An Experimental Investigation of
the Antecedent Preferences of
Hebrew Subject Pronouns
Hamutal Meridor
Master of Science
Cognitive Science and Natural Language Processing
School of Informatics
University of Edinburgh
2006
AbstractA magnitude estimation study was conducted in order to investigate the likelihood of
different pronominal forms in Hebrew to be resolved to different antecedents. Subjects
were presented with sentences such as ”Dana wrote Nina whenφ stayed in the U.S.”
and were asked to rate the likelihood of a following sentence resolving the anaphor, e.g.
”Dana stayed in the U.S.”. The results showed that null and demonstrative pronouns
have a significant preference for subject and object antecedents, respectively, a result
in keeping with previous work (by Carminati, 2002; Bosch et al., 2006, among others).
The overt pronoun exhibited a significant bias only when used in logophors, where
it preferred a subject antecedent over an object one. This stands in contrast to what
Carminati (2002) and Sorace and Filiaci (2006) have found in Italian whereby the
object antecedent was preferred in these cases. The results are discussed in light of
Hebrew’s special pro-drop pattern, alongside the implications of the cross-linguistic
variance attested for.
i
AcknowledgementsI’d like to thank the many people who provided great help during the course of my
dissertation, whether professional, technical, inspirational, or other. In particular I
owe a special thank you to Antonella Sorace for devoting both time and thought in
giving me fruitful advice accompanied by genuine enthusiasm. Thank you to all the
people who took the time and effort to participate in my experiment. An infinitely
huge gratitude is devoted to my family - my brothers Mattan, Shaul and Avishai, and
especially my parents Dan & Leora (abba ve’imma), without whom this whole year in
general, and this dissertation in particular, would truly not have been possible. Last but
not least, the biggest thank you of all is granted to my supervisor, Frank Keller, for the
immense amounts of knowledge, understanding, patience, inspiration and kindness he
offered me throughout the whole project. I have no doubt this thesis would not have
taken the form it did if it were not for his superb guidance.
ii
DeclarationI declare that this thesis was composed by myself, that the work contained herein is
my own except where explicitly stated otherwise in the text, and that this work has not
been submitted for any other degree or professional qualification except as specified.
(Hamutal Meridor)
iii
Table of Contents
1 Introduction 1
2 Background 3
2.1 The pro-drop phenomenon . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.2 Processing null pronouns . . . . . . . . . . . . . . . . . . . . 4
2.2 Null vs. overt pronouns cross-linguistically . . . . . . . . . . . . . . 5
2.3 Demonstrative pronouns cross-linguistically . . . . . . . . . . . . . . 13
2.4 Pronouns in Hebrew . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Experiment 18
3.1 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.1 Participants . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2.2 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4 Discussion 34
5 Conclusions 37
5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
A Experimental Material 41
B Descriptive Statistics 49
Bibliography 50
iv
Chapter 1
Introduction
Anaphoric expressions have long been a focus of different language-related research
domains. Linguists have investigated the different constraints on potential antecedents
for anaphora in a variety of means, among them syntactic frameworks such as binding
theory (Chomsky, 1981), semantic-pragmatic frameworks like centering theory (Grosz
et al., 1995) and discourse-related ones as discourse representation theory (Kamp,
1981). Alongside linguists, computational linguists, in their pursuit of reference reso-
lution algorithms of noun phrases in general, have also striven to tackle the problem of
anaphora resolution in particular. Different algorithms explore the effects of syntactic,
semantic and other factors that might affect the antecedent preferences of pronouns
(Hobbs, 1978; Lappin and Leass, 1994). These factors have also been extensively in-
vestigated within the domain of psycholinguistics, where pronouns and anaphora have
indeed been a main research topic (Gernsbacher and Hargreaves, 1988; Cooreman and
Sanford, 1996).
While anaphora resolution appears to have been extensively researched in the separate
realms of linguistics, computational linguistics and psycholinguistics, other phenom-
ena in natural language have not received as much unified attention. Whereas linguists
have long been concerned with the typologically wide spread phenomenon of pronoun
dropping (pro-drop), psycholinguists on the other hand, as they have focused mainly
on English processing, have conducted surprisingly little research on the topic. This is
especially noticeable in light of the great attention the processing of pronouns in gen-
eral has received by psycholinguists. The existence of two separate forms to express
pronouns, both overt as well as covert, has led researchers to believe that they carry
1
Chapter 1. Introduction 2
different roles, whether in grammar or in processing. Despite grammarians’ notable
advances in depicting the anaphoric properties of the null pronouns, they have not been
able to produce a satisfying answer to the puzzling question of their existence in the
first place. Why would a language have two sets of pronominals, null and overt, both
filling the same role?
This project investigates the hypothesis that different types of pronouns have different
preferences in selecting their antecedents. It looks into the possibility that the exis-
tence of null versus overt pronouns can be accounted for by the anaphora resolution
mechanism. The project further tests another class of pronominals, the demonstrative
pronouns, which have a clear bias toward less salient antecedents (Bosch et al., 2003,
2006; Kaiser and Trueswell, 2004, 2005). In order to investigate the referential proper-
ties of the pronouns, the project makes use of the experimental paradigm of magnitude
estimation in judgement experiments, which enables to determine the difficulty in re-
solving anaphora of different forms.
Following this brief introduction, section 2 offers a review of the phenomenon of null
subjects, demonstratives and pronouns in general, alongside a critical review of the
relevant research that has been conducted in the field in other languages. Subsequently,
section 3 presents the experiment conducted under the scope of this dissertation. Then,
section 4 offers an analysis of the data obtained in light of the past research presented
in section 2 and the predictions made in section 4. Section 5 concludes by highlighting
the findings and offering some interesting unresolved issues for future research.
Chapter 2
Background
2.1 The pro-drop phenomenon
2.1.1 Review
Linguists have long been concerned with the typologically widespread phenomenon of
pronoun dropping (pro-drop). Pro-drop languages have two seemingly redundant sets
of pronouns, null and overt, both filling the same role. Thus, the two Hebrew sentences
(2.1 a, b) below carry the same meaning:
(2.1) a. φ axalti tapuax
φ ate1SG apple
b. ani axalti tapuax
I ate 1SG apple
’I ate an apple’
Pronouns may be dropped from different syntactic positions, carrying different gram-
matical roles. Not only the dropping of subjects as this study investigates (e.g. 2.1)
has instigated much theoretical interest, but also that of object pronouns (Cole, 1987;
Huang, 1995; Keller and Lapata, 1998).
Different languages exhibit different variations on pronoun dropping. Whereas lan-
guages such as Italian allow pronoun dropping ubiquitously, languages like English
regularly prohibit it. Traditionally, pronouns are believed to provide lexical informa-
tion that aids in resolving anaphora. Therefore, it seems reasonable that languages
3
Chapter 2. Background 4
as Italian, in which verbs carry marked agreement features that render the pronouns
superfluous, may drop the pronouns’ overt form. Jaeggli and Safir (1989) noted that
null subjects may be found either in languages in which the agreement morphology
is indeed sufficient without the overt subject, as well as in languages which have no
agreement or person feature at all. Chinese, for instance, has but a number feature
indicated by its pronouns in spoken language and allows a far greater null pronoun
distribution (Huang, 2000) than languages such as Italian and Spanish which do have
agreement features marked on their verbs. Hebrew, a special kind of pro-drop lan-
guage, lies somewhere between those two extremes as it generally allows null subjects
only in 1st and 2nd persons, in the past and future tenses. Furthermore, Hebrew has a
relatively free word order and allows both the canonical SVO1 order as well as VSO
and other less frequent constructions.
2.1.2 Processing null pronouns
Carminati (2002) points to two possible hypotheses one might conjecture in trying
to understand the reasoning underlying the existence of the two separate pronomi-
nal forms. First, since null subjects usually carry information regarding person and
number, but not gender2, one might speculate that there exists a preference to use the
overt subject in order to avoid ambiguity in a context with two gender-differentiated
referents. In other words, when an overt pronoun may disambiguate the reference, it
seems reasonable that it would be preferred over a null pronoun which maintains the
ambiguity. The second, somewhat contradictory, possible hypothesis Carminati brings
forward is one found on the general principal of economy of processing. This account
predicts that the use of the null form might be preferred as it is more economical both
in size and in the content it carries (less features).
Albeit the seeming plausibility of these hypotheses, Carminati’s findings render both
insufficient in explaining the perplexing phenomena of the processing of null and overt
subjects in intra-sentential anaphora in Italian. She proposes and provides evidence for
a processing hypothesis to account for the phenomenon, stipulating that the pronouns
1Although most linguistic theories follow the tradition of Greenberg (1966) in stipulating ModernHebrew word order to follow that of Biblical Hebrew and be underlying VSO, on the surface it iscommonly SVO, which is the crucial factor for the inherently serial processing task.
2This is true mainly in Romance languages but not Semitic languages, including Hebrew, whereseparate inflictions exist for the two genders
Chapter 2. Background 5
differ from each other in having distinct biases for antecedents. Her hypothesis, the
Position of Antecedent Hypothesis, states the following:
The Position of Antecedent Hypothesis for the Italian null and overt pro-nouns in intra-sentential anaphora: the null pronoun prefers an antecedentwhich is in the Spec IP position, while the overt pronoun prefers an an-tecedent which is not in the Spec IP position. (p. 33).
Carminati finds her hypothesis on the notion that pronouns have preferences for an-
tecedents based on a scale of prominence. She stipulates that null pronouns prefer the
most prominent antecedent, while their overt counterparts prefer the less prominent
one, if such is available. In particular, Carminati claims the prominence scale is one
of structural, syntactic, nature and that the [SPEC, IP] position (Chomsky, 1981) is the
most prominent one in the syntactic structure of the sentence. Hence, she claims the
null pronoun in Italian intra-sentential anaphora will prefer to take the argument in the
[SPEC, IP] position as an antecedent. It seems that Carminati’s strong commitment
to a specific syntactic framework (that offered by Chomsky and followers) might be a
rather binding approach theoretically.
2.2 Null vs. overt pronouns cross-linguistically
Null Pronouns in Italian
While linguists have tried to account for pro-drop by syntactic, lexical, discourse-
related and other means, psycholinguists have conducted surprisingly little research
on the topic. Rather, most psycholinguistic research has focused on the processing of
overt pronouns. Arnold et al. (2000), for instance, used an eye-tracking study to show
that gender information and accessibility influence the initial processing of overt pro-
nouns in English. Carminati (2002) provides evidence from Italian for her hypothesis
by which the existence of null pronouns derives from a division of labour between the
two pronominal forms with respect to anaphora resolution. She used self-paced read-
ing questionnaire studies to show that null and overt pronouns in Italian intra-sentential
anaphora have complementary biases in choosing their antecedents, as shown in (2.2):
(2.2) Marta scriveva frequentemente a Piera quandoφ/lei era negli Stati Uniti
’Martha wrote frequently to Piera whenφ/she was in the U.S.’
Chapter 2. Background 6
a. Quando Marta era negli Stati Uniti
’When Marta was in the States’
b. Quando Piera era negli Stati Uniti
’When Piera was in the States’
Carminati demonstrated that people are more likely to resolve the null pronoun as re-
ferring to the subject and the overt ones as referring to the object in her questionnaires.
In (2.2) above this preference translates into different interpretations of who was the
person that was in the U.S. according to the use of the different pronouns. When the
pronoun is absent (null pronoun), Martha is gathered to have been in the U.S. (par-
ticipants chose option a), whereas when the overt pronoun is used Piera is the one
understood to have been there (option b).
In addition to this same-gender case, Carminati also shows that the gender information
carried by the overt pronoun does not render it preferable on its covert counterpart even
in a gender-ambiguous context. This suggests that not only is the human processor
affected by the form of the pronoun, but that processing might also account for the
nature of the distribution of pronouns, alongside other linguistic factors.
Sorace and Filiaci (2006) have studied the resolution of null and overt pronominal
anaphora in near-native and native speakers of Italian. Their results provide more fine-
grained observations as to the nature of the different pronouns’ preferences. Specifi-
cally, they provide evidence for differences between forward (anaphor) and backward
(logophor) anaphora with respect to the processing of pronouns. In processing terms,
there exists an important difference between the two types of anaphors in that in for-
ward anaphora, where the main clause precedes the subordinate clause, the proces-
sor encounters both potential antecedents before having encountered the pronoun. In
backward anaphora on the other hand, where the subordinate clause precedes the main
one, the processor first encounters the pronoun and does not encounter either of the
potential antecedents until after having encountered their assignment. As Sorace and
Filiaci note, this type of anaphora poses certain requirements for the processor that are
of particular interest. Since the pronoun linearly precedes its potential antecedents,
there are no prior discourse constraints that are imposed on the resolution in advance.
Research on logophoric anaphora resolution in English has shown that the processor
tries to resolve the pronoun with the closest possible antecedent, even in the absence
of bottom-up information (Cowart and Cairns, 1987; Kazanina, 2005; Kazanina et al.,
Chapter 2. Background 7
2005). Furthermore, there seems to be a particular expectation by the English proces-
sor to resolve the backward anaphoric dependency in the matrix subject position (van
Gompel and Liversedge, 2003).
In Italian, Sorace and Filiaci found a different pattern of behaviour for both type of
pronouns in the two anaphor forms. They used a picture verification task in which
the participants were asked to indicate the picture corresponding to the meaning of the
subordinate clause which resolves the anaphor, thus identifying the antecedent for the
pronoun. For null pronouns, they found that while in backward anaphora there is a
clear overwhelming preference for subject antecedents as Carminati had found, this
preference does not exist in forward anaphora. In the latter sentences, preferences of
the native speakers were equally divided between the subject and the complement of
the matrix clause. Thus, in (2.3) below, when the null pronoun is used, in (a) the one
wearing the coat is more likely to be the mother while in (b) it is as likely to be the
mother as it is the daughter.
(2.3) a. Mentre lei/φ si mette il capoto la mamma da un bacio
while she/φ wears the coat the mother gives a kiss
alla filgia
to the daughter
’While she/φ is wearing her coat, the mother kisses her daughter.’
b. La mamma da un bacio alla filgia mentre lei/φthe mother gives a kiss to the daughter while she/φsi mette il capoto
wears the coat
’The mother kisses her daughter, while she/φ is wearing her coat.’
In the overt pronoun case, Sorace and Filiaci also found different preferences for the
anaphor and logophor cases. In forward anaphora cases, native speakers significantly
preferred the complement over the subject and extralinguistic referent. In (2.3b) above,
that translates to preferringlei to refer to the daughter rather than the mother. In the
backward anaphora case however, a strong preference toward an extralinguistic an-
tecedent was found, followed by the complement as a distant second possibility, and
the subject as an even less likely one. Hence, in (2.3a) above, people are more likely to
understandlei as referring to an extralinguistic antecedent, then the daughter, and very
Chapter 2. Background 8
infrequently the mother. As Sorace and Filiaci note, it is quite surprising and not clear
why the extralinguistic antecedent is so acceptable in this context that it supersedes the
complement antecedent.
Null Pronouns in Spanish
Evidence for a division of labour between null and overt pronouns in anaphora resolu-
tion exist in other Romance languages as well. Alonso-Ovalle et al. (2002) investigated
the anaphoric behaviour of pronouns in Spanish, both in intra-sentential as well as in
inter-sentential anaphora. They used written questionnaires to show that the null pro-
noun in Spanish prefers a subject antecedent, while the overt pronoun does not. In ex-
ample (2.4a) participants tended to assign Juan, which is the subject, as an antecedent
of the null pronoun in the majority of the trials, while in (2.4b)El was assigned either
Juan or Pedro equally. This stands in contrast to Carminati’s findings in Italian by
which the overt pronoun is more likely to be linked to the less salient antecedent.
(2.4) a. Juan pego a Pedro. φ Esta enfadado
Juan hit Pedro. φ is angry
b. Juan pego a Pedro. El Esta enfadado
Juan hit Pedro. He is angry
’Juan hit Pedro. He is angry’
A possible methodological issue Alonso-Ovalle et al. have not considered and might
be influencing their results lies in their choice of experimental material. Most of the
sentences they used feature statives in the second, anaphor resolving, sentence (e.g.
’He is angry.’). This may have a potential effect on the antecedent preferences. It
might be argued, for example, that statives are less inclined to a topic shift than verbs
and are therefore more likely to be resolved by the more salient antecedent, the subject
(this might provide an explanation as to why no preference toward object antecedents
was found for the overt pronouns in this study).
Furthermore, some of Alonso-Ovalle et al.’s lexical choices do not intuitively seem
pragmatically neutral, such as sentence (2.5) below.
(2.5) Maite entretuvo a Elena.Φ/Ella esta cansada.
’Maite entertained Elena. She is tired.’
In fact, in the absence of pre-testing of the stimuli, it is rather hard to determine whether
the bias Alonso-Ovalle et al. report stems from the pronouns themselves or is it rather
Chapter 2. Background 9
an artefact of the experimental material chosen.
Aside from extending the validity of Carminati’s theory regarding null subjects to
Spanish, Alonso-Ovalle et al. (2002) also present evidence suggesting that the topic-
focus articulation of the sentence containing a pronoun affects the pronouns’ general
anaphoric preferences. Specifically, they show that when an overt pronoun is a pre-
verbal subject3 it is more likely that the pronoun refers to prominent antecedents (sub-
jects), thus overriding the overt pronoun’s normal neutrality with respect to its an-
tecedent. In (2.6), the pronominal subject of the embedded sentence can either have
Pedro as an antecedent, or have an obviative reading (a reading in which the antecedent
for the pronoun is found outwith the sentence).
(2.6) a. Pedro piensa que esta cansado el
Pedro thinks that is tired he
b. Pedro piensa queel esta cansado
Pedro thinks that he is tired
’Pedro thinks that he is tired’
After having established that the syntactic position dictates topicality, Alonso-Ovalle
et al. turned to testing whether the properties of the null pronoun also stem from
its syntactic position, the preverbal subject position. By manipulating the position of
the pronoun and thus its topicality (or lack of), they predict and ascertain that Pedro
is chosen as an antecedent more frequently in the topical (2.6b) than the non-topical
(2.6a). Alonso-Ovalle et al. deem this finding as implying that the preferences of the
pronouns in fact result from a wider principle of the interpretation associated with the
special syntactic position the null pronouns typically occupy rather than the lexical or
semantic properties of the pronouns themselves.
Null Pronouns in Romanian
Another Romance language in which evidence of different pronoun forms affect the
preference for their antecedents is Romanian. Diaconescu and Goodluck (2004) stud-
ied the d(iscourse)-linked attraction effect in Romanian, another Romance pro-drop
language. A d-linked wh-phrase is a phrase such aswhich boy, while a non d-linked
phrase is a phrase such aswho. The d-linked effect, first introduced by Frazier and
Clifton (2002) in English sentences, stipulates that d-linked and non d-linked phrases
3Alonso-Ovalle et al. (2002) separately show the preverbal subject is interpreted as a topic in Span-ish.
Chapter 2. Background 10
have different biases in choosing their pronoun antecedents. Frazier and Clifton point
out that d-linked phrases imply the existence of a set of contextually determined enti-
ties from which the speaker chooses. For instance, when introducingwhich boy, the
speaker refers back to a set of already contextually existent boys. Non d-linked phrases
such aswho, on the other hand, carry no such implication. Frazier and Clifton showed
that d-linked phrases like (2.7b) are more likely to attract the reference of a pronoun in
a subordinate clause than non d-linked phrases such as (2.7a).
(2.7) a. Rick knew who Janice sang a song to before he went to sleep.
b. Rick knew which brother Janice sang a song to before he went to sleep.
Diaconescu and Goodluck were able to replicate the pronoun attraction effect Frazier
and Clifton had found in English for both null and overt subjects in Romanian by
using a similar questionnaire type experiment the latter have used. In direct questions,
subjects attached both types of pronouns to the wh-phrase more often when it was a
d-linked phrase than when it was not. For instance, in (2.8) below,φ andhe chose
which brotheras an antecedent more frequently thanwho.
(2.8) Cui/carui baiat I -a trimis Barbu o pusca atunci candφ/el a fost amenintat?
’Who/Which guy did Barbu send a rifle to whenφ/he was threatened?’
The prediction that the overt pronoun might carry a stronger effect due its being a
focused element was not confirmed. In fact, it was an unpredicted result of their ex-
periment that has come to provide evidence relevant to our study. The null subject had
a higher proportion of co-reference with the wh-phrase than its overt counterpart did.
This is exactly as Carminati and Alonso-Ovalle et al. would have it, the null pronoun
being more likely to be linked to the most prominent element.
Importantly, a different kind of experimental sentence did not strictly follow the syntactically-
driven model offered by Carminati (2002). In embedded question sentences such as
(2.9), Carminati would wrongly predict the null pronouns to refer to the main clause
subject Barbu, whereas they were found to actually refer to the wh-phrase.
(2.9) Radu a stiut cui/carui frate i -a cantat Ioana un cantecinainte caφ/el sa se fi
dus la culcare
’Radu knew who/which brother Ioana sang a song to beforeφ/he went to
sleep’
Chapter 2. Background 11
Diaconescu and Goodluck offer an explanation to this exception in that the search for
an antecedent for the null pronoun might be restricted within its clausal domain. They
suggest that since the only possible antecedent in the immediate domain is not a good
candidate for reference of the overt pronoun, a search for its antecedent may proceed
upwards beyond the domain more rapidly than the search for a null pronoun antecedent
would. As they point out, this account also explains their near-significant finding that
the null pronoun was more inclined to attach to the wh-phrase in the embedded ques-
tions than the non-embedded sentences, since in the former, the wh-phrase is the sole
possible antecedent in the local domain.
Null Pronouns in Chinese
Self-paced reading-time experiments combined with questionnaires conducted by Yang
et al. (1999, 2001) show that the Chinese null pronouns also have a preference as to
their antecedent, although the overt ones do not. Yang et al. followed the prevalent
methodological assumption by which faster reading times indicate greater acceptabil-
ity (e.g. Arnold et al., 2000), and in this case greater likelihood of the antecedent
to refer to the pronoun. Yang et al. studied the use of null and overt pronouns in
extra-sentential anaphora in Chinese, a pro-drop language. As previously pointed out,
unlike the rich morpho-syntactic systems of the Romance languages Chinese belongs
to a different class of languages, such that do not have verb agreement and do not
superfluously encode agreement features thus rendering the overt pronoun redundant.
Therefore, one might expect a different mechanism in the processing of the pronouns
in Chinese.
Yang et al. (1999) used materials such as (2.10), translated into English:
(2.10) Context sentence: Xiaomei (feminine) told Xiaorong (feminine) that veg-
etables, instead of flowers, should be planted in the garden.
Critical Sentence:
Continue Condition: She/φ thought vegetables are of more utility than
flowers.
Shift Condition : She/φ thought, however, that both vegetables and flowers
should be planted.
Final Sentence: The usage and planning of a garden are worth studying.
The critical sentence containing either form of the pronoun was manipulated such that
it was pragmatically disambiguated toward the same-gender subject or object in the
Chapter 2. Background 12
continue and shift conditions, respectively. Yang et al. found that the form of the
pronominal does not interact with the continue/shift conditions. Their findings sug-
gested that in Chinese the null pronoun indeed prefers a subject antecedent, but they
did not find the complementary behaviour for the overt pronoun, as Carminati had
found in Italian. The overt pronoun has no antecedent bias and can equally easily
retrieve subject and object antecedents, much like the Spanish examples.
In a similar experiment, whose results were later replicated (Yang et al., 2001), the
gender of the referents was manipulated so it could be disambiguated by an overt
pronoun, as shown in (2.11):
(2.11) Context sentence: Daxing (masculine) told Xiaorong (feminine) that veg-
etables, instead of flowers, should be planted in the garden.
Critical Sentence:
Continue Condition: He/φ thought vegetables are of more utility than flow-
ers.
Shift Condition : He/φ thought, however, that both vegetables and flowers
should be planted.
In the shift condition, but not in the continue condition, Yang et al. found that the
overt, gender disambiguating, pronoun was read significantly faster than the null pro-
noun, which, in turn, was read significantly faster in the continue condition than in
the shift condition. These findings indicate a preference to provisionally assign the
null pronoun to a subject antecedent and thus the retrieval of the object antecedent is
penalized(in the shift condition). The Chinese overt pronoun, on the other hand, dif-
fers from the Italian one’s behaviour in that it is more flexible in retrieving either kind
of referent. Whereas in Italian Carminati found evidence of overt pronouns assigning
subjects as their antecedents only in gender disambiguated or one-referent contexts but
not in general, the Chinese overt pronoun was found to retrieve subjects also in other
contexts.
Specifically, the overt pronoun showed no special preference as to its antecedent.
Carminati notes that while the connection of the null pronoun with a highly salient ref-
erent might well be a universal feature, the preference of the overt pronoun antecedent
might be a continuum between the two extreme options exemplified by Italian and Chi-
nese. It might therefore be interesting to see whether and where Hebrew, a language
that exhibits distinctively different null pronoun distribution than both these languages,
Chapter 2. Background 13
lies somewhere on that continuum.
2.3 Demonstrative pronouns cross-linguistically
Several recent studies have investigated the distribution of demonstrative pronouns
and their referential properties. Bosch et al. (2003, 2006) used reading time and sen-
tence completion experiments as well as corpus evidence to show that German demon-
stratives have a bias toward non-subject antecedents, such as the following example
demonstrates:
(2.12) Paul wollte mit Peter laufen gehen. Aber er/der war erkaltet.
’Paul wanted to go running with Peter. But he/DEM had a cold.’
The demonstrative pronounder is naturally interpreted to refer to the grammatical
object of the previous sentence,Peter, while the personal pronouner seems to have a
preference to refer to the grammatical subjectPaul.
In their corpus study, Bosch et al. (2003, 2006) investigated the distribution of pro-
nouns and demonstratives within sentences, as well as the distribution of their an-
tecedents in prior sentences. They found a significant difference in use between demon-
strative and personal pronouns. Their findings suggested that the typical use of the
German demonstrative is as a pre-verbal subject referring back to an NP object of the
prior sentences, while the typical use of the German pronoun is as a pre or post-verbal
subject referring back to the NP subject of the prior sentence. Bosch et al. (2006) car-
ried on by conducting self-paced reading-time experiments followed by questionnaires
or completion questions testing the preferences of the demonstrative versus the overt
pronoun. The completion studies repeated the sentence containing the anaphor, leav-
ing a definite article followed by a gap at the position of the pronoun, hence forcing
the subjects to resolve the anaphor with a specific noun phrase of the two potential
antecedents. The questionnaire studies used a comprehension question rather than the
completion sentence, however the form of that question and its possible answers, if at
all, remains vague in their paper. Aside from manipulating the pronouns themselves,
they also manipulated the pragmatic environment in which the sentences were read
based on the intuitive observation that world-knowledge may heavily affect the resolu-
tion of the pronouns, as demonstrated in (2.13):
Chapter 2. Background 14
(2.13) a. Im Krankenhaus.
At the hospital.
b. Der Oberarzt untersucht den Notfallpatienten.
The senior doctor is examining the emergency patient.
c. Er/Der ist namlich Herzspezialist.
He/DEM is a heart specialist.
d. Er/Der muß sofort operiert werden.
He/DEM must be operated on at once.
While in (2.13c) the world-knowledge clearly dictates a preference for the subject an-
tecedent, in (2.13d) the preferred reading is the one where the object is the antecedent.
When using a demonstrative, that is argued to prefer object antecedents, with a reading
that calls for a subject antecedent such as (2.13c), we would expect a difficulty for the
reader. Indeed, Bosch et al. found longer reading times for those conditions where
world knowledge supports the subject of the context sentence. Moreover, in their com-
pletion questionnaires they observed that in these conditions more participants tended
to disregard world knowledge and choose the object when the pronoun was a demon-
strative than when the pronoun was a personal pronoun. However, all the results they
report lack the crucial support of statistical data, rendering them untestable at best, if
not questionable.
The central observation Bosch et al. made was that in those cases where there is a world
knowledge bias for the interpretation of the pronoun, world knowledge is clearly the
dominating factor in pronoun comprehension. In the unbiased neutral cases they found
a clear preference for one interpretation only where the demonstrative pronoun referred
to the context sentence object, suggesting that the demonstrative does indeed carry its
own lexical preference as to its antecedent, unlike the personal pronoun. However, as
Bosch et al. point out it has hard to determine whether this finding does indeed stem
from some special property of the demonstrative correlated with its grammatical func-
tion, or whether it is but a mere side effect. Specifically, the observation which casts
doubt on the idea that the use of demonstrative pronouns is exclusively determined by
the grammatical role of the antecedent is the clear preference of demonstrative pro-
nouns to occur in the pre-verbal position of the sentence Bosch et al. have found in
their corpus study.
Chapter 2. Background 15
The demonstrative preference toward a non-salient antecedent is indeed quite a preva-
lent cross-linguistic finding. Further reenforcing evidence for the referential properties
of demonstrative pronouns come from Kaiser and Trueswell (2004, 2005). They used
completion studies and eye-tracking experiments in order to present evidence that the
choice of a demonstrative over an overt pronoun in Finnish (Kaiser and Trueswell,
2005) and Dutch (Kaiser and Trueswell, 2004) is triggered by the salience of the refer-
ents. Interestingly, they found that this is not the case for the choice of reduced versus
overt pronouns in Dutch4.
2.4 Pronouns in Hebrew
As previously mentioned, Hebrew is a partial pro-drop language as it generally al-
lows null pronouns only in 1st and 2nd persons, in tensed clauses (past and future), as
depicted in Table 2.1, adapted from Hacohen and Schaeffer (2005).
Pronoun Person Past Present Future
ani 1 sg m amarti *omer omar
ani 1 sg f amarti *omeret omar
ata 2 sg m amarta *omer tomar
at 2 sg f amart *omeret tomri
hu 3 sg m *amar *omer *yomar
hi 3 sg f *amra *omeret *tomar
anaxnu 1 pl m amarnu *omrim nomar
anaxnu 1 pl f amarnu *omrot nomar
atem 2 pl m amartem *omrim tomru
aten 2 pl f amarten *omrot tomarna
hem 3 pl m amru *omrim *yomru
hen 3 pl f amru *omrot *tomarna
Table 2.1: Pro-drop distribution
However, although Hebrew generally does not allow pronoun dropping in third person
e.g. (2.14) below, the phenomenon is quite prevalent in certain complex structures and
specifically, in sentences involving subordinate clauses, such as (2.15).
4In the accessibility literature (e.g. Ariel, 1990), reduced pronouns are considered to be higher thannull pronouns and lower than overt ones.
Chapter 2. Background 16
(2.14) *φ telex la-mis’ada
* φ go FUT 3SGF to-restaurant
* ’will go to the restaurant’
(2.15) Dana amra se-φ-telex la-mis’ada
Dana said3SGF that-φ-go FUT 3SGF to-restaurant
’Dana said that she will go to the restaurant’
Third person null subject constructions potentially give rise to different preferences.
Examining the overt form of (2.15), shown in (2.16), we notice thathi might refer
either to Dana or to a different female antecedent, provided that such exists in the
context.
(2.16) Dana amra se-hi telex la-mis’ada
Dana said3SGF that-she goFUT 3SGF to-restaurant
’Dana said that she will go to the restaurant’
While this ambiguity does exist in potential in sentences such as (2.16), as Carminati
pointed out it is much more evident in cases where there exists more than one possible
antecedent within the sentence boundaries, such as (2.17):
(2.17) Dana katva le-nina kse-φ/hi hayta be-skotland
Dana katva3SGF to-Nina when-φ/she was3SGF in-Scotland
’Dana wrote Nina when she was in Scotland’
Both the null as well as the overt pronoun in the relative clause are potentially ambigu-
ous and can refer to either one of the two antecedents Dana and Nina. In Italian, as
Carminati (2002) showed, the null pronoun is more likely to be resolved to the subject,
whereas the overt one is more likely to be resolved to the object. According to this
logic, in (2.17) Dana would be the one understood to be in Scotland when the null
pronoun is used, and Nina the one when the overt pronoun is used.
Carminati further showed that in gender-ambiguous contexts the overt form is less
acceptable than in gender-unambiguous contexts equivalent to (2.17). The main dif-
ference between Italian and Hebrew in this respect is that the Hebrew 3rd person verb
infliction marks the gender of the subject, unlike Italian where both male and female
forms of the verb are identical in 3rd person. Therefore, in Hebrew the gender ambigu-
ous contexts such as (2.18) below will in fact be resolved upon the processing of the
Chapter 2. Background 17
verb, and are therefore predicted to produce results similar to the gender-unambiguous
contexts.
(2.18) Dana katva le-ro’i kse-φ/hi hayta be-skotland
Dana katva3SGF to-Roy when-φ/she was3SGF in-Scotland
’Dana wrote Roy when she was in Scotland’
* ’Dana wrote Royi when shei was in Scotland’
In addition to the null and overt personal pronouns Hebrew makes use of, there also
exists a set of demonstrative pronouns in the language. Replacing the overt pronoun in
(2.17) with a demonstrative will result in a preferred resolution of the pronoun to the
object, as shown in (2.19):
(2.19) Dana katva le-nina kse-zo hayta be-skotland
Dana katva3SGF to-Nina when-sheDEM was3SGF in-Scotland
’Dana wrote Ninai when shei was in Scotland’
This intuitive preference is in keeping with the findings of Bosch et al. (2003, 2006) and
Kaiser and Trueswell (2004, 2005). It suggests that different pronouns might indeed
have different preferences as to their antecedents and specifically, that the demonstra-
tives seem to have a strong preference toward non-subject antecedents.
Chapter 3
Experiment
The experiment aimed at testing the existence of different pronouns’ preferences in
Hebrew in choosing their antecedents. This was investigated by testing the accept-
ability of different pronouns being resolved to subject or (indirect) object antecedents.
Specifically, the experiment elicited magnitude estimation judgements in two-sentence
discourses, where the pronouns - null, overt or demonstrative - were resolved to one of
the two potential antecedents.
3.1 Predictions
Following the research that has been conducted in the area in other languages, however
preliminary it might be, a certain amount of the investigated data was predicted to
adhere to the cross-linguistic trends found.
Demonstrative Pronouns
Bosch et al. (2003, 2006) and Kaiser and Trueswell (2004, 2005) have investigated
the referential properties of demonstrative pronouns in German, Dutch and Finnish,
respectively. The demonstratives in all three languages exhibited similar behaviour in
their preference for less salient antecedents (i.e.: non-subject antecedents).
In Hebrew, demonstrative pronouns have a clear preference toward non-subject an-
tecedent, to the extent of the ungrammaticality of (3.1) below1:
1This sentence is ungrammatical when standing on its own. As a part of a discourse, however, thedemonstrative could potentially be understood as referring to an extra-sentential entity, provided that
18
Chapter 3. Experiment 19
(3.1) *Dana katva le-roi kse-zo hayta be-skotland
Dana wrote3SGF to-Roy when-sheDEM was3SGF in-Scotland
* ’Dana wrote Roy when she was in Scotland’
Since the demonstrative pronounzo is female, it must refer back to a feminine an-
tecedent. The only possible feminine entity in this sentence is Dana, who is in the
subject position. Yet the sentence is ungrammatical, suggesting that the demonstrative
cannot refer back to the subject. The object antecedent in (3.1) is masculine and thus
renders the sentence ungrammatical. When the gender of the object antecedent is in
agreement with the pronoun as in (3.2), where the masculine form of the demonstra-
tive zeis used, the sentence is completely grammatical, suggesting the pronoun indeed
refers to the object.
(3.2) Dana katva le-roi kse-ze haya be-skotland
Dana wrote3SGF to-Roy when-heDEM was3SGM in-Scotland
’Dana wrote Roy when he was in Scotland’
Based on these observations, alongside the cross-linguistic findings presented above,
we would be on safe grounds to expect the demonstrative pronouns in Hebrew as well
to have a strong preference toward object antecedents.
Null Pronouns
As previously shown in section 2, there exists a wide range of cross-linguistic evi-
dence from languages that exhibit distinctively different lexical, syntactic and other
characteristics (e.g. Chinese, Italian) pointing all to the null pronouns’ preference to-
ward subject, more salient, antecedents. In light of these findings, it seems fairly sound
to expect that Hebrew too will adhere to the same behaviour of its null pronouns an-
tecedent preferences. Thus, in sentences such as (3.4), we would expect people to
resolve Dana as the one that was in Scotland.
(3.3) Dana katva le-nina kse-φ hayta be-skotland
Dana katva3SGF to-Nina when-φ was3SGF in-Scotland
’Dana wrote Nina when she was in Scotland’
However, due to Sorace and Filiaci’s (2006) finding suggesting that this is not in fact
the case in forward anaphora, but only in backward anaphora, we might find a differ-
ence in the antecedent preferences between the different anaphora directions in Hebrew
such is given by the context.
Chapter 3. Experiment 20
null pronouns as well. Although, even if not indifferent to their antecedent’s grammat-
ical function in the forward anaphora case, we would still not be surprised to find the
null pronouns exhibit a stronger preference effect for subject antecedents in the back-
ward anaphora case.
Overt Pronouns
Unlike the null pronoun case, the evidence in light of a strong preference of the overt
pronoun is hardly as uniform and convincing as that of the null pronoun. Despite
Carminati’s (2002) finding that overt pronouns in Italian resolve to non-subject an-
tecedents, other studies suggest differently. Yang et al. (1999, 2001) found that Chinese
overt pronouns are not biased as to their antecedent altogether and tend to prefer sub-
ject antecedents equally as they do object antecedents. Such is the case with Spanish
overt pronouns as well (Alonso-Ovalle et al., 2002). Therefore, it might seem reason-
able to predict that Hebrew will also show no such preference. However, since there
do seem to be cross-linguistic differences regarding this phenomenon, no prediction
could be soundly made.
Any result regarding the behaviour of the overt pronouns would be plausible, although
different results would carry with them different implications. Should the overt pro-
noun in the forward anaphora case in Hebrew be found to prefer object antecedents,
it would support Carminati’s claim, causing it hold in a wider cross-linguistic setting.
On the other hand, should the overt pronoun reveal no preference as to its antecedent
or if it would reveal a preference toward subject antecedents, it would seem that the
overt pronoun preferences are not as universal as the null pronoun’s ones. In this
case, Carminati’s suggestion of an existence of a cross-linguistic continuum of overt
pronoun preferences along which different languages are situated would seem suit-
able. The latter scenario, where Hebrew overt pronouns prefer subject antecedents in
forward anaphora, will render Italian and Hebrew to reside on both extremes of that
continuum, while Chinese will be located somewhere in the middle between them, its
overt pronoun having no referential bias.
Furthermore, a closer look into the results from Italian might shed some light on other
potential factors that may be affecting the anaphora resolution. Sorace and Filiaci
(2006), who conducted their study on Italian as well, found that Carminati’s find-
ings by which overt pronouns prefer less salient antecedents holds only for forward
anaphora, but not for logophors (backward anaphora). In logophors the most plausible
Chapter 3. Experiment 21
referent is an extra-linguistic one, and not the less salient of the two antecedents within
the boundaries of the sentence. With respect to the null pronouns, on the other hand,
Sorace and Filiaci were able to replicate Carminati’s findings only in the backward
anaphora case, and not the forward one (as she has with the overt pronouns). This
suggests the type of the anaphor might also play a role in the Hebrew overt pronoun
resolution, but due to the little, somewhat contradictory, results regarding the direction
of the anaphor’s effect, it seems rather unsafe to predict what it will be in our experi-
ment. It will suffice to say that any result will undoubtedly be interesting and call for
further investigation of the matter.
3.2 Method
3.2.1 Participants
Forty-eight native speakers of Hebrew participated in the experiment. The subjects
were recruited over the Internet by posting emails. Participation was voluntary and
unpaid. The data of five subjects were excluded. One was excluded due to lack of
completion of the personal details section. Another was excluded as she turned out
to be bilingual, having been brought up outwith Israel and exposed to more English
than Hebrew throughout the majority of her childhood. The three remaining subjects’
data were excluded due to them not following the instructions in such a manner that
did not allow the statistical processing of their data2. Furthermore, not following the
instructions made their entire data set questionable as it suggests they might have not
read the instructions carefully and might have misunderstood the nature of the task
altogether. Thus, forty-three subjects were left for analysis, of which twenty-three
were male and twenty female. Thirty-six subjects were right-handed and seven left-
handed. The age of the subjects ranged from twenty-one to fifty-nine years, the mean
being 31.9 years.
2These subjects assigned a value of zero to their reference item. Since all data were normalised bydividing the judgements by the reference number, as described in section 3.4, a reference value of zeroprevented their data from being normalised, and subsequently, used.
Chapter 3. Experiment 22
3.2.2 Materials
Training Materials
The experiment included a set of training materials whose goal was to familiarise sub-
jects with the nature of the magnitude estimation task. The training set contained six
horizontal lines, ranging between largest and smallest at a ratio of 1:10. The lines were
evenly distributed over this range, with the largest item covering the maximal window
width of the web browser. The modulus used was a line of intermediate length.
Practice Materials
After having introduced the subjects with the magnitude estimation method using vi-
sual estimation of line lengths, a set of practice items was used in order to familiarise
subjects with applying magnitude estimation to linguistic stimuli of the type they were
to later encounter in the test section. The practice set consisted of six sentences that
were representative of the test materials in some crucial respects. The items included
two proper names serving as potential antecedents, and some form of anaphora that
refers to one of them, either using a forward anaphor, or a backward one. The anaphor
type included pronouns, alongside case markers and marked prepositions. The gram-
matical functions of the potential antecedents in the items varied from subject or indi-
rect object (similar to the experimental materials) to unrelated constructions (similar
to the fillers). A wide spectrum of acceptability was covered, ranging from fully ac-
ceptable to severely unacceptable (the acceptability range in the practice items was
representative of that in the experimental items). The item used as modulus was a sen-
tence in the middle range, i.e., a sentence that was neither fully acceptable nor fully
unacceptable.
Test Materials
Twenty four two-sentence discourses similar in form to example (3.4, brought again
below for convenience) were constructed.
(3.4) Dana katva le-nina kse-φ hayta be-skotland
Dana katva3SGF to-Nina when-φ was3SGF in-Scotland
’Dana wrote Nina when she was in Scotland’
Each occurred in one of twelve forms, similar to the ones presented in Table 3.1. The
forms differed in three variables that were manipulated: the type of the pronoun (null,
Chapter 3. Experiment 23
overt or demonstrative), the antecedent (subject or indirect object) and the type of
the anaphor (forward or backward). A full list of materials is given in Appendix A.
The first sentence comprised of two clauses. The main clause included two proper
names serving at either one of two grammatical functions: subject or indirect object.
Both names were unambiguously of the same gender (12 sentences with male names,
12 with female names). The subordinate clause included a pronoun that referred to
one of the entities in the main clause. All three types of pronouns: null, overt or
demonstrative, could potentially refer to either one of the two proper names in the
main clause.
First Second Antecedent
Dana wrote Nina whenφ was in Scotland. Dana was in Scotland.
Dana wrote Nina whenshewas in Scotland. Dana was in Scotland.
Dana wrote Nina whensheD was in Scotland. Dana was in Scotland.
Dana wrote Nina whenφ was in Scotland. Nina was in Scotland.
Dana wrote Nina whenshewas in Scotland. Nina was in Scotland.
Dana wrote Nina whensheD was in Scotland. Nina was in Scotland.
Whenφ was in Scotland Dana wrote Nina. Dana was in Scotland.
Whenshewas in Scotland Dana wrote Nina. Dana was in Scotland.
WhensheD was in Scotland Dana wrote Nina. Dana was in Scotland.
Whenφ was in Scotland Dana wrote Nina. Nina was in Scotland.
Whenshewas in Scotland Dana wrote Nina. Nina was in Scotland.
WhensheD was in Scotland Dana wrote Nina. Nina was in Scotland.
Table 3.1: Experimental conditions
Both the proper names and the verbs were partly taken from the Hebrew treebank
(Simaan et al., 2001), based on their frequency, and partly from miscellaneous web-
sites. All verbs appeared in past tense, and no statives were used. The verbs were
carefully selected so as not to carry any potential pragmatic influences that are inher-
ent to the verbs themselves3. Furthermore, the items were chosen so as to be as neutral
as possible. In particular, the actions depicted in the main and subordinate clauses
were chosen such that the main clause verbs will not be implicative of the antecedent
of the subordinate clause from world-knowledge etc. All items were spelled usingktiv
3For instance, a subject in a pilot study pointed out that when using the verbhizdahata, empathized,it is more likely that the pronoun refers to the indirect object, describing the reason for the empathy,e.g.’Dana empathized with Nina when she got divorced’.
Chapter 3. Experiment 24
male, meaning the optional vowel letters appeared, in order to avoid ambiguities4. All
sentences were pre-tested in a pilot study so as to rule out biased or unclear stimuli.
In order to investigate the resolution of the anaphor to one of the potential antecedents,
the stimuli had to contain the resolution of the anaphor in some manner. As mentioned
in section 2.2, in Carminati’s 2002 studies for instance, the subjects were presented
with the two possible options to resolve the anaphor and had to chose the more nat-
ural option between them. A possible manner of translating this methodology into a
magnitude estimation study was simply to ask the subjects to rate the likelihood of
resolving the anaphor into both antecedents. However, this path was not pursued as
it was believed to carry a potential biasing effect on the subjects’ judgement. Since,
as studies have shown (Yang et al., 1999, 2001; Alonso-Ovalle et al., 2002; Sorace
and Filiaci, 2006), in some cases both antecedents are as likely to be possible resolu-
tions of the pronoun, perhaps when forced to rate both resolutions participants might
tend to stick to their choice and not acknowledge the possibility of the other resolu-
tion being possible as well, and sometimes even to the same extent. Therefore, the
resolution of the anaphor was taken as another experimental condition, and the stimuli
contained a second sentence resolving the anaphor. Each sentence appeared twice, the
two versions differing in the second sentence which resolves the anaphor (once with
the subject antecedent and once with the object one).
Fillers
The filler stimuli were similar to the test items in some respects. They too were con-
structed of two sentences, the first introducing two potential same-gender entities. The
second sentence in the stimuli offered some statement regarding one of the entities,
which was at times a resolution to an anaphor, though not necessarily explicitly (i.e.
not a repetition of the subordinate clause containing the pronoun as it was in the test
material). The fillers were not controlled for grammatical function of the antecedents,
nor for any other syntactic, semantic or other constraint.
4This form of spelling is the one normally used e.g. in newspapers, books and on the web, as thesedo not feature the vowel signs,nikkud.
Chapter 3. Experiment 25
3.3 Procedure
The use of overt pronouns in the discussed example (brought again below in 3.5 for
convenience) is optional in nature and suggests different degrees of grammatical ac-
ceptability and likelihood rather than strict grammatical constraints.
(3.5) Dana katva le-nina kse-hi hayta be-skotland
Dana katva3SGF to-Nina when-she was3SGF in-Scotland
’Dana wrote Nina when she was in Scotland’
This optionality allowed for by natural language calls for an experimental method
which correspondingly permits gradient judgements. Such a methodology is magni-
tude estimation, an experimental paradigm adopted from psychophysics (as proposed
by Stevens, 1975) into the linguistic judgements domain by Bard et al. (1996).
In a magnitude estimation experiment subjects assign numerical values to linguistic
stimuli in a proportional manner. The subjects are first presented with an initial stan-
dard stimulus, the modulus, to which they assign some fixed numerical value. They are
subsequently asked to elicit numerical values of the stimuli they perceive in proportion
to the number they assigned to the modulus. In other words, if a subject perceives
a certain item to be twice better than the modulus, he is required to assign it a nu-
merical value that is two times bigger than the one he assigned to the modulus. Bard
et al. (1996) showed that magnitude estimation of linguistic acceptability solves the
measurement scale problems which afflict conventional judgement techniques. The
magnitude estimation methodology provides data that makes fine distinctions in a ro-
bust enough manner which yields statistically significant results of linguistic interest.
They also showed that at the same time, magnitude estimation allows for replication
across groups of subjects.
Each subject participated in an experimental session lasting approximately 15 minutes.
First, the subjects were presented with a set of instructions regarding the experiment
they were about to take part in. They then completed a short questionnaire regard-
ing their personal details which was followed by three experimental phases: a training
phase, a practice phase and an experimental phase. The experiment was self-paced,
though response times were recorded in order to enable screening the data for sub-
jects with anomalous response times. The experiment was run over the World Wide
Web, and subjects accessed the experiment remotely, using their web browser. The
Chapter 3. Experiment 26
browser established an Internet connection to the experimental server, which was run-
ning WebExp 2 (2005), an interactive software package for administering web-based
psychological experiments. The experimental methodology of running web-based psy-
chological experimental studies has been previously used and its validity attested for
(Keller et al., 2002; Keller and Sorace, 2003).
Instructions
A set of instructions in Hebrew was presented to the subjects prior to the beginning of
the experimental phase. The instructions commenced with a short explanation regard-
ing the technical requirements needed in order for the Hebrew fonts to properly appear
in the web browser. The latest version of java run time environment (cite by link) had
to be downloaded, alongside a configuration file. Once the subjects completed these
two operations they were able to view the experiment correctly.
Following a statement of the personal details that were to be gathered from the sub-
jects, the experimental instructions appeared. These first introduced the concept of
numerical magnitude estimation of line length. The notion of relative judgements was
explained and subjects were told they would first see a reference line, to which they
would have to assign some arbitrary number. They were then instructed they would
have to estimate the length of the following lines they will be presented with relative
to the first line they had seen, the reference line. Assigning numbers in a relative man-
ner was explained to be representative of how long the current line is in proportion to
the reference line. Several example lines and corresponding numerical estimates were
provided to illustrate the concept of proportionality.
Subjects were subsequently told that linguistic acceptability could be judged in the
same way as line length. The concept of linguistic acceptability was not defined; in-
stead, examples of acceptable and unacceptable sentences were provided, together with
possible numerical estimates to them. Subjects were made clear that they could use any
range of positive numbers for their judgments, including decimals. It was stressed that
there was no upper or lower limit to the numbers that could be used (except for them
being positive). Subjects were urged to use a wide range of numbers and to make
as fine grained distinctions as possible, distinguishing many degrees of acceptability.
It was also emphasized that there were no correct answers, and that subjects should
base their judgments on first impressions, not spending long to think about any one
sentence.
Chapter 3. Experiment 27
Personal details questionnaire
After the instructions were given, a short personal details questionnaire was admin-
istered. The questionnaire included name, email address, age, sex, handedness, and
academic subject or occupation. Handedness was defined as ”the hand you prefer to
use for writing”. The subjects were also requested to provide their email address for
future reference, if needed.
Training phase
The training phase was meant to familiarise subjects with the concept of numeric mag-
nitude estimation using line lengths. Items were presented as horizontal lines, centered
in the window of the subjects web browsers. After viewing an item, subjects had to
provide a numerical judgment via the computer keyboard. After they pressed Return,
the current item disappeared and the next item was displayed. There was no possibil-
ity of revisiting previous items or changing responses once Return had been pressed.
No time limit was set for either the item presentation or for the response. Subjects
first judged the modulus item, and then all the items in the training set. The modulus
was the same for all subjects, and it remained on the screen all the time to facilitate
comparison. Items were presented in random order, with the same fixed randomisation
being presented to all subjects.
Practice phase
This phase allowed subjects to practice magnitude estimation of linguistic acceptabil-
ity. Presentation and response procedure was similar to the one used in the training
phase, with linguistic stimuli being displayed instead of lines. The first sentence of the
stimulus was presented first, after which a question was presented -’How likely is the
following sentence?’The second sentence of the stimulus was then presented, offering
a resolution to the anaphor. Each subject judged the whole set of practice items. As in
the training phase, subjects first judged the modulus item, and then all the items in the
practice set. The modulus was the same for all subjects, and it remained on the screen
for the entire duration of the experiment in order to facilitate comparison. Items were
presented in random order, with the same fixed randomisation for all subjects.
Experimental phase
Presentation and response procedures in the experimental phase were the same as in the
practice phase. Twelve test sets were used: each test set contained two lexicalisations
Chapter 3. Experiment 28
for each of the twelve cells in the design, resulting in twenty-four stimuli. Lexicalisa-
tions were assigned to test sets using a Latin square. As in the practice phase, subjects
first judged the modulus item, which was the same for all subjects and remained on the
screen all the time. Then they were presented with forty-eight test items: twenty-four
experimental items and twenty-four fillers. Items were presented in random order, with
a fixed randomization of each of the twelve test sets for all subjects. The randomisation
of the test set was such that it did not include more than two experimental items in a
row. Each subject was randomly assigned one of the test sets.
3.4 Results
The magnitude estimation data were first normalised on a per subject basis by dividing
each of the judgements by the modulus given by the subject to the reference item.
The normalisation of all numerical judgements was carried out in order to transform
the scales of the different subjects into one common scale. Then, the results were log
transformed in order to receive the geometric means of the normalised judgments. The
use of geometric means ensures that the data are normally distributed and is standard
practice for magnitude estimation data (Bard et al., 1996; Keller and Sorace, 2003).
An overall 3×2×2 ANOVA, Pronoun type (null, overt, demonstrative)× Antecedent
(subject, object)× Anaphor type (anaphor, logophor), was conducted, once with the
participants as the random variable and once with the items as the random variable.
A significant main effect for the pronoun type was found by subjects (F1(2;84) =
6.257, p− .003), and by items (F2(2;46) = 4.408, p = .018). The more interesting
interaction effect found to be significant both by subjects as well as by items is the in-
teraction between pronoun and antecedent (F1(2;84) = 48.875, p< .0005,F2(2;46) =
62.035, p< .0005), as graphed by the (normalised, log-transformed) mean judgements
in Figure 3.1 below.
A significant effect was also found for the interaction between direction and antecedent
(F1(1;) = 32.785, p< .0005,F2(1;23) = 17.833, p< .0005), as graphed in Figure 3.2.
It suggested that anaphors prefer object antecedents, while logophors prefer subject
ones.
Although a significant effect was not found for the interaction between pronoun and
direction, examining the 3-way interaction for the relation between direction and an-
Chapter 3. Experiment 29
Figure 3.1: Pronoun - Antecedent interaction. Judgements for pronouns in the experi-
ment by antecedent.
tecedent with respect to the pronouns reveals more interesting insights, as graphed in
Figures 3.3, 3.4 and 3.5 below. The three-way interaction was found to be significant
(F1(2;84) = 10.984, p < .0005,F2(2;46) = 3.445, p = .040).
Indeed, a post-hoc Tukey test which was conducted for the three-way interaction be-
tween pronoun, antecedent and direction (p < .01 by subjects5) revealed further in-
sights into the nature of the interaction between the pronouns and their antecedents.
The results show that the interaction between the pronoun and its antecedent is depen-
dent upon the direction of the anaphor, as is clearly depicted by Figures 3.3, 3.4 and
3.5.
The Tukey test results of differences in mean judgements presented in Table 3.2 re-
veal significant interactions for the different pronouns between subject and object an-
tecedents depending on the anaphor direction, for all cases but for the overt pronoun
in forward anaphora. In other words, all pronouns have a marked preference for their
antecedent depending on the anaphor direction, apart from the overt pronoun in the
5The Tukey test of the by item ANOVA revealed insignificant results, but for two cases. Thedemonstrative pronouns in forward anaphors significantly preferred object antecedents over subjectones (p < .01). The null pronouns in backward anaphors significantly preferred subject antecedentsover objects (p < .01)
Chapter 3. Experiment 30
Figure 3.2: Antecedent - Direction interaction. Judgements for anaphor direction by
antecedent.
forward anaphor condition. We notice, however, that the overt pronoun does have a
significant bias in the logophor condition, and it is only in the anaphor case that it is
equally likely to attract both types of antecedents. Table 3.3 presents the significant
biases of the different pronouns depending on directionality.
Pronoun Anaphor (subject-object) Logophor (subject-object)
Null 0.2400** 0.4146**
Overt 0.0335 0.2292**
Demonstrative 0.9190** 0.3359**
Table 3.2: Tukey table results for difference in acceptability of antecedents by pronouns
and anaphor direction. ** significant at p = 0.01
A subject antecedent was found to be significantly more acceptable than an object one
in the both null pronouns (regardless of whether it is an anaphor or a logophor) as
well as overt pronouns in logophors. No preference as to an antecedent was revealed
solely for the overt pronoun in the anaphor case. An object antecedent was found to be
significantly more acceptable than a subject one with the use of demonstratives (both
anaphor and logophor).
Furthermore, although the preference of demonstrative pronouns for an object an-
Chapter 3. Experiment 31
Figure 3.3: Antecedent - Direction interaction for null pronouns. Judgements for null
pronouns’ anaphor direction by antecedent.
Pronoun Anaphor Logophor
Null subject subject
Overt no preference subject
Demonstrative object object
Table 3.3: Antecedent preference by pronoun and anaphor direction
tecedent was found to be significant in both anaphors and logophors, the Tukey test
revealed that the demonstrative preferred anaphors significantly more than logophors
with both types of antecedents, and so the preference for its object antecedent was
significantly more so in the anaphor case. While the null pronouns showed no such
significant preference for the direction of the anaphor, the overt pronoun did show
a significant bias for forward anaphors when an object antecedent was used. These
results are shown in Table 3.4.
Chapter 3. Experiment 32
Figure 3.4: Antecedent - Direction interaction for overt pronouns. Judgements for overt
pronouns’ anaphor direction by antecedent.
Pronoun Subject (anaphor-logophor) Object (anaphor - logophor)
Null 0.0612 0.1134
Overt 0.0035 0.1992**
Demonstrative 0.3386** 0.2445**
Table 3.4: Tukey table results for difference in acceptability of anaphor direction by
pronoun and antecedent. ** significant at p = 0.01
Chapter 3. Experiment 33
Figure 3.5: Antecedent - Direction interaction for demonstratives. Judgements for
demonstrative pronouns’ anaphor direction by antecedent.
Chapter 4
Discussion
The results presented in the previous section suggest that some pronouns are indeed
biased toward their antecedent, as hypothesised.
Demonstrative pronouns
The demonstratives revealed a significant preference in favour of object antecedents,
as predicted and in keeping with Kaiser and Trueswell (2004, 2005) and Bosch et al.
(2003, 2006). The effect was found to be significantly larger for anaphors than lo-
gophors, which is compliant with the use of demonstratives in a logophoric manner
not being intuitively natural and in fact rather odd. Furthermore, subject antecedents
were also found to be significantly preferred in anaphors over logophors. It seems
that the logophors received lower acceptability ratings in general when used with the
demonstrative pronouns and this might have induced even the significantly less ac-
ceptable subject antecedent to be found to be significantly more acceptable in forward
anaphora.
This observation, of the lower acceptability of demonstratives in logophoric sentences,
might also shed some light on the nature of the overall significant result of the interac-
tion between direction and antecedent. As reported, anaphors were found to prefer ob-
ject antecedents while logophors prefer subject ones. This might be partly explained by
the fact that demonstratives significantly prefer anaphors, and are significantly inclined
to take object antecedents. Perhaps the strong unacceptability of the demonstrative lo-
gophor, coupled with the strong acceptability of demonstrative anaphors with an object
antecedent, can partly account for the significance of the overall direction-antecedent
interaction finding.
34
Chapter 4. Discussion 35
Null pronouns
The null pronoun, as predicted and in keeping with Carminati’s (2002) findings, showed
a strong preference toward the subject antecedent, in both anaphor and logophor con-
ditions. The former finding demonstrates that the evidence Sorace and Filiaci (2006)
show by which Italian null pronouns have no marked preference in forward anaphora
does not hold for Hebrew null pronouns. Rather, it offers supporting evidence to
Carminati’s hypothesis of a universal null pronoun preference. However, as will be
discussed below, the results regarding the overt pronoun shed a different light on these
findings.
Overt pronouns
The overt pronoun revealed a more complex bias, which was dependent on the anaphor
direction. In logophors, an effect similar to the one the null pronoun demonstrated was
found - a preference toward subject antecedents. This finding is relatively surprising,
especially with respect to Carminati’s (2002) findings and the findings by Yang et al.
(1999, 2001). By contrast, in the anaphor (forward anaphora) case, the overt pro-
noun revealed no marked preference. This correlates to some extent with Sorace and
Filiaci’s (2006) findings, whereby in Italian the overt pronoun has no preference be-
tween subject and object antecedents in anaphors, but has a marked preference only in
the logophor case. Importantly though, Sorace and Filiaci found that the Italian overt
pronoun has a significant preference toward extralinguistic antecedents, followed by
objectantecedents in logophors. This does not correlate with what was found for the
Hebrew overt pronoun which is biased towardsubjectantecedents in logophors. How-
ever, since an extralinguistic referent was not one of the optional antecedents in our
experimental design, this might have biased the results in an unwanted and unpre-
dictable manner. Perhaps, if offered, subjects would have opted for an extralinguistic
antecedent rather than both subject and object ones.
That being said, the current results obtained concerning the overt pronoun behaviour
suggests that it behaves more similarly to the Hebrew null pronoun than to the Italian
overt pronoun. This sheds a different light on the null pronoun’s behaviour, and specif-
ically with respect to Carminati’s suggestion whereby there exists a division in labour
between the null and the overt pronouns. A possible reasoning for that is that perhaps
the special pro-drop pattern in Hebrew renders the overt pronoun to behave more along
the lines of non pro-drop languages’ pronouns. If this is indeed the case, the question
Chapter 4. Discussion 36
remains what role is there for a double set of pronominal forms in a partial pro-drop
language.
General Discussion
The hypothesis this study aimed at investigating is that different pronominal forms have
different referential properties and are thus biased toward different antecedents. While
the results seem to have indicated that different pronouns indeed tend to take different
syntactic elements as their antecedent, it cannot strictly determine that it is indeed a
property of the pronouns themselves that renders them biased toward a specific an-
tecedent, or whether it is some other factor that co-occurs with certain constructions
that provides this bias. Indeed, it seems that other factors might also affect the reso-
lution of the pronoun, namely the direction of the anaphor seems to play a significant
role.
Moreover, our results cannot confirm what it is about the antecedents that renders them
preferable for a specific type of pronoun. Is it the grammatical function they occupy?
Is it their linear position? A wide variety of factors and qualities may come into mind,
among them are the grammatical function and the recency ones mentioned, as well as
topicality (as put forward by Bosch et al., 2003), theme-hood and others. In light of
these it appears that Carminati’s commitment to a specific, theoretically-binding, syn-
tactic construction might be a rather narrow approach. Although the syntactic position
has been shown to affect the antecedent bias, disregarding the other factors which seem
to be of importance, renders Carminati’s theory quite restricted.
In this light, other existing theories which are more suitable in accounting for different
influencing factors come into mind for the resolution of anaphors. Importantly, our
results are in keeping not only with Carminati’s purely syntactic-oriented model, but
also with a wide variety of theories such as accessibility theory (Ariel, 1990), center-
ing theory (Grosz et al., 1995) and other salience-scale theories (Gundel et al., 1993).
With regard to these, our results seem to suggest that Hebrew’s special pro-drop pat-
tern renders its overt pronoun to be situated in a special position on the accessibility
scale, as it does not adhere to the same logic that overt pronouns of ’normal’ pro-drop
languages such as Italian follow. The Hebrew overt pronoun seems to be closer in
behaviour to the null pronoun, which is arguably closer to the behaviour of overt pro-
nouns from non pro-drop languages. Therefore, the mystery of the existence of both
forms of pronominals in Hebrew remains to be investigated.
Chapter 5
Conclusions
5.1 Summary
Following an examination of the phenomenon of pronoun dropping, section 2 offered
an extensive review of the cross-linguistic research that has been conducted in the res-
olution of null, overt, as well as demonstrative pronouns. It also depicted the pronom-
inals behaviour in Hebrew, laying grounds for the experiment. Then, section 3 de-
scribed the experiment conducted, paying particular attention to the conceptual design
decisions made. Section 4 continues by highlighting the interesting findings and of-
fering a critical analysis of the results. It further offers interesting unresolved research
questions this study has shed light upon.
This study has provided evidence for different biases of Hebrew null, overt and demon-
strative pronouns in choosing their antecedents. While the null and demonstrative
pronouns’ biases seem to adhere to a wider cross-linguistic trend, the overt pronoun
was found to exhibit a rather distinct behaviour. The cross-linguistic differences were
pointed out to possibly stem from Hebrew’s special pro-drop pattern rendering its overt
pronoun’s role quite different from the one traditionally associated with it. It was also
suggested, following Carminati, that there might exist a scale of typologically different
overt pronouns, on which Hebrew takes a special place.
The study has further pointed to the role that the direction of the anaphor plays in the
antecedent preference. This suggests that different factors might also affect the res-
olution of the anaphor, apart from the pronominal form. Since the direction of the
anaphor is argued to be closely related to processing issues, further experiments using
37
Chapter 5. Conclusions 38
processing-oriented methodologies such as reading-time might shed further light on
the resolution of the different pronouns. It may also provide an opportunity to test other
factors potentially affecting the resolution, whether syntactic, semantic, discourse-
related or other.
5.2 Future Work
As was clearly evident from our results, the direction of the anaphor affects the bias
of the pronoun for its antecedent. This was definitely the case with the overt pronoun,
where the pronoun was biased toward a subject antecedent in anaphors, but no such
preference was detected in the logophor case. Furthermore, the demonstrative pronoun
revealed a significant preference toward anaphors, showing that pronouns are indeed
sensitive to the direction of the anaphor. This might therefore suggest that other factors
may also play a role in the pronouns’ preferences toward their antecedents, if at least in
the overt pronoun case. Not only does the direction of the anaphor change the order in
which the pronoun and the potential antecedents are encountered by the processor, but
it also manipulates theirrecency, which has been proved to affect anaphora resolution
(Gernsbacher et al., 1989; Bever and Townsend, 1978). Therefore, it might be inter-
esting to examine other factors that manipulate the recency factor as well. One such
factor is word order. As Hebrew is a language that allows relatively free word order,
a potential follow-up experiment could manipulate the order in which the potential
antecedents are encountered by using, for instance, OSV rather than SVO construc-
tions. The interaction between the different, potentially conflicting, factors that affect
anaphora resolution in general has been extensively discussed in the literature (e.g.
see Gernsbacher et al., 1989, among others), and it might prove even more so with
respect to, and in comparison to, different pronouns, namely null and demonstratives
pronouns.
The previously mentioned finding regarding the significant preference of anaphors
over logophors when using demonstrative pronouns suggests that the direction of the
anaphor also involves acceptability rather than grammaticality in their use and calls for
further investigation on its own right, either within a certain language (e.g. Hebrew),
but also cross-linguistically. Different factors might be proven to have an effect on
the acceptability of the direction of the anaphor, as was demonstrated for the pronoun
type in Hebrew. The same line of research may also be pursued with word order in
Chapter 5. Conclusions 39
general. As Hebrew does allow relatively free word order but not entirely, and as the
less-canonical word orders (other than SVO) are less frequently used, it might prove
interesting to study the factors that affect the distribution and acceptability of these
word-orders.
Another experimental path that was previously suggested and should be pursued in
later studies is adding an extralinguistic antecedent as a possible resolution of the
anaphor. Not only has it been proved to hold in other languages such as Italian (Sorace
and Filiaci, 2006), but it might also prove to alter the results we have obtained in this
study. Indeed, not only is the overt pronoun susceptible to an extralinguistic interpreta-
tion, but the demonstrative as well. This is especially true in the less natural logophor
case, which intuitively seems to be a lot more acceptable when viewed as referring to
an extralinguistic entity.
In addition to the different research questions our data has raised that were put forward
above, it also points to a few further methodologies which may now come into use.
As this experiment confirmed the existence of a bias in the referential properties of
the different classes of Hebrew pronouns, it may serve as a baseline for other method-
ologically motivated studies into the same question. For instance, it may be followed
with a reading-time and/or eye tracking experiments which will point as to whether the
mechanism that is responsible for the pronouns’ referential bias is a processing one, or
whether it is not processing that is involved. Such methodologies enable tracking the
subject’s online linguistic behaviour and can thus offer evidence in support of process-
ing accounts which judgement experiments such as the one conducted in this study
fail to provide. Using these methods, one would make the prediction that the readings
where the preferred antecedent resolves the anaphor would result in shorter reading
times in comparison to those with the disfavoured antecedents. Eye-tracking stud-
ies may also potentially detect saccades back to the preferred antecedents in forward
anaphora.
A further line of research to be pursued is to make use of the Hebrew treebank (Simaan
et al., 2001) for a corpus study in order to study the distribution of different phenom-
ena. The use of corpus linguistics in order to account for and investigate different lin-
guistic phenomena has become common practice in computational linguistics. Bosch
et al. (2003, 2006), for instance, used antecedent preference in anaphora resolution
in accounting for the distribution of German demonstrative pronouns. In addition to
their experimental data, they also present data gathered regarding the distribution of
Chapter 5. Conclusions 40
the demonstrative pronouns in written corpora. As they successfully show, corpus
based findings may provide reinforcing evidence that complements the judgement and
processing data. Accordingly, data from a Hebrew corpus might provide added, com-
plementing, insight into the ’real-world’ use of the different pronouns. Frequencies of
the resolution of the pronouns to their antecedents can point to the statistically prefer-
able antecedent in written text and will thus indicate the referential bias of the pro-
nouns. Moreover, the results from the conducted experiments can later be evaluated
against the findings from the corpus, and different factors (such as word order) may be
further investigated.
Appendix A
Experimental Material
Each of the experimental stimuli appeared in one of the twelve experimental condi-
tions, demonstrated by (A.1 a-l) below:
(A.1) a. Me’ir silem le-david lifney se-φ-tas le-xul.
Me’ir paid 3SGM to-David before that-φ-flew 3SGM to-abroad.
Me’ir tas le-xul.
Me’ir flew 3SGM to-abroad.
b. Me’ir silem le-david lifney se-hu tas le-xul.
Me’ir paid 3SGM to-David before that-he flew3SGM to-abroad.
Me’ir tas le-xul.
Me’ir flew 3SGM to-abroad.
c. Me’ir silem le-david lifney se-ze tas
Me’ir paid 3SGM to-David before that-heDEM flew 3SGM
le-xul.
to-abroad.
Me’ir tas le-xul.
Me’ir flew 3SGM to-abroad.
’Me’ir paid David before he flew abroad.’
’Me’ir flew abroad.’
d. Me’ir silem le-david lifney se-φ-tas le-xul.
Me’ir paid 3SGM to-David before that-φ-flew 3SGM to-abroad.
41
Appendix A. Experimental Material 42
David tas le-xul.
David flew 3SGM to-abroad.
e. Me’ir silem le-david lifney se-hu tas le-xul.
Me’ir paid 3SGM to-David before that-he flew3SGM to-abroad.
David tas le-xul.
David flew 3SGM to-abroad.
f. Me’ir silem le-david lifney se-ze tas
Me’ir paid 3SGM to-David before that-heDEM flew 3SGM
le-xul.
to-abroad.
David tas le-xul.
David flew 3SGM to-abroad.
’Me’ir paid David before he flew abroad.’
’David flew abroad.’
g. lifney se-φ-tas le-xul Me’ir silem le-david.
before that-φ-flew 3SGM to-abroad Me’ir paid3SGM to-David.
Me’ir tas le-xul.
Me’ir flew 3SGM to-abroad.
h. lifney se-hu tas le-xul Me’ir silem le-david.
before that-he flew3SGM to-abroad Me’ir paid3SGM to-David.
Me’ir tas le-xul.
Me’ir flew 3SGM to-abroad.
i. lifney se-ze tas le-xul Me’ir silem
before that-heDEM flew 3SGM to-abroad Me’ir paid3SGM
le-david.
to-David.
Me’ir tas le-xul.
Me’ir flew 3SGM to-abroad.
’Before he flew abroad Me’ir paid David.’
’Me’ir flew abroad.’
j. lifney se-φ-tas le-xul Me’ir silem le-david.
before that-φ-flew 3SGM to-abroad Me’ir paid3SGM to-David.
Appendix A. Experimental Material 43
David tas le-xul.
David flew 3SGM to-abroad.
k. lifney se-hu tas le-xul Me’ir silem le-david.
before that-he flew3SGM to-abroad Me’ir paid3SGM to-David.
David tas le-xul.
David flew 3SGM to-abroad.
l. lifney se-ze tas le-xul Me’ir silem
before that-heDEM flew 3SGM to-abroad Me’ir paid3SGM
le-david.
to-David.
David tas le-xul.
David flew 3SGM to-abroad.
’Before he flew abroad Me’ir paid David.’
’David flew abroad.’
The remaining 23 lexicalisations will appear below only in the form of (A.1 a), i.e. in
the null pronoun, forward anaphor, subject antecedent condition. The other 11 forms
of each of them may be deducted from (A.1 b-l).
(A.2) Yael katva le-dana kse-φ sahata be-arcot habrit.
Yael katva3SGF to-Dana when-φ stayed3SGF in-states united.
Yael sahata be-arcot habrit.
Yael stayed3SGF in-states united.
’Yael wrote Dana while she stayed in the United States’.
’Yael stayed in the United States.’
(A.3) Mixal nofefa le-raxel kse-φ-xacta et ha-kvis.
Michal waved3SGF to-Rachel when-φ-crossed3SGF ACC the-road.
Mixal xacta et ha-kvis.
Michal crossed3SGF ACC the-road.
’Michal waved Rachel while she was crossing the road.’
’Michal crossed the road’
Appendix A. Experimental Material 44
(A.4) Miryam yi’aca le-xana axrey se-φ-nitka et
Miriam advised3SGF to-Hannah after that-φ-hung-up3SGF ACC
ha-telefon.
the-phone.
Miryam nitka et ha-telefon.
Miriam hung-up3SGF ACC the-phone.
’Michal gave Hannah advice after she had hung up the phone.’
’Miriam hung up the phone.’
(A.5) Mixael xagag im cvi kse-φ-histaxrer
Michael celebrated3SGM with Tzvi when-φ-got 3SGM released
me-ha-cava.
from-the-army.
Mixael histaxrer me-ha-cava.
Michael got3SGM released from-the-army.
’Michael celebrated with Tzvi when he got released from the army.’
’Michael got released from the army.’
(A.6) Uri tilpen le-mose lifney se-φ-azav et
Uri phoned3SGM to-Moshe before that-φ-left 3SGM ACC
Yerusalayim.
Jerusalem.
Uri azav et Yerusalayim.
Uri left 3SGM ACC Jerusalem.
’Uri phoned Moshe before he left Jerusalem.’
’Uri left Jerusalem.’
(A.7) Dalya nifreda mi-slomit kse-φ-avra le-tel aviv.
Dalia parted3SGF from-Shlomit when-φ-moved3SGF to-Tel Aviv.
Dalya avra le-tel aviv.
Dalia moved3SGF to-Tel Aviv.
’Dalia parted from Shlomit when she had moved to Tel Aviv.’
’Dalia moved to Tel Aviv.’
(A.8) Yardena hit’alma me-rut kse-φ-hegi’a la-hofa’a.
Yardena ignored3SGF from-Ruth when-φ-arrived3SGF to-the-concert.
Yardena hegi’a la-hofa’a.
Yardena arrived3SGF to-the-concert.
Appendix A. Experimental Material 45
’Yardena ignored Ruth when she had arrived to the concert.’
’Yardena arrived to the concert.’
(A.9) Ilana hiskima im anat kse-φ-sava me-ha-yeshiva.
Ilana agreed3SGF with Anat when-φ-returned3SGF from-the-meeting.
Ilana sava me-ha-yeshiva.
Ilana returned3SGF from-the-meeting.
’Ilana agreed with Anat when she had returned from the meeting.’
’Ilana returned from the meeting.’
(A.10) Yoni ana le-ido lifney se-φ-xazar la-basis.
Yoni answered3SGM to-Ido before that-φ-returned3SGM to-the-base.
Yoni xazar la-basis.
Yoni returned3SGM to-the-base.
’Yoni answered Ido before he had returned to the base.’
’Yoni returned to the base.’
(A.11) Yoram hitkaser le-dani’el axrey se-φ-hitpater me-ha-avoda.
Yoram called3SGM to-Daniel after that-φ-resigned3SGM from-the-job.
Yoram hitpater me-ha-avoda.
Yoram resigned3SGM from-the-job.
’Yoram called Daniel after he had resigned from his job.’
’Yoram resigned from his job.’
(A.12) Smu’el azar le-eli kse-φ-nixnas la-xeder.
Shmuel helped3SGM to-eli when-φ-entered3SGM to-the-room.
Smuel nixnas la-xeder.
Shmuel entered3SGM to-the-room.
’Shmuel helped Eli when he had entered the room.’
’Shmuel entered the room.’
(A.13) Tamar nafsa im sira axrey se-φ-hitgarsa.
Tamar went3SGF on vacation with Shira after that-φ-divorced3SGF.
Tamar hitgarsa.
Tamar divorced3SGF.
’Tamar went on vacation with Shira after she had divorced.’
’Tamar divorced.’
Appendix A. Experimental Material 46
(A.14) Na’ama hictarfa le-talya axrey se-φ-higi’a be-ixur.
Na’ama joined3SGF to-talya after that-φ-arrived3SGF in-delay.
Na’ama higi’a be’ixur.
Na’ama arrived3SGF in-delay.
’Na’ama joined Talya after she had arrived late.’
’Na’ama arrived late.’
(A.15) Sani xiyxa le-no’omi kse-φ-nifreda le-salom.
Shani smiled3SGF to-Naomi when-φ-parted3SGF to-goodbye.
Sani nifreda le-salom.
Shani parted3SGF to-goodbye.
’Shani smiled at Naomi when she said goodbye.’
’Shani said goodbye.’
(A.16) Yuval siye’a le-rami axrey se-φ-sav habayta.
Yuval assisted3SGM to-Rami after that-φ-returned3SGM to-home.
Yuval sav habayta.
Yuval returned3SGM to-home.
’Yuval assisted Rami after he had returned home.’
’Yuval returned home.’
(A.17) Alon hoda le-omri lifney se-φ-yaca me-ha-misrad.
Alon thanked3SGM to-Omri before that-φ-left 3SGM from-the-office.
Alon yaca me-ha-misrad.
Alon left 3SGM from-the-office.
’Aon thanked Omri before he had left the office.’
’Alon left the office.’
(A.18) Itay kara le-ro’i kse-φ-lavas et ha-me’il.
Itay called3SGM to-Roy when-φ-wore 3SGF ACC the-coat.
Itay lavas et ha-me’il.
Itay wore3SGM ACC the-coat.
’Itay called Roy when he was putting the coat on.’
’Itay was putting the coat on.’
Appendix A. Experimental Material 47
(A.19) Hila sixaka im sheli lifney se-φ-savra et ha-buba.
Hila played3SGF with Shelly before that-φ-broke3SGF ACC the-doll.
Hila savra et ha-buba.
Hila broke3SGF ACC the-doll.
’Hila played with Shelly before she had broken the doll.’
’Hila broke the doll.’
(A.20) Hadas avda im avigayil lifney se-φ-hitmanta
Hadas worked3SGF with Abigail before that-φ-was-appointed3SGF
le-menahelet bxira.
to-manager executive.
Hadas hitmanta le-menahelet bxira.
Hadas was-appointed3SGF to-manager executive.
’Hadas worked with Abigail before she was appointed executive manager.’
’Hadas was appointed executive manager.’
(A.21) Le’a naska le-efrat beteremφ-halxa haxuca.
Leah kissed3SGF to-Efrat before φ-went 3SGF to-out.
Le’a halxa haxuca.
Leah went3SGF to-out.
’Leah kissed Efrat before she went outside.’
’Leah went outside.’
(A.22) Doron hebit be-yonatan kse-φ-hexel ledaber.
Doron looked3SGM at-Jonathan when-φ-started3SGM speaking.
Doron hexel ledaber.
Doron started3SGM speaking.
’Doron looked at Jonathan when he had started speaking.’
’Doron started speaking.’
(A.23) Nadav hitpayes im Eytan kse-φ-hitxaten.
Nadav made-peace3SGM with Eitan when-φ-got-married3SGM.
Nadav hitxaten.
Nadav got-married3SGM.
’Nadav made peace with Eitan when he got married.’
’Nadav got married.’
Appendix A. Experimental Material 48
(A.24) Yisra’el viter le-asaf lifney se-φ-araz et
Yisrael let-go3SGM to-Assaf before that-φ-packed3SGM ACC
ha-mizvadot.
the-suitcases.
Yisra’el araz et ha-mizvadot.
Yisrael packed3SGM ACC the-suitcases.
’Yisrael let Assaf go before he packed the suitcases.’
’Israel packed the suitcases.’
Appendix B
Descriptive Statistics
Pronoun Antecedent Direction Mean SD SE
null subject anaphor 0.0717 0.2671 0.0407
null subject logophor 0.1329 0.2924 0.0446
null object anaphor -0.1683 0.3392 0.0517
null object logophor -0.2817 0.4342 0.0662
overt subject anaphor -0.0298 0.3342 0.0510
overt subject logophor -0.0333 0.4167 0.0636
overt object anaphor -0.0633 0.2756 0.0420
overt object logophor -0.2625 0.3610 0.0551
demonstrative subject anaphor -0.6752 0.6978 0.1064
demonstrative subject logophor-0.3366 0.5377 0.0820
demonstrative object anaphor 0.2438 0.3168 0.0483
demonstrative object logophor-0.0007 0.3772 0.0575
Table B.1: Descriptive statistics for the experiment
49
Bibliography
Alonso-Ovalle, L., Fernandez-Solera, S., Frazier, L., and Clifton, C. (2002). Null
versus overt pronouns and the topic-focus articulation in spanish.Journal of Italian
Linguistics, 14:151–169.
Ariel, M. (1990). Accessing Noun-phrase Antecedents. Routledge, London.
Arnold, J., Eisenband, J., Brown-Schmidt, S., and Trueswell, J. (2000). The rapid
use of gender information: Evidence of the time course of pronoun resolution from
eye-tracking.Cognition, 76:B13B26.
Bard, E. G., Robertson, D., and Sorace, A. (1996). Magnitude estimation of linguistic
acceptability.Language, 72:32–68.
Bever, T. and Townsend, D. (1978). Perceptual mechanisms and formal properties
of main and subordinate clauses. In Walker, R. and Wales, R., editors,Studies in
Sentence Processing. Hillsdale, New Jersey: Laurence Erlbaum.
Bosch, P., Katz, G., and Umbach, C. (2006). The non-subject bias of german demon-
strative pronouns. In Schwartz-Friesel, M. and Consten, M., editors,Anaphors in
Texts(to appear).
Bosch, P., Rozario, T., and Zhao, Y. (2003). Demonstrative pronouns and personal
pronouns. german der vs. er. InProceedings of the EACL2003, Budapest. Workshop
on The Computational Treatment of Anaphora.
Carminati, M. N. (2002).processing of Italian subject pronouns. PhD thesis, Univer-
sity of Massachusetts Amherst.
Chomsky, N. (1981).Lectures on Government and Binding. Foris, Dordrecht.
Cole, P. (1987). Null objects in universal grammar.Linguistic Inquiry, 18:597–612.
50
Bibliography 51
Cooreman, A. and Sanford, A. J. (1996). Focus and syntactic subordination in dis-
course. Technical Report HCRC/RP-79, Human Communication Research Centre
(HCRC).
Cowart, W. and Cairns, H. (1987). Evidence for an anaphoric mechanism within syn-
tactic processing. some reference relations defy semantic and pragmatic constraints.
Memory and Cognition, 15:318–331.
Diaconescu, R. and Goodluck, H. (2004). The pronoun attraction effect for d(iscourse)-
linked phrases: Evidence from speakers of a null subject language.Journal of Psy-
cholinguistic Research, 33:303–319.
Frazier, L. and Clifton, C. J. (2002). Processing ’d-linked’ phrases.Journal of Psy-
cholinguistic Research, 31:633–659.
Gernsbacher, M. and Hargreaves, D. (1988). Accessing sentence participants: The
advantage of first mention.Journal of Memory and Language, 27:699–717.
Gernsbacher, M., Hargreaves, D., and Beeman, M. (1989). Building and access-
ing clausal representaions: The advantage of first mention versus the advantage of
clause recency.Journal of Memory and Language, 28:735–755.
Greenberg, J. (1966). Some universals of grammar with particular reference to the
order of meaningful elements. In Greenberg, J., editor,Universals of Language,
pages 73–113. MIT Press, Cambridge, Massachusetts and London, England.
Grosz, B. J., Joshi, A. K., and Weinstein, S. (1995). Centering: A framework for
modeling the local coherence of discourse.Computational Linguistics, 21(2):203–
225.
Gundel, J., Hedberg, N., and Zacharski, R. (1993). Cognitive status and the form of
referring expressions in discourse.Language, 69:274–307.
Hacohen, A. and Schaeffer, J. (2005). Subject realization in early hebrew/english
bilingual acquisition: The role of crosslinguistic influence. In Deen, K., Nomura, J.,
Schulz, B., and Schwartz, B., editors,The Proceedings of the Inaugural Conference
on Generative Approaches to Language Acquisition. North America, Honolulu, HI,
volume 4 ofUniversity of Connecticut Occasional Papers in Linguistics, pages 113–
124.
Hobbs, J. R. (1978). Resolving pronoun references.Lingua, 44:311–338.
Bibliography 52
Huang, Y. (1995). On null subjects and null objects in generative grammar.Linguistics,
page 1081.
Huang, Y. (2000).Anaphora. A Cross-linguistic Study. Oxford University Press, Ox-
ford.
Jaeggli, O. and Safir, K. (1989). The null subject parameter and parametric theory. In
Jaeggli, O. and Safir, K., editors,The Null Subject Parameter. Dordrecht, Kluwer.
Kaiser, E. and Trueswell, J. (2004). The referential properties of dutch pronouns and
demonstratives: Is salience enough? In Weisgerber, M., editor,Proceedings of Sinn
und Bedeutung 8, University of Konstanz linguistics working papers, Germany.
Kaiser, E. and Trueswell, J. (2005). Investigating the interpretation of pronouns and
demonstratives in finnish: Going beyond salience. In Gibson, E. and Pearlmutter,
N., editors,The Processing and Acquisition of Reference. MIT Press, Cambridge,
Massachusetts.
Kamp, H. (1981). A theory of truth and semantic representation. In Groenendijk, J.,
Janssen, T., and Stokhof, M., editors,Formal Methods in the Study of Language,
pages 277–322. Mathematisch Centrum, Amsterdam.
Kazanina, N. (2005). The acquisition and processing of backward anaphora. Unpub-
lished PhD Dissertation, University of Maryland, College Park.
Kazanina, N., Lieberman, M., Phillips, C., and Yoshida, M. (2005). Constraints on
coreference in the online processing of backwards anaphora. Poster presented at
the 18th Annual CUNY Sentence Processing Conference, University of Arizona,
Tuscon.
Keller, F. and Lapata, M. (1998). Object drop and discourse accessibility. In Shahin,
K., Blake, S., and Kim, E.-S., editors,Proceedings of the seventeenth West Coast
conference on Formal Linguistics, pages 362–374. Stanford, CA: CSL1.
Keller, F., Lapata, M., and Ourioupina, O. (2002). Using the web to overcome data
sparseness. InProceedings of EMNLP-02, page 230237.
Keller, F. and Sorace, A. (2003). Gradient auxiliary selection and impersonal passiviza-
tion in German: An experimental investigation.Journal of Linguistics, 39(1):57–
108.
Bibliography 53
Lappin, S. and Leass, H. J. (1994). An algorithm for pronominal anaphora resolution.
Computational Linguistics, 20(4):535–561.
Simaan, K., Itai, A., Winter, Y., Altman, A., and Nativ, N. (2001). Building a tree-bank
of modern hebrew text.Traitment Automatique des Langues, 42. Special Issue on
Natural Language Processing and Corpus Linguistics.
Sorace, A. and Filiaci, F. (2006). Anaphora resolution in near-native speakers of italian.
Second Language Research (to Appear), 22,3:1–30.
Stevens, S. (1975).Psychophysics: Introduction to its Perceptual, Neural, and Social
Prospects. John Wiley, New York.
van Gompel, R. and Liversedge, S. (2003). The influence of morphological infor-
mation on cataphoric pronoun assignment.Journal of Experimental Psychology:
Learning, Memory and Cognition, 29:128–139.
Webexp2 (2005). Webexp2 experimental software, version 1.01.
http://www.webexp.info/.
Yang, C., Gordon, P., Hendrick, R., and Wu, J. (1999). Comprehension of referring
expressions in chinese.Language and Cognitive Processes, 14:715–743.
Yang, C., Gordon, P., Hendrick, R., Wu, J., and Chou, T. (2001). The processing of
coreference for reduced expressions in discourse integration.Journal of Psycholin-
guistic Research, 30:21–35.