1 Linguistic fundamentals

This first chapter of the book offers an introduction to anaphora and the concepts
associated with it. It outlines the related phenomenon of coreference and classifies the
various types of anaphora.

This chapter does not aim to provide an all-encompassing theoretical linguistics

account of the pervasive phenomenon of anaphora. It seeks to provide the basics for
those who wish to familiarise themselves with the field of automatic resolution of
anaphora or who plan to undertake practical work in this field, with particular
reference to the types of anaphora most widely used.

1.1 Basic notions and terminology

Cohesion is a phenomenon accounting for the observation (and assumption) that

what people try to communicate in spoken or written form1 under ‘normal
circumstances’ is a coherent whole, rather than a collection of isolated or unrelated
sentences, phrases or words. Cohesion occurs where the interpretation of some
element in the discourse is dependent on that of another and involves the use of
abbreviated or alternative linguistic forms which can be recognised and understood
by the hearer or the reader, and which refer to or replace previously mentioned items
in the spoken or written text.

Consider the following extract from Jane Austen's Pride and Prejudice:

(1.1) Elizabeth looked archly, and turned away. Her resistance had not injured her
with the gentleman.2

Although it is not stated explicitly, it is normal to assume that the second sentence is
related to the first one and that her refers to Elizabeth. It is this reference which
ensures the cohesion between the two sentences. If now the text is changed by
replacing her with his in the second sentence or the whole second sentence is
replaced with This book is about anaphora, cohesion does not occur any more: the
interpretation of the second sentence in both cases no longer depends on the first
sentence.

Discourse (1.1) features an example of anaphora with the possessive pronoun her
referring to the previously mentioned noun phrase Elizabeth. Halliday and Hasan
(1976) describe anaphora3 as ‘cohesion which points back to some previous item’.4

1 Or in any other appropriate mode of communication such as gestural or more generally multimodal
communication, sign language etc.
2 Jane Austen, Pride and Prejudice, Ch.6, p.23. Penguin: London, 1995.
3 The etymology of the term anaphora goes back to Ancient Greek: anaphora (αναϕορα) is a
compound word consisting of the separate words ana (ανα) − back, upstream, back in an upward di-

4
The ‘pointing back’ word or phrase5 is called an anaphor6 and the entity to which it
refers or for which it stands is its antecedent. The process of determining the
antecedent of an anaphor is called anaphora resolution.7 When the anaphor refers to
an antecedent and when both have the same referent in the real world, they are termed
coreferential. Consider the following example from Huddleston (1984):

(1.2) The Queen is not here yet but she is expected to arrive in the next half an
hour.

In this example, the pronoun she is an anaphor, the Queen is its antecedent and she
and the Queen are coreferential. Note that the antecedent is not the noun Queen but
the noun phrase (NP) the Queen.

The relation between the anaphor and the antecedent is not to be confused with that
between the anaphor and its referent; in the above example the referent the Queen is a
person in the real world (e.g. Queen Elizabeth) whereas the antecedent the Queen is a
linguistic form. Next, consider (1.3):

(1.3) This book is about anaphora resolution. The book is designed to help
beginners in the field and its author hopes that it will be useful.

In this example there are three anaphors referring to the antecedent this book – the
noun phrase the book, the possessive pronoun its and the personal pronoun it (section
1.4 below will discuss different varieties of anaphora). For all three anaphors, the
referent in the real world is the book being read and therefore the anaphors and their
antecedent(s)8 are coreferential.

On the other hand, look at this example:

(1.4) Stephanie balked, as did Mike.9

This sentence features the verb anaphor did (see also section 1.4.4) which is a
substitution for the antecedent balked; however, since the two terms in this anaphoric

rection and phora (ϕορα) - the act of carrying. Anaphora thus denoted the act of carrying back
upstream.
4 Note that anaphora is not merely the act of referring to a previously mentioned item in a text: as will
be seen later, not every type of anaphora is referential, that is, has a referring function (e.g. verb
anaphora).
5 The ‘pointing back’ word (phrase) is also called a referring expression if it has a referential
function.
6 As a matter of accuracy, note that anaphora is a linguistic phenomenon and not the plural of
anaphor (the latter is the word/phrase pointing back), as it has been wrongly referred to as in some
work on anaphora resolution so far.
7 In the literature both terms anaphora resolution and anaphor resolution have been used. Perhaps one
can argue that anaphor resolution is a no less precise term since (i) it would be logical to say that the
anaphor is resolved to its antecedent and (ii) it is acceptable to say pronoun resolution (which would
be the ‘parallel form’ to anaphor resolution) but not pronominalisation resolution (the parallel form to
anaphora resolution). However, anaphora resolution has established itself as a more widespread term
and therefore has been adopted throughout this book.
8 In this example, both this book and the book can be regarded as antecedents of the anaphors it and its
(see also section 1.3).
9 Ian MacMillan, Light and Power Stories, Story 5 ‘Idiot’s Rebellion’, p.51. University of Missouri
Press: Columbia and London, 1980.

5
relation do not have a common referent, one cannot speak of coreference between the
two.

1.2 Coreference

The previous section introduced examples of coreference, which is the act of picking
out the same referent in the real world. As seen in (1.3), a specific anaphor and more
than one of the preceding (or following) noun phrases may be coreferential thus
forming a coreferential chain of entities which have the same referent. As a further
illustration, in (1.5) Sophia Loren, she (from the first sentence), the actress, her and
she (second sentence) are coreferential. Coreferential chains partition discourse
entities into equivalence classes. In (1.5) the following coreferential chains can be
singled out: {Sophia Loren, she, the actress, her, she}, {Bono, the U2 singer}, {a
thunderstorm}, {a plane}.10

(1.5) Sophia Loren says she will always be grateful to Bono. The actress revealed
that the U2 singer helped her calm down when she became scared by a
thunderstorm while travelling on a plane.11

Definite noun phrases in copular relation are considered as coreferential, hence in the
example

(1.6) David Beckham is the Manchester United midfielder.12

the proper name David Beckham and the definite description the Manchester United
midfielder are coreferential.13 Coreferential are also David Beckham and the second
best player in the world in (1.7)

(1.7) David Beckham was voted the second best player in the world behind
Rivaldo.14

Other examples of copular relations include the relation of apposition illustrated by

(1.8):

(1.8) Dominique Voynet, the French Environment Minister, launched a bitter attack
on Mr. Prescott’s ‘chauvinism’.15

10 The notion of coreference can be formally defined as a relation and the coreference chains can be
described as equivalence classes. In particular, if we introduce the relation t antecedes x between an
anaphor x and an antecedent t (note that this definition would apply to identity-of-reference anaphora
only), then two discourse entities x and t are said to be coreferential (notated as coref (x,t)) if any of
the following holds (Lappin and Leass 1994): (i) t antecedes x; (ii) x antecedes t; (iii) s antecedes x
for some discourse entity s and coref (s,t) and (iv) s antecedes t for some s and coref (s,x). Also, coref
(x,x) is true for any discourse entity x. The coref relation defines equivalent classes of discourse
entities: each class corresponds to a coreferential chain equiv (x) = { y | coref (x,y)}.
11 Adapted from Now, 31 October 2001.
12 Times, 16 May 2000, p.7.
13 In addition to establishing coreference between two definite noun phrases in copular relation,
another interpretation would be that the definite noun phrase after the verb to be has a predicative,
rather than a referential function. See Lyons (1977), volume 2, p. 185 for related discussion.
14 The Express, 15 April 2000, p. 119.
15 The Independent, 28 November 2000, p.1.

6
In this example the definite noun phrase the French Environment Minister is
coreferential with the NP to which it applies, in this case Dominique Voynet. Since
proper names are regarded as definite, in the example

(1.9) Bulger is a fugitive and his sister, Jean Holland, had tried to stop the Justice
Department from seizing Bulger's winnings, one-sixth of a 1991 $14.3 million
jackpot.16

the NP Jean Holland is coreferential with the NP to which it applies (his sister). On
the other hand, the indefinite predicate nominal a fugitive is not normally regarded as
coreferential17 with Bulger: the fact that it is not specific enough means that it cannot
be viewed as an NP having the same referent in the real world as Bulger.18

It is important to point out that in some cases an NP without a ‘definiteness’ modifier

(such as the, this, that) can still be regarded as specific and definite, and therefore
coreferential with the NP with which it is in a copular relation:

(1.10) Nicolas Clee, editor of the Bookseller, describes him as a journalist’s dream
contact.19

In this example editor of the Bookseller is specific enough to be regarded as definite,

and therefore coreferential with Nicolas Clee.

Coreference is typical of anaphora realised by pronouns and non-pronominal definite

noun phrases (see varieties of anaphora in 1.4), but does not apply to varieties of
anaphora that are not based on referring expressions, such as verb anaphora.
However, as was already seen with indefinite noun phrases, not every NP triggers
coreference. Bound anaphors which have as their antecedent quantifying NPs such as
every man, most computational linguistics, nobody etc. are another example where
the anaphor and the antecedent do not corefer. As an illustration, the relation in (1.11)
is only anaphoric, whereas in (1.12) it is both anaphoric and coreferential.

(1.11) Every man has his own agenda.

(1.12) John has his own agenda.

A substitution test can be used to establish coreference in (1.12) resulting in the

semantically equivalent sentence

(1.13) John has John’s own agenda.

No such equivalence can be yielded with (1.11) however, where a substitution test
produces

16 Example from (Hirschman et al. 1997).

17 My interpretation is different from that adopted in the MUC (Message Understanding Conference)
coreference task (see Chapter 7) where indefinite predicate nominals are regarded as coreferential with
the NP they apply to.
18 In fact, since the indefinite NP designates an entire class of entities, it cannot properly have a
referent (point made by Linda C. Van Guilder).
19 Telegraph Magazine, 8 April 2000, p.26.

7
(1.14) Every man has every man’s agenda.

which is not the same statement as (1.11).

Finally, in the example

(1.15) The man who gave his paycheck to his wife was wiser than the man who gave
it to his mistress.20

the anaphor it and the antecedent paycheck do not correspond to the same referent in
the real world but to one of a similar description (such type of anaphora is called
identity-of-sense anaphora as opposed to identity-of-reference anaphora in
examples 1.3 and 1.5; see also section 1.6 for more details). Therefore, it and his
paycheck are not coreferential.21

On the other hand, there may be cases where two items are coreferential without
being anaphoric. Cross-document coreference is an obvious example: two mentions
of the same person in two different documents will be coreferential, but will not stand
in anaphoric relation.

Having seen some of the differences between anaphora and coreference, it is worth
emphasising that identity-of-reference nominal anaphora22 involves coreference by
virtue of the anaphor and its antecedent having the same real-world referent.
Consequently, for anaphora of that type, it would be logical to regard each of the
preceding lexical noun phrases23 that are coreferential with the anaphor(s) as a
legitimate antecedent. In the light of this observation, the task of automatic anaphora
resolution will be considered successful, if any of the preceding non-pronominal
entities in the coreferential chain24 is identified as an antecedent. Consider again
(1.5). Here the antecedent of the anaphors she (first sentence) and the actress is the
noun phrase Sophia Loren; both Sophia Loren and the actress can be considered
antecedents for the anaphors her and she from the second sentence.

This book will focus more on the task of anaphora resolution and less on coreference
resolution.25 Whereas the task of anaphora resolution has to do with tracking down an
20 Karttunen (1969).
21 There are other examples where the anaphor does not trigger coreference such as My neighbour has
a monster Harley 1200. They are really huge but gas-efficient bikes. (Sidner, 1983). To account for
such cases, Sidner introduces the relationship co-specification. She regards the relationship anaphor-
antecedent as kind of cognitive pointing to the same ‘cognitive element’, called specification. Co-
specification allows one to construct abstract representations and define relationships between them
which can be studied in a computational framework.
22 Nominal anaphora is the type of anaphora where the anaphor is a pronoun or a non-pronominal
(lexical) definite noun phrase and the antecedent is a non-pronominal noun phrase; this class of
anaphora is most crucial to Natural Language Processing (see sections 1.4.1 and 1.4.2).
23 Lexical noun phrases are non-pronominal noun phrases such as definite noun phrases and proper
names (see section 1.4.2).
24 We can also speak about anaphoric chains as opposed to coreferential chains. In the case of
identity-of-reference nominal anaphora the anaphoric chain would be a coreferential chain as well;
however, there may be ‘pure’ anaphoric chains that are not coreferential (e.g. anaphoric chains
featuring verb anaphora, noun (one-) anaphora etc. ). Such classes of anaphora are considered in more
detail in 1.4.3 and 1.7 below.
25 Several coreference resolution approaches will be outlined in Chapter 5.

8
antecedent of an anaphor, coreference resolution seeks to identify all coreference
classes (chains). For more on coreference resolution, it is suggested that the reader
consult the Message Understanding Conference (MUC) Proceedings in which
coreference resolution is covered extensively (Hirschman and Chinchor 1997).

1.3 Discourse entities

When the antecedent is an NP, it becomes convenient to abstract away from its
syntactic realisation in order to capture certain subtleties of its semantics. The
abstraction, termed a discourse entity, allows the NP to be modelled as a set of one
or more elements and provides a natural metaphor for describing what may on the
surface seem to be grammatical number conflicts.

For example, consider (1.16):

(1.16) Lisa could almost see the stars in the black sky, how they had looked that
night.26

The discourse entity described by the noun phrase Lisa consists of one element - the
specific person in question, whereas the discourse entity represented by the noun
phrase the stars incorporates all the stars in the sky that Lisa could ‘almost see’.

Consider now (1.17):

(1.17) The teacher gave each child a crayon. They started drawing colourful pictures.

The discourse entity represented by the noun phrase each child comprises all children
in the teacher's class and is therefore referred to by a plural anaphor.

Finally, in (1.18) the antecedent of the plural anaphor they is the police, which as a
noun phrase is singular:

(1.18) Had the police taken all the statements they needed from her?27

If the discourse entity associated with the NP the police is now considered, it is easy
to explain the number ‘mismatch’: this discourse entity as a set contains more than
one element.

Therefore, the anaphor agrees with the number of the discourse entity (for more on
agreement, see Chapter 2, section 2.1.1) associated with its antecedent rather than the
number of the NP representing it.28

For the sake of simplicity, I shall often limit the treatment of the antecedent to its
classical definition as a linguistic form (e.g. surface constituent such as noun phrase)

26 Esther Freud, ‘Lessons in Inhaling’; in GRANTA 43 Best of Young British Novelists, ed. Bill
Buford, Spring 1993, p.71. Granta Publications Ltd.: London, 1993.
27 S. Paretsky, Indemnity only, p. 131. London: Penguin Books, 1982.
28 More formally, if the cardinal number of the set representing the discourse entity is greater than 1,
then the reference can be made by a plural anaphor.

9
and therefore, refrain from searching for an associated discourse entity (e.g. semantic
set). This is an approach widely adopted by a number of anaphora resolution systems
that do not have recourse to sophisticated semantic analysis. It should be borne in
mind however, that there are cases where more detailed semantic description or
processing is required for the successful resolution (see Chapter 2, section 2.1.3).

1.4 Varieties of anaphora according to the form of the anaphor

Nominal anaphora arises when a referring expression - pronoun, definite noun

phrase or proper name, has a non-pronominal noun phrase as its antecedent. This
most important and frequently occurring class of anaphora has been researched and
covered most extensively, and is the best understood in the Natural Language
Processing (NLP) literature. As a consequence, this book will be looking mainly at
the computational treatment of nominal anaphora.

1.4.1 Pronominal anaphora

The most widespread type of anaphora is that of pronominal anaphora. Pronominal

anaphora occurs at the level of personal pronouns (The most difficult for Dalí was to
tell her, between two [sic] of nervous laughter, that he loved her.29), possessive
pronouns (But the best things about Dalí are his roots and his antennae.30), reflexive
pronouns (Dalí once again locked himself in his studio …31) and demonstrative
pronouns (Dalí, however, used photographic precision to transcribe the images of his
dreams. This would become one of the constraints of his work…32). Relative
pronouns are regarded as anaphoric too (Dalí, a Catalan who was addicted to fame
and gold, painted a lot and talked a lot.33).

The set of anaphoric pronouns consists of all 3rd person personal (he, him, she, her,
it, they, them), possessive (his, her, hers, its, their, theirs) and reflexive (himself,
herself, itself, themselves) pronouns plus the demonstrative (this, that, these, those)
and relative (who, whom, which, whose) pronouns both singular and plural (where
and when are anaphoric too, see section 1.4.4 for locative and temporal anaphora).
Pronouns 1st and 2nd person singular and plural are usually used in a deictic
manner34 (I would like you to show me the way to San Marino.) although their
anaphoric function is not uncommon in reported speech or dialogues as the use of I in
(1.19) and (1.25), and the use of you in (1.20)

(1.19) ‘He is beautiful,’ Isabel told the woman, of her own son. ‘I feel incomplete
when I am not with him.’35

29 Gilles Neret, Dalí, Ch.2, p. 23. Benedikt Taschen Verlag GmbH: Germany, 1994.
30 Gilles Neret, Dalí, Ch.1, p. 8. Benedikt Taschen Verlag GmbH: Germany, 1994.
31 Gilles Neret, Dalí, Ch.2, p. 26. Benedikt Taschen Verlag GmbH: Germany,1994.
32 Gilles Neret, Dalí, Ch.2, p. 23. Benedikt Taschen Verlag GmbH: Germany, 1994.
33 Gilles Neret, Dalí, Ch.1, p.6. Benedikt Taschen Verlag GmbH: Germany, 1994.
34 Deictic words are those whose interpretation is derived from specific features of the context
surrounding an utterance (e.g. who is the speaker, who is the addressee, where and when the utterance
takes place) and not from previously introduced words, as is the case with anaphors. For a brief outline
of deixis see section 1.11.
35 John Updike, Brazil, p. 34. Penguin Books: London, 1994.

10
(1.20) James, don’t cross-examine me. You sound like a prosecuting counsel.36

(i) Pleonastic it

In addition to the 1st and 2nd person pronouns, the pronoun it can often be non-
anaphoric. For example, in (1.21) it is not specific enough to be considered
anaphoric:

(1.21) It is dangerous to be beautiful – that is how women have learned shame. 37

Non-anaphoric uses of it are also referred to as pleonastic38 (Lappin and Leass 1994)
or prop it (Quirk et al. 1985). Examples of pleonastic it include non-referential
instances of

a) It appearing in constructions with modal adjectives such as It is dangerous, It is

important, It is necessary, It is sufficient, It is obvious, It is useful etc.
b) It in various constructions with cognitive verbs such as It is believed that..., It
appears that ..., It should be pointed out that... etc.
c) It appearing in constructions describing weather conditions such as It is raining, It
is sunny, It is drizzling etc.
d) It in temporal constructions such as It is five o’clock, It is high time (we set off), It
is late, It is tea time, It is winter, What day is it today? etc.
e) It in constructions related to distance such as How far is it to Wolverhampton?, It
is a long way from here to Tokyo.
f) It in idiomatic constructions such as At least we've made it, Stick it out, Call it
quits, How's it going?39
g) It in cleft constructions such as It was Mr.Edgar who recruited Prudence Adair.40

Non-anaphoric uses of it are not always a clear cut case and some occurrences of it
appear to be less unspecified than others and are therefore a matter of debate in
linguistics. For further discussion of this issue see Morgan (1968).

The automatic identification of pleonastic it in English is not a trivial task. For further
discussion see section 2.2.1.

(ii) Other non-anaphoric uses of pronouns

In addition to pleonastic it, there are other non-anaphoric uses of 3rd person pronouns
in English. The generic use of pronouns is frequently observed in proverbs or
sayings:

(1.22) He that plants thorns must never expect to gather roses.

(1.23) He who dares wins.

36 P.D. James, Original Sin, Ch. 8, p.6. Faber and Faber: London, 1995.
37 John Updike, Brazil, p. 7. Penguin Books: London, 1994.
38 Semantically empty.
39 Quirk et al. 1985.
40 Susan Sallis, Come Rain or Shine, Ch.1, p.9. Transworld Publishers: London, 1988.

11
The deictic use (see footnote 31; see also section 1.11) of 3rd person pronouns is not
uncommon in conversation. For example, some time ago I went shopping with my
son, then 3 years old. Upon reaching the till he explained to me that we had spent a
lot of money so that we now had less money than we had started the shopping trip
with. The cash assistant must have overheard his comments and I was chuffed when
she said:

(1.24) He seems remarkably bright for a child of his age.

In this case he was not used anaphorically but deictically; in fact there had been no
mention of the little boy prior to the utterance.

1.4.2 Lexical noun phrase anaphora

Lexical noun phrase anaphors are realised syntactically as definite noun phrases,
also called definite descriptions (Russell 1905), and proper names. Although
personal, reflexive, possessive and demonstrative pronouns41 as well as definite
descriptions and proper names are all considered definite expressions, only lexical
noun phrases and not pronouns, have a meaning independent of their antecedent.
Furthermore, definite descriptions do more than just refer. They convey some
additional information, as in (1.25), where the reader can learn more about Roy Keane
through the definite description Alex Ferguson’s No.1 player.

(1.25) Roy Keane has warned Manchester United he may snub their pay deal.
United's skipper is even hinting that unless the future Old Trafford Package
meets his demands, he could quit the club in June 2000. Irishman Keane, 27,
still has 17 months to run on his current £23,000-a-week contract and wants to
commit himself to United for life. Alex Ferguson's No. 1 player confirmed: ‘If
it's not the contract I want, I won't sign’.42

In this text, Roy Keane has been referred to by anaphoric pronouns (he, his, himself,
I), but also by definite descriptions (United's skipper, Alex Ferguson's No. 1 player)
and a proper name (Irishman Keane).43 Furthermore, Manchester United is referred to
by the definite description the club and by the proper name United.

The additional information conveyed by definite referring expressions frequently

stands in predictable semantic relation to the antecedent, and thus increases the
cohesiveness of the text. Lexical noun phrase anaphors may have the same head as
their antecedents (these footprints and the footprints, see example 1.27) or the
relationship between the referring expression and its antecedent may be that of

41 As opposed to indefinite pronouns such as some, every, any etc.

42 The Sun, 12 January 1999.
43 A number of authors restrict lexical noun phrase anaphora to references which have the same head
as their antecedents, whereas references which have different heads are regarded as form of
substitution (Halliday and Hasan 1976). Others (Coulson 1995; Grishman 1986) regard substitution
with coreferential noun phrases (see the above example) as lexical noun phrase anaphora and we are
taking this line too. In fact substitution includes, among other things, the phenomenon identity-of-
sense anaphora (see section 1.7) and anaphora realised by non-referring expressions (such as in the
case of verb anaphora). For a detailed description of substitution and the distinction between
coreference and substitution see Quirk et al. (1985).

12
synonymy (shop… the store), generalisation/hypernymy (boutique… the shop, also
Manchester United … the club as in 1.25) or specialisation/hyponymy (shop … the
boutique, also their pay deal … his current £23,000-a-week contract as in 1.25).44
Proper names45 often refer to antecedents whose names they match in whole or in
part (Manchester United…United) with exact repetition not being uncommon:

(1.26) Alice was as nervous as a kitten on the eve of Mile’s party. That’s Alice
for you.46

Certain determiners such as the, this, these, that and those signal that the noun phrase
they modify is coreferential to a previous noun phrase.

(1.27) Both noses went down to the footprints in the snow. These footprints were
very fresh.47

We have already seen that coreferential noun phrases may have identical heads, but
also that noun phrases may be coreferential even if their heads are not identical. On
the other hand, identity of heads does not necessarily imply coreference of two noun
phrases. For example:

(1.28) The rooms on the first floor and ground floor did not reveal anything odd.48

In this example, the first floor is not coreferential with ground floor. Similarly, in
(1.25) his current £23,000-a-week contract and the contract I want are not
coreferential.

Finally, definite descriptions are not always anaphoric and their generic use is not
uncommon:

(1.29) No one knows precisely when the wheel was invented.

(1.30) George enjoys playing the piano.

1.4.3 Noun anaphora

Noun phrase anaphora should not be confused with noun anaphora - the anaphoric
relation between a non-lexical proform and the head noun or nominal group49 of a
noun phrase. Noun anaphora represents a particular case of identity-of-sense
anaphora (see example 1.15 above).

44 It should be noted that these are only the basic relationships between the anaphoric definite NP and
the antecedent but not all.
45 It should be noted however, that the distinction between proper names and definite descriptions can
often be blurred. Whereas Roy Keane (1.25) is a ‘pure’ proper name, the same cannot be said for
Irishman Keane or for the noun phase the great adventurer John Smith.
46 Sarah Jackson, Staying Alive, Ch. 8, p.8. London: Chamelon Books, 1996.
47 Jack London, White Fang, Ch.1, p.36. Parragon Book Service Ltd: London, 1994.
48 Enid Blyton, The Famous Five and the Stately Homes Gang, Ch. 19, p.140. Knight Books: London,
1985.
49 N-bar in the X-bar notation, see Jackendoff (1977).

13
(1.31) I don’t think I’ll have a sweet pretzel, just a plain one.50

The non-lexical proform one constitutes an example of a noun anaphor. Note that one
points to the noun pretzel and not to the noun phrase a sweet pretzel.

1.4.4 Verb anaphora, adverb anaphora

Among the other varieties of anaphora according to the form of the anaphor, verb
anaphora should be mentioned. In the sentence:

(1.32) When Manchester United swooped to lure Ron Atkinson away from the
Albion, it was inevitable that his midfield prodigy would follow, and in 1981
he did.51

the interpretation of did is determined by its anaphoric relation52 to its antecedent in

the preceding clause. Whereas in (1.32) the anaphor did stands for the verb followed,
the verb anaphor did in (1.33) replaces the verb phrase begged for reinforcements:

(1.33) Romeo Dallaire, the Canadian general in charge, begged for reinforcements;
so did Boutros-Ghali.53

We also distinguish adverb anaphora which can be locative such as there (1.34) or
temporal anaphora such as then (1.35).

(1.34) Will you walk with me to the garden? I’ve got to go down there and Bugs has
to go to the longhouse.54

(1.35) For centuries archaeologists have argued over descriptions of how

Archimedes used concentrated solar energy to destroy the Roman fleet in
212BC. Historians have said nobody then knew enough about optics and
mirrors.55

As previously illustrated with first and second person pronouns, adverbs of this type
are frequently used not anaphorically but deictically, taking their meaning from
contextual elements such as the time or location of utterance.

It has already been shown that the anaphors can be verbs and adverbs, as well as
nouns and noun phrases56, and thus span the major part-of-speech categories.

1.4.5 Zero anaphora

50 Paulina Simons, Eleven Hours, p.5. Flamingo: Great Britain, 1999.

51 Hotline, Autumn 1999, p.9.
52 Note that while did can be regarded as substitution, it does not have a referring function.
53 The Sunday Times, 14 May 2000, p. 20.
54 Alex Garland, The Beach, Prisoners of the Sun, p. 213. Penguin: 1997.
55 The Sunday Times, 14 May 2000, p.8.
56 Note that pronouns belong to the syntactical category NP.

14
Another important class of anaphora according to the form of the anaphor is the so-
called zero anaphora or ellipsis. Zero anaphors (signalled below by ∅) are
‘invisible' anaphors - at first glance they do not appear to be there because they are
not overtly represented by a word or phrase. Since one of the properties and
advantages of anaphora is its ability to reduce the amount of information to be
presented via abbreviated linguistic forms, ellipsis may be the most sophisticated
variety of anaphora.57

Ellipsis is the phenomenon associated with the deletion of linguistic forms thus
enhancing rather than damaging the coherence of a sentence or a discourse segment.
The resultant ‘gap’ (zero anaphor) signals the necessity of recovering the meaning via
its antecedent.

The most common forms of ellipsis are zero pronominal anaphora, zero noun
anaphora and verb (phrase) ellipsis.

(i) Zero pronominal anaphora

Zero pronominal anaphora occurs when the anaphoric pronoun is omitted but is
nevertheless understood. This phenomenon occurs in English a somewhat restricted
environment in but is so pervasive in other languages such as Spanish, Italian,
Portuguese, Polish, Chinese, Japanese, Korean and Thai, that NLP applications
covering these languages cannot circumvent the problem of zero anaphora
resolution.

Consider the first sentence in the previous paragraph.

(1.36) Zero pronominal anaphora occurs when the anaphoric pronoun is

omitted but ∅ is nevertheless understood.

The third clause in this sentence features zero pronominal anaphora (The expected
full form would have been but it is nevertheless understood).58

Similarly the second clause of the sentence

(1.37) Willie paled and ∅ pulled the sock up quickly.59

contains a zero pronominal anaphor.

In some languages verb agreement points to a zero pronoun. As an illustration,

consider the following example in Spanish:

57 This is the view expressed by Coulson (1995).

58 Note that pronominal zero anaphora overlaps with ‘zero noun phrase’ anaphora. Since zero
pronominal anaphora is realised by a missing pronominal constituent and since pronouns replace noun
phrases, one could argue that the missing pronoun could well have been a missing noun phrase. As an
illustration, the second clause of example (1.36) can be reconstructed as to it is nevertheless
understood but also as the pronoun is nevertheless understood and even as this pronoun is neverthe-
less understood. To describe cases such as (1.36) - (1.39), the terms zero pronominal anaphora or
zero pronoun have been adopted extensively in the literature due probably to the fact that the pronoun
would have been the most natural overt expression.
59 M. Magorian, Goodnight Mister Tom, p.13. Penguin: London, 1981.

15
(1.38) Marta está muy cansada. ∅ Ha estado trabajando todo el día.
Marta is very tired. (She) Has been working all day long.

Japanese, Chinese and Korean are languages with extensive use of zero pronouns.60
The following is an example of zero pronominal anaphora in Japanese.

(1.39) Nihongo o hanasu no wa kantan desu ga kaku no wa muzukashii desu.

Speaking Japanese is easy but writing ∅ (=it) is difficult.

A study of anaphoric pronouns in parallel English and Japanese texts conducted by

Uehara (1996) exemplifies the pervasive distribution of zero pronouns in Japanese.
This study found61 that 14.5% of the English anaphoric pronouns were retained in
Japanese as overt pronouns, 29% were replaced by overt noun phrases and 56.5%
were ‘deleted’ as zero pronouns.62

(ii) Zero noun anaphora

Zero noun anaphora arises when the head noun only - and not the whole NP - is
elliptically omitted (the reference is realised by the ‘non-omitted’, overt modifiers).
Typical overt modifiers of zero anaphoric nouns in English are the indefinites
several, few, some, many, more.

(1.40) George was bought a huge box of chocolates but few ∅ were left by the end
of the day.
(1.41) Jenny ordered three copies of the document and Conny ordered several ∅ too.

In (1.40) and (1.41) the empty set sign ∅ stands for the elliptically omitted chocolates
and copies respectively.

(iii) Zero verb anaphora

Zero verb anaphora occurs when the verb is omitted elliptically and the zero
anaphor points to a verb in a previous clause or sentence:

(1.42) Win a Golf GTi or ∅ a week in Florida or ∅ weekend in Paris.63

The zero verb anaphors, ∅, stand for the verb win in the clause Win a Golf GTi.

60 Many linguists (Foley and van Valin 1984; Hinds 1978; Tsujimura 1996) highlight the difference
between zero anaphora in Japanese which is controlled by inference (pragmatically controlled zero
anaphora) and zero anaphora in Latin and Slavonic languages which is controlled by agreement.
Nariyama (2000), however, argues that zero anaphora in Japanese is not controlled so much by
inference but more importantly by the interaction of a number of different grammatical factors such as
morphological agreement, syntax constraints and discourse topic.
61 The study was based on O’Henry’s story ‘The Last Leaf’. Note that in this case the original English
text was translated into Japanese.
62 One has to bear in mind that the Japanese texts were translations from English. In non-translated
Japanese texts the frequency of overt pronouns is typically much lower (personal communication, S.
Nariyama).
63 Daily Mail, 4 August 1999, page 20.

16
(iv) Verb phrase zero anaphora (ellipsis)

Verb phrase zero anaphora, also termed ellipsis, is the omission of a verb phrase
which leaves a gap pointing to a verb phrase antecedent, usually in a previous clause,
and which enhances the readability and coherence of the text by avoiding repetition.

(1.43) I have never been to Miami but my father has ∅, and he says it was wonderful.

In this example ∅ stands for the verb phrase been to Miami.

Finally, it is interesting to note that the antecedent can be elliptically omitted too as in
(1.44):

(1.44) I have not got a car myself but Tom has ∅, and I think I'll be able to persuade
him to let us borrow it.64

1.5 Types of anaphora according to the locations of the anaphor and the
antecedent

The varieties of anaphora discussed so far are based on the different types of words
which refer back to (or replace) a previously mentioned item. Depending on the
location of the antecedent, intrasentential (sentence) anaphora and intersentential
(discourse) anaphora can be observed.

Intrasentential anaphora arises if the anaphor and its antecedent are located in the
same sentence. On the other hand, intersentential anaphora is exhibited when the
antecedent is in a different sentence from the anaphor. Reflexive pronouns are typical
examples of intrasentential anaphors. Possessive pronouns can often be used as
intrasentential anaphors too, and can even be located in the same clause as the
anaphor. In contrast, personal pronouns and noun phrases acting as intrasentential
anaphors usually have their antecedents located in the preceding clause(s) of the same
complex sentence.

(1.45) Pop superstar Robbie Williams hid his secret heartbreak as he picked up three
Brit awards last night. He was stunned to discover that his ex-fiancée, All
Saints beauty Nicole Appleton, is dating a New York rapper. Robbie, 25, was
distraught after being dumped by the love of his life Nicole at Christmas.65

In the first sentence of (1.45) the anaphoric pronouns his and he are examples of
intrasentential anaphors having their antecedent in the same sentence (the antecedent
of he is in a preceding clause but still in the same sentence). On the other hand, he
and his in the second sentence, and Robbie in the third sentence, act as intersentential
anaphors since their antecedent is in a preceding sentence.

64 The following would be an alternative interpretation: ∅ also acts as a zero anaphor with antecedent
car and since it is coreferential with the anaphor it, then car is regarded as the antecedent of it. Note
that in this would be a case of identity-of-sense anaphora.
65 The Mirror, 17 February 1999.

17
The distinction between intrasentential and intersentential anaphora is of practical
importance for the design of an anaphora resolution algorithm. As pointed out in
5.3.1, 5.4 and 7.4.2, syntax constraints could play a key role in the resolution of
intrasentential anaphors.

1.6. Indirect anaphora

Indirect anaphora66 arises when a reference becomes part of the hearer’s or reader’s
knowledge indirectly rather than by direct mention, as in (1.46):

(1.46) Although the store had only just opened, the food hall was busy and there
were long queues at the tills.67

In (1.46) the noun phrase the store is regarded as antecedent of the indirect anaphors
the food hall and the tills. It can be inferred that the tills make an indirect reference to
the store because it is known that stores have tills and because the store has already
been mentioned. Similarly, the food hall is understood to be part of the store. The
inference may require more specialised ‘domain’ knowledge, however, and in the
example:

(1.47) When Take That broke up, the critics gave Robbie Williams no chance of
success.68

one must know that Robbie Williams was a member of the former pop group Take
That in order to be able to infer the indirect reference.69

The above examples feature relationships such as part-of (1.46) and set membership
(1.47) between the anaphor and its antecedent.70 The latter includes the relationship
subset-set between the anaphor and its antecedent as in (1.53) which are also
instances of indirect anaphora. The distinction between direct and indirect anaphora is
not clear-cut. Many definite descriptions can serve as examples of indirect anaphora
and the amount of knowledge required to establish the antecedent may vary
depending on whether the relation between the anaphor and the antecedent is that of
generalisation, specialisation or even synonymy.71 In example (1.25), for instance,
some of the coreferential links can be established only on the basis of the knowledge
that Roy Keane is Irish or that he is Manchester United's skipper. Hence some
researchers (Vieira and Poesio 2000b) use the term direct anaphora to refer
exclusively to the cases when the definite description and the antecedent have
identical heads.

1.7. Identity-of-sense anaphora and identity-of-reference anaphora

66 This class of anaphora is also known as bridging or associative anaphora.

67 Bill Bryson, Notes from a Small Island, Ch.10, p.135. BCA: England, 1995.
68 The Mirror, 17 February 1999.
69 Or alternatively, to know that musical bands are things that break up, have critics, have members
who may or may not achieve success etc.
70 Or more precisely between the discourse entities associated with the anaphor and the antecedent.
71 As mentioned earlier, this is not an exhaustive list of the possible relationships between a definite
description and its antecedent.

18
In all preceding examples of pronominal and lexical noun phrase anaphora (except
examples 1.15 and 1.31) the anaphor and the antecedent have the same referent in the
real world and are therefore coreferential. These examples demonstrate identity-of-
reference anaphora, with the anaphor and the antecedent denoting the same entity.
For example:

(1.48) In Barcombe, East Sussex, a family had to flee their cottage when it was
hit by lightning.72

The anaphor it and their cottage have the same referent: the cottage that belonged to
the family and that was hit by a lightning. In (1.15), however:

(1.15) The man who gave his paycheck to his wife was wiser than the man who
gave it to his mistress.

paycheck and it do not refer to the same entity but to one of a similar description. In
particular, it refers to the paycheck of the second (less wise) man. Similarly, in (1.49)

(1.49) The physicians who had eaten strawberries were much happier than the
physicians who had eaten egg sandwiches for lunch.73

the two mentions of the physicians are not coreferential.

This type of anaphora is called identity-of-sense anaphora. An identity-of-sense

anaphor does not denote the same entity as its antecedent, but one of a similar
description. Clearly identity-of-sense anaphora does not, by definition, trigger
coreference because the anaphor and the antecedent do not have the same referent.

A further example of identity-of-sense anaphora is the sentence:

(1.50) The man who has his hair cut at the barber's is more sensible than the one
who has it done at the hairdresser's. 74

Note the identity-of-sense anaphors it and the one. The latter refers to an item of
similar description (man) that is different from the man who has his hair cut at the
barber's .

The following sentences supply yet more examples of identity-of-sense anaphora:

(1.51) George picked a plum from the tree. Vicky picked one too.
(1.52) Jenny ordered five books. Olivia ordered several too.

In (1.51) and (1.52) the anaphors one and several75 refer to entities of a different
description from their antecedents (Vicky picked a different plum from George; the
books ordered by Olivia are different from those ordered by Jenny).

72 Daily Mail, 9th October 2001.

73 Jerome K Jerome, Three Men in a Boat, Ch. 1, p. 8. Penguin: London, 1994.
74 Note the verb anaphor done.
75 Note that one and several act as noun anaphors; note also the zero noun anaphor after several
(apples elliptically omitted) in 1.53.

19
Note, on the other hand, that several in

(1.53) Jenny bought 10 apples. Several were rotten.

is still an example of an identity-of-reference anaphor. In addition, (1.53) can be

regarded as an instance of indirect anaphora since the discourse entity associated with
the anaphor (several apples) is a subset of the discourse entity associated with the
antecedent (10 apples).

Finally, it is worth mentioning that it is possible to come across anaphors that can be
read either as identity-of-reference or as identity-of-sense anaphora, thus rendering
the text ambiguous:

(1.54) John likes his hair short but Jenny likes it long.76

It can be either John's hair (identity-of-reference anaphora) or Jenny's hair (identity-

of-sense anaphora).

1.8 Types of antecedents

This book, like most NLP projects, concentrates on anaphors whose antecedents are
noun phrases. As already seen, however, even though these are the most common
and best studied types of anaphors, they are not the only ones. An anaphor can
replace/refer to a noun (example 1.31), verb (1.32) and verb phrase (1.33). Also, the
antecedent of a demonstrative pronoun77 or the antecedent of the personal pronoun it
can be a noun phrase, clause (1.55), sentence (1.56), or sequence of sentences (1.57).

(1.55) Owen tried to help her with something: this made indeed for disorder.78
(1.56) They will probably win the match. That will please my mother.79
(1.57) Many years ago their wives quarrelled over some trivial matter, long
forgotten. But one word led to another, and the quarrel developed into a
permanent rupture between them. That’s why the two men never visit
each other’s houses.80

In some cases, anaphors may have co-ordinated antecedents - two or more noun
phrases co-ordinated by and or other conjunctions.81 The anaphor in this case must be
plural, even if each of the noun phrases is singular.

(1.58) The cliff rose high above Paul and Clara on their right hand. They stood
against the tree in the watery silence.82

76 Adapted from Hirst (1981).

77 See also section 1.4.1.
78 Henry James, The Spoils of Poynton, p. 139. Penguin: London, 1987.
79 Quirk et al. 1985.
80 Quirk et al. 1985.
81 Such antecedents are also referred to as split antecedents in the literature.
82 D.H.Lawrence, Sons and Lovers, p. 377. Penguin: London, 1973.

20
Similarly, a co-ordinated antecedent can arise when a list of noun phrases is separated
by commas and/or a conjunction.

(1.59) Among the newspaper critics present, at that time unknown to each other
and to James, were three men shortly destined to become the most
celebrated writers of the age – George Bernard Shaw, Arnold Bennett and
H.G.Wells. They appreciated James’s intelligent dialogue…. 83

1.9 Location of the antecedent

Information about the expected/possible distance between the anaphor and the closest
antecedent84 is not only interesting from the point of view of theoretical linguistics,
but can be very important practically and computationally in that it can narrow down
the search scope of candidates for antecedents.85 Empirical evidence suggests that the
distance between a pronominal anaphor and its antecedent in most cases does not
exceed 2-3 sentences. Hobbs (1978) found that the 98% of the pronoun antecedents
were in the same sentence as the pronoun or in the previous one. Pérez (1994) studied
the SUSANNE manually tagged corpus86 and reported that out of 269 personal
pronouns, 83 had their antecedents in the same sentence, whereas 126 referred to an
entity in the preceding sentence. Moreover, 16 pronouns had their antecedents two
sentences back, whereas 44 pronouns had their antecedent three sentences back. A
study based on 4 681 anaphors from the UCREL Anaphoric Treebank corpus
conducted by McEnery, Tanaka and Botley (1997) established that in 85.64% cases
the antecedent was within a window of 3 sentences (current, previous and prior to the
previous), whereas 94.91% of the antecedents were no further than 5 sentences away
from the anaphor. Fraurud’s (1988) study of novels, reports of court procedures and
articles about technological innovations in Swedish found that in about 90% of the
cases the antecedent was located in the same sentence as the anaphor or in the
preceding one. Guindon (1988) obtained similar results for spoken dialogues as did
Dahlbäck’s (1992) findings for Swedish.

Both Fraurud and Guindon note that there is a small class of long-distance anaphors
whose antecedents are not in the same or the preceding sentence. The greatest
distance between a pronominal anaphor and its antecedent reported in (Hobbs 1978)
is 13 sentences and in (Fraurud 1988) - 15 sentences. Fraurud’s investigation also
established that the animacy of the antecedent is a factor for long-distance
pronominalisation: usually pronouns referring to humans can have their antecedents
further away. This tendency was especially evident in the stories and it looks as if
long distance anaphors are more typical of certain genres. Biber et al. (1998)
concluded that in news reportage and academic prose the distance between anaphors
and their antecedents is greater than in conversation and public speeches.87 Hitzel and
Poesio (1998) analysed a small corpus of oral descriptions of museum items and
found that the long-distance pronouns comprise about 8.4% in this kind of data.
However, for more conclusive results further analysis involving larger and more

83 Introduction to Henry James The Spoils of Poynton by David Lodge, p. 1. Penguin: London, 1987.
84 I use the term ‘closest antecedent’ because as I explained in 1.3 each preceding coreferential non-
pronominal entity is regarded as a possible antecedent.
85 See also section 2.2.2.
86 It consists of 130 000 words and is a subcorpus of Brown's Corpus of American English.
87 Biber measures the distance as the number of intervening NPs between anaphor and antecedent.

21
representative samples is needed. Hitzeman and Poesio’s analysis looked at 83
pronouns only; Fraurud’s findings were based on a sample consisting of 600
pronouns, and so cannot be regarded as definitive either.

Ariel (1990) conducted a corpus-based analysis and concluded that demonstrative

anaphors88 were normally longer-distance anaphors than pronouns, but the distance
between definite descriptions or proper names and their antecedents may be even
greater. In fact the present writer found it quite common for proper names to refer to
antecedents which are thirty or more sentences away. For example, in one newspaper
article89 President Ronald Reagan’s national security adviser Robert McFarlane was
referred to by the proper name McFarlane thirty five sentences (many of which were
long and with complicated syntax) and fourteen paragraphs after it was last
mentioned.

For practical reasons most pronoun resolution systems restrict their search to the
preceding 2-3 sentences when looking for an antecedent (see Kameyama 1997;
Mitkov 1998b). On the other hand, since anaphoric definite noun phrases may have
their antecedents further away, strategies for their resolution have involved the search
of the 10 preceding sentences (Kameyama 1997).

1.10 Anaphora and cataphora

Cataphora arises when a reference is made to an entity mentioned subsequently in

the text.

(1.60) She is now as famous as her ex-boyfriend. From the deserts of Kazakhstan to
the south seas of Tonga, everyone knows Monica Lewinsky.90

In this example she refers to Monica Lewinsky, mentioned subsequently. Cataphora is

similar to anaphora, the difference being the direction of the pointing (reference).

Where cataphora occurs, anaphoric reference is also possible and can be obtained by
reversing the positions of the anaphor and the antecedent.91 The new sentence is
synonymous to the original one.92

(1.61) Monica Lewinsky is now as famous as her ex-boyfriend. From the deserts of
Kazakhstan to the south seas of Tonga, everyone knows her.

Example (1.60) illustrates intersentential cataphora, but in English intrasentential

cataphora is more usual.

(1.62) The elevator opened for him on the 14th floor, and Alec stepped out quickly.93

88 A thorough study of the distance between demonstrative anaphors and their antecedents is presented
in Botley (1999).
89 ‘Captured warlord’s cry for help fell on deaf US ears’, The Sunday Times, 28 October 2001.
90 Adapted from The Mirror, 4 March 1999.
91 This does not necessarily apply to possessive pronouns: for example, reversing the positions of the
anaphora and the antecedent in (1.63) would not produce a synonymous sentence.
92 However, the rhetorical effect is different.
93 John Burnham Schwartz, Bicycle Days, p.13. Mandarin Paperbacks : London, 1989.

22
Typically, intrasentential cataphora occurs where the cataphoric pronoun is in a
subordinate clause.94

(1.63) Lifting his feet high out of the sand, Ralph started to stroll past.95

Intrasentential cataphora is exhibited only by pronouns96, as opposed to

intersentential cataphora which can be signalled by non-pronominal noun phrases
too97:

(1.64) The former White House intern is now as famous as her ex-boyfriend. From
the deserts of Kazakhstan to the south seas of Tonga, everyone knows Monica
Lewinsky.

The nature of cataphora has been discussed and disputed by a number of researchers,
both within the generative framework and outside it.98 Some linguists such as Kuno
(1972, 1975), Bolinger (1977) and Cornish (1996) argue against the genuine
existence of cataphora claiming that alleged cataphoric pronouns must have, located
in the previous text, corresponding coreferential items. Their observations are based
on examples such as

(1.65) Though her party comprised 20 supporters, Hillary and a female colleague
were the only two eating and the bill was $6.99

where even though the occurrence of her appears to be cataphoric, this is not the case
if the extract is examined within the context (1.66) of the whole document, rather
than in isolation (1.65).

(1.66) At about 10am, two men in suits appeared, asking to talk to the manager. It
turned out they were Secret Service agents wanting to know if Hillary Clinton
could pop in for breakfast […] Though her party comprised 20 supporters,
Hillary and a female colleague were the only two eating and the bill was $6.

On the other hand researchers such as Carden (1982) and Tanaka (2000) demonstrate
that genuine cataphora does exist. Carden (1982) supports his argument with
approximately 800 examples of cataphoric cases where such pronouns are, as he
claims, the ‘first mention of its referent in the discourse’100. Such a type of cataphora
is described as ‘first-mention’ cataphora and counteracts the aforementioned
scepticism that assumes that each pronoun acting cataphorically must possess a
previously mentioned discourse referent.

94 Or more generally, at a lower level of syntactic structure than the antecedent.

95 William Golding, Lord of the Flies, p.164. Faber and Faber Limited: London, 1974.
96 Including demonstrative pronouns such as in the case He told me a story like this: ‘Once upon a
time... ‘ (Quirk et al. 1985).
97 I argue that in (1.64) The former White House intern is perceived to refer to a discourse entity
(person) not yet introduced and is therefore, viewed as cataphoric.
98 For a comprehensive account and update see (Tanaka 2000).
99 The Times: Times 2, 21 March 2000.
100 Carden, 1982, p.366.

23
The use of cataphoric references is typical in literary and journalistic writing and the
following is an example of genuine cataphora.

(1.67) From the corner of the divan of Persian saddle-bags on which he was lying,
smoking, as was his custom, innumerable cigarettes, Lord Henry Wotton
could just catch the gleam of the honey-sweet blossoms of a laburnum...101

As this text occurs in the second paragraph of the first chapter of the book and there
is no direct or indirect mention Lord Henry Wotton in the first paragraph of this
chapter, its title or in the title of the book, it would not be possible to analyse the
pronouns he and his as anything other than cataphoric.

1.11 Anaphora and deixis

In the example previously quoted

(1.24) He seems remarkably bright for a child of his age.

the pronoun he was not used anaphorically, but deictically: he did not refer to an item
previously mentioned in the discourse, but pointed to a specific person in a given
situation. The information that could have been derived from a potential antecedent
was not necessary on this occasion and the statement was not dependent on
information explicitly present in a text or discourse. However, if the above sentence
had been preceded by the sentence George is only 4 but can read and write in both
English and Bulgarian, the pronoun he would have been interpreted anaphorically.
Deixis is the linguistic phenomenon of picking out a person, object, place, etc. in a
specific context or situation. The interpretation of the deictically used expression is
determined in relation to certain features of the utterance act, such as the identity of
the speaker and addressee together with the time and place at which it occurs
(Huddleston, 1984). As an illustration, consider the utterance:

(1.68) I want you to be here now.

The deictic pronoun I refers to whoever is uttering the sentence and the pronoun you -
to whoever the addressee is. Similarly, the interpretations of here and now are
associated respectively with the place and time of the utterance.

Among the words typically used in a deictic way are the personal pronouns I, we, you
and their reflexive and possessive counterparts; the demonstratives this and that; the
locatives here and there and a variety of temporal expressions such as now, then,
today, tomorrow, yesterday, next week, last month, next year, in the last decade, this
century, last century, on Sunday etc.:

(1.69) I know that you will enjoy reading this chapter.

(1.70) I bet you were expecting that example.
(1.71) It was very fashionable to wear long hair then. (then deictic, e.g. uttered while
watching a film)

101 Oscar Wilde, Picture of Dorian Gray, Chapter 1.

24
(1.72) Last century has witnessed a real technological revolution. (Last century
deictic, e.g. uttered at the beginning of the 21st century)

I have already shown that 3rd person pronouns are usually anaphoric but sometimes
they can be used deictically (1.24); on the other hand most uses of 1st and 2nd person
are not anaphoric. Demonstrative pronouns such as that are used both deictically
(1.70) and anaphorically (When I used to ask my then102 two-and-a-half-year old son
‘George, would you like to eat a green pepper?’ he would reply ‘I don’t like that’).
Similarly, adverbs such as then can be both deictic (1.71) and anaphoric (1.35).
Finally, there are uses that are simultaneously anaphoric and deictic:

(1.73) Maggie came103 to England when she was four, and has lived here ever
since.104

In (1.73) here is deictic in that it refers to the place where the utterance occurs but at
the same time it is anaphoric to England, previously introduced in the text.

1.12 Anaphora and ambiguity

Many anaphors like she in (1.74)

(1.74) Jane told Mary she was in love.

are ambiguous - she could be either Jane or Mary. Equally ambiguous is the example

(1.75) Jane convinced Mary she was in love.

Often the level of ambiguity in similar examples depends on the semantics of the verb
or other components in the sentence or discourse.

(1.76) Jane informed Mary she was in love.

In this example it is more likely that Jane was in love because if Mary were in love
herself, perhaps she would not have needed to be informed of it.

Similarly,

(1.77) Jane told Mary she was in danger.

is ambiguous whereas in

(1.78) Jane warned Mary she was in danger.

Mary is by far the more probable antecedent because of the semantics of the verb to
warn which focuses on the person being warned (and hence, the danger to the
addressee).

102 Note the deictic use of then.

103 Note the deictic function of the verb to come as opposed to the verb to go.
104 Adapted from Huddleston (1984).

25
In practice, however, some readings are much more probable than others

(1.79) Jane told Sarah she was the nicest person she knew of.105

Even though this sentence is theoretically ambiguous (with four different meanings:
each she can be either Jane or Sarah), in practice it is much more probable that Jane
would praise somebody else rather than showing off so immodestly, therefore, Sarah
would be the preferred antecedent of the first she. Similarly, Jane is inevitably the
antecedent of the second she, since Jane cannot have ‘inside knowledge’ of what
Sarah knows.

These examples illustrate that in many cases of ambiguous anaphors there is a

probable, preferred or a default antecedent106, which is taken as the correct one ‘in the
absence of contradicting context or knowledge’ (Hirst 1981).

In many cases the preferred reading relies on extralinguistic knowledge such as

(1.80) Prime Minister Tony Blair had a fruitful meeting with President Yeltsin.
The old man has just recovered from a heart attack.

The antecedent of The old man is most probably President Yeltsin who is known to be
much older than Tony Blair and has poor health at the time of writing.

1.13 Anaphora and the resolution moment

The interpretation of anaphora may be delayed until other discourse elements

intervene to elucidate the anaphoric reference. This becomes clear in the following
example (Tanaka 2000, p. 221):

(1.81) Police officer David Cheshire went to Dillard's home. Putting his ear next to
Dillard's head, Cheshire heard the music also.

The disambiguation moment of the pronoun his is the moment the reader processes
Dillard's head. At this moment the reader would have no difficulty to instantiate
David Cheshire to the anaphor his instead of Dillard, since one cannot put one’s ear
next to one’s own head. Therefore, the resolution moment is not that of the pronoun
reading but a later one. Example (1.81) suggests that there is a distinction between the
point when a reader encounters an anaphor and begins to interpret it (initiation point),
and the point when the reader completes the interpretation of the pronoun (completion
point). As Sanford and Garrod (1989) note, the gap between the two points can be
almost nil as the case when a reader resolves a pronoun immediately after she/he
encounters it. In other cases, the gap can be extended to the end of the phrase, clause,
or sentence in which the pronoun is included. The problem of delayed resolution is
also discussed in (Cristea and Dima 2000).

105Adapted from Hirst (1981).

106 Not all ambiguous anaphors have a default such as in the examples (1.76), (1.78) and (1.79).

26
Summary

This chapter introduces the linguistic phenomenon of anaphora (the act of pointing
back to a previously mentioned item) and related phenomena and concepts.107 I have
shown that anaphora and coreference (the act of referring to the same referent in the
real world) are not the same thing even though important classes of anaphora involve
coreference. I have also outlined the related phenomena of cataphora (backwards
anaphora) and deixis (non-textual reference in a specific situation). The classification
of the varieties of anaphora proposed in this chapter aims to be simple enough for the
purpose of Natural Language Processing (NLP).108 I have pointed out that nominal
anaphora, that is, anaphora exhibited by pronouns and lexical noun phrases109 that
refer to noun phrases, is the most crucial and best understood class in NLP. I have
distinguished varieties of anaphora (i) according to the form of the anaphor
(pronominal, lexical noun phrase, noun, verb, zero anaphora etc.), (ii) according to
the location of the anaphor and the antecedent (intrasentential as opposed to
intersentential), (iii) according to the inference needed (indirect as opposed to direct)
and (iv) according to whether the anaphor and the antecedent have the same referent
in the real world or one of a similar description (identity-of-reference or identity-of-
sense anaphora). Finally, I have briefly discussed the typical distance between the
different varieties of nominal anaphora and their antecedents, and have alerted the
reader to the fact that anaphors may be ambiguous.

107 For detailed accounts (but not necessarily using the same terminology) see Brown and Yule
(1983), Halliday and Hasan (1976), Huddleston (1974), Quirk et al. (1985) and Lyons (1977).
108 For alternative and more comprehensive classifications see Hirst (1981) and Quirk et al. (1985).
Also see Cornish’s (1986) classification of anaphora based on the type of antecedent.
109 Lexical noun phrases include definite descriptions and proper names but not pronouns.

27