLE find that significantly more ness forms than ity forms receive a higher proportion of Yes responses, as we predict. Furthermore, the proportion stays more or less constant, regardless of instructions. This latter result is expected. Differences in instructions should affect the total proportion of positive responses, but should not affect the ratio of positive n e s to positive ity responses. In TABLE 5 , the data are broken down for individual subjects. Here our findings are parallel. Out of 141 subjects, the number (98) giving more yes responses for ness items is significantly greater than the number (36) giving more yes responses to ity items. Seven subjects give equal numbers of responses to both sets. Here the proportions are almost exactly constant across instructions, as we predict. Summary We have shown how an experimental technique borrowed from cognitive psychology, the Lexical Decision Task, can be used to study morphology. In particular, we have shown that native speakers of English respond positively more often to novel words of the form Xiveness than they do to words of the form Xivity, a result which was predicted from the greater productivity of the former pattern, as determined by techniques of linguistic analysis. Our finding holds true under a variety of instructions. While the result of our study is not particularly surprising, we hope that in demonstrating that the Lexical Decision Task is a reliable tool in the study of morphology we will encourage further research along the lines that we have developed. zyxwvut zyxwvu zyxwvu REFERENCES 1. JESPERSEN, 0. 1924. The Philosophy of Grammar. George Allen and Unwin. London. 2. ARONOFF, M. 1976. Word Formation in Generative Grammar. Linguistic Inquiry Monograph 1. MIT Press. Cambridge, Mass. 3. BLOOMFIELD, L. 1933. Language. Holt. New York. 4. RUBENSTEIN, H., S.S. LEWIS& M.A. RUBENSTEIN. 1,971. Evidence for phonemic recoding in visual word recognition. Journal of verbal learning and verbal behavior 10: 645-57. 5. SCHVANEVELDT, R.W. & D.E. MEYER.1973. Retrieval and comparison processes in semantic memory. In Attention and Performance IV. S. Kornblum, Ed. Academic Press. New York. 6. SCHVANEVELDT, R.W., D.E. MEYER& C.A. BECKER.1976. Lexical ambiguity, semantic 7. 8. 9. 10. 11. 12. context and visual word-recognition. Journal of Experimental Psychology. Human Performance and Perception: 243-56. MACKAY,D. 1974. Derivational rules: dead souls or living people in the perception, production and storage of words? Paper presented at the Psycholinguistic Circle of New York. TAFT, M. & K. FORSTER. 1975. Lexical storage and retrieval of prefuted words. Journal of verbal learning and verbal behavior 14: 638-47. BERKO,J. 1958. The child’s learning of English morphology. Word 14: 150-77. STEINBERG, D. 1973. Phonology, reading, and Chomsky and Halle’s optimal orthography. Journal of Psycholinguistic Research 2: 239-58. CHOHSKY, N. & M. HALLE.1968. The Sound Pattern of English. Harper and Row. New York. WALKER.J. 1936. Walker’s Rhyming Dictionary, revised and enlarged by L. H. Dawson. Dutton. New York.