Demonstrative Pronouns and Personal Pronouns. German der vs. er.

Peter Bosch Tom Rozario Yufan Zhao

Inst. of Cognitive Science Inst. of Cognitive Science Inst. of Cognitive Science
University of Osnabrück University of Osnabrück University of Osnabrück
49069 Osnabrück, Germany 49069 Osnabrück, Germany 49069 Osnabrück, Germany
pbosch@uos.de trozario@uos.de yzhao@uos.de

In (1) the demonstrative der can only refer to

Abstract Peter, while the personal pronoun er can only
refer to Paul. In (2) both the personal pronoun sie
This paper deals with the distribution and and the demonstrative die can be used equally
reference of the German demonstrative well to refer to the woman.
pronoun der and its case and gender vari- There is, to the best of our knowledge, no
ants, in comparison with the behaviour of computational nor any explicit formal linguistic
personal pronouns. After introducing the account of German demonstratives. – One obsta-
relevant facts about both pronoun types, cle is that the same forms are also used as defi-
we develop our Complementarity Hy- nite articles and relative pronouns, which makes
pothesis, which claims that demonstra- it hard at times to single out demonstrative uses
tives differ from personal pronouns automatically. Some part of speech taggers, even
mainly in that they prefer less salient ref- very good ones, sytematically go wrong on de-
erents. The hypothesis is tested against a monstratives1. A further problem for a systematic
newspaper corpus and is found to be fun- account is that demonstratives, as we already saw
damentally correct. The paper concludes in (2), overlap in their distribution as well as in
with the discussion of the empirical re- their referential options with the personal pro-
sults and their effect for future work. nouns and the distinction occasionally looks like
no more than a stylistic one.
The main reason that no computational ac-
1 Introduction count has been attempted, however, is probably
that there seems no urgent need for it: Demon-
Unlike English, German has demonstrative pro- strative pronouns of the der paradigm – different
nouns that are inflected for number, gender, and from other German demonstrative pronouns like
case and can refer also to persons as well as other dieser, jener, derjenige, derselbe, etc. – are com-
individuals, very much like personal pronouns. paratively rare. In a large German newspaper
Here are some examples: corpus 94.8% of all occurrences of der were
found to be occurrences of the definite article, a
(1) Paul wollte mit Peter laufen gehen. Aber further 4.8% were occurrences of the relative
{er/der} war erkältet. pronoun, and a mere 0.4% were occurrences of
[Paul wanted to go running with Peter. But the demonstrative pronoun2. Because of its clear
{he/[DEMONSTR]} had a cold.] deictic potential, similar to English this or that,
one might think that the demonstrative would be
(2) Paul sah eine Frau hereinkommen. {Sie/Die} more frequent in spoken language. But an unbi-
trug einen schwarzen Mantel. ased comparison would require written and spo-
[Paul saw a woman enter. She was wearing a ken corpora of the same discourse type, which
black coat.]
1
An example is the tree tagger by Helmut Schmid (1994).
2
These figures are based on a 41 Mio word corpus of Frankfurter
Rundschau, a German national daily newspaper.
we have not been able to acquire. So we have to 2 Demonstrative and personal pronouns
leave this point to speculation3.
So, why would one want a computational At a first glance there is considerable overlap in
identification of demonstratives, and why would the distribution of demonstrative and personal
one want a better understanding of their refer- pronouns. In many contexts, as already in exam-
ence? On the one hand, there is of course the mo- ple (2), either form seems acceptable and seman-
tivation from applications, similar as for all other tic, pragmatic, or even stylistic differences are
work on reference resolution: A treatment of hard to pinpoint. This is also true in the follow-
demonstratives is clearly needed, e.g., for ma- ing cases:
chine translation. Nearly no German demonstra-
tive can be adequately translated into English, or (3) Am Dienstag, 16. Juni, können dann Falken-
most other languages, without reference resolu- steiner ihren Sondermüll von 9 bis 10 Uhr zu
tion. dem Wagen bringen. {Er/Der} parkt auf
On the other hand, independent of applica- dem Parkplatz beim Bürgerhaus.
tions, we believe that an improved understanding [On Tuesday, 16 June, the Falkensteiners can
of demonstratives may contribute both to a better take their hazardous waste to the van be-
understanding of processes of pronominal refer- tween 9 and 10 a.m. It is parked in the park-
ence and to theoretical accounts of information ing lot at the Civic Centre.]
structure and the dynamics of discourse.
The interesting point about German demon- (4) Zunächst waren die Mietforderungen des
stratives in this connection is that they allow for Investors (38 Mark pro Quadratmeter) zu
a fairly direct comparison of pronominal ana- hoch. {Er/Der} reduzierte sie dann auf unter
phoric and demonstrative reference in a way this 30 Mark.
comparison is not possible for other languages. [At first the rent demands of the investor (38
In English we could look at differences between marks per square meter) were too high. Then
it, this, and that. But here the range of potential he reduced them to under 30 Marks.]
referents brings in a number of additional com-
plications (cf. Passonneau 1989, Webber 1991, The impression that there is an area of overlap
Eckert & Strube 2000) that are avoided when we where the function of the demonstrative pronoun
compare personal and demonstrative pronouns in is indistinguishable from the anaphoric personal
German. pronoun and where the two forms could be sub-
This paper takes a few modest steps towards stituted for each other without a clear grammati-
an account of the distribution and reference of cal or semantic difference was confirmed in two
demonstratives, but what we have to offer is still psycholinguistic experiments carried out by our
work in progress. We have not yet looked at all project group (see Cummins et al. (in prepara-
forms of the demonstrative. What we have tion)). Self-paced reading as well as native
worked out so far is an hypothesis about the dis- speakers' acceptability preferences yielded no
tinction between different uses of the demonstra- significant difference.
tive and personal pronouns, and we have put this This functional overlap of personal and de-
hypothesis to the test with a small 350,000 word monstrative pronouns may well be an artefact of
corpus of written German. underspecified data though. If, as we shall argue,
the difference between the two types of pronouns
is a matter of information structure, then the dis-
tinction may not show up when not enough in-
formation structure is visible – most clearly
when we consider odd sentences in isolation, but
3
The Verbmobil corpus of spoken German (http:// verbmo-
often also in sentence pairs.
bil.dfki.de/) has a ratio of personal pronouns to demonstratives of Still, there are occurrences of the two pro-
1:4, while the Negra corpus of written German yields about 8:1. noun types for which the difference is quite clear
However, this comparison remains inconclusive: the Verbmobil
corpus consists of dialogues negotiating the time and place of ap- as already in (1), repeated here:
pointments, while the Negra corpus is a regular daily newspaper
corpus with obviously vastly different discourse structure.
(1) Paul wollte mit Peter laufen gehen. Aber tary preference of the unstressed counterpart." –
{er/der} war erkältet. where the referents of both stressed and un-
[Paul wanted to go running with Peter. But stressed pronouns come from the current focus of
{he/[DEMONSTR]} had a cold.] attention of the discourse, the Forward-looking
Centre of the preceding utterance, in terms of
The er can only be interpreted with reference to Centering Theory (Grosz e.a. 1995).
Paul, while der must refer to Peter – a difference But the distinction between German personal
that is hard to emulate in an English translation. and demonstrative pronouns is not the same. The
But also a difference in grammaticality can be demonstrative in (1) does not translate into Eng-
found. In (5) only the demonstrative is allowed lish as a stressed personal pronoun: While der in
while the personal pronoun would be ungram- (1) can only refer to Peter, HE in (1') is just
matical: oddly ambiguous:

(5) Die Gelder sollen nicht aus dem Etat des (1') Paul wanted to go running with Peter. But
Umweltministeriums, sondern aus {dem/ HE had a cold.
*ihm} des Entwicklungsministeriums
fließen. Kameyama (1999) is probably right characteris-
[The money should not come out of the ing the function of English stressed personal
budget of the Ministry of Environment, but pronouns in terms of an Alternative Semantics
from [the budget] of the Ministry of Devel- approach to focus (Rooth 1992), and hence in
opment.] terms of contrast, and an analysis of stressed
German and English personal pronouns in terms
The way that demonstratives differ from personal of contrast and markedness was also put forward
pronouns may in some respects be reminiscent of in Bosch (1988). The difference between per-
differences found in English between stressed sonal pronouns and demonstratives in German,
and unstressed personal pronouns, as docu- however, cannot be captured along these lines:
mented by Akmajian and Jackendoff (1970). Cf.4 Even though complementarity plays a role, the
category of contrast is here absent.
(6) When {he/HE} came home, John was tired. The basic regularity that we find in the be-
haviour of demonstratives is that they
where the unstressed anaphoric pronoun clearly i. choose from several antecedents (that obey
refers to John and the stressed HE cannot be so the relevant number and gender constraints
interpreted. In addition we have cases like those as well as subcategorisation constraints of
brought up by Lakoff (1971): the relevant predicate) the one whose referent
is preferentially not at the top salience rank
(7) John called Fred a Republican. Then he in- (not a topic); Cf. (1), (3), (4).
SULted him. ii. pick up a unique antecedent (that obeys the
relevant number and gender constraints as
(7') John called Fred a Republican. Then HE in- well as subcategorisation constraints of the
sulted HIM. relevant predicate), even though it stands for
the most salient referent (the topic). They
Also here the referential options of the pronouns may also accommodate a generic referent
change when the pronouns are accentuated. Ob- that obeys the subcategorisation constraints
servations like these suggest something like a of the relevant predicate (cf. (8) below). In
complementarity between stressed and un- this condition their referential options are
stressed personal pronouns in English as was identical to those of referential anaphoric
claimed in Kameyama's (1999) Complementary personal pronouns, except for a certain stylis-
Preference Hypothesis: "A focused pronoun [i.e. tic markedness; cf. (2).
stressed personal pronoun] takes the complemen- iii. when accompanied by a pointing gesture
they refer to a unique referent that was not
4
salient before.
Stressed syllables are marked typographically by capitals.
Condition (iii), even though without direct rele- 3 Identifying demonstrative uses of der
vance for written language, probably provides
the clearest intuitive understanding for the func- The first problem is the identification of those
tion of demonstratives: They highlight a previ- occurrences of the relevant forms that are actu-
ously non-salient referent, and in a sense they ally used as demonstratives. The forms of the
presuppose the non-salience of their referents. – demonstrative der in German are largely identi-
If their reference defaults to salient discourse cal with those of the relative pronoun and the
entities, as under condition (ii), when no other definite article. So, how do we identify
suitable discourse referents are available, we ob- demonstrative uses?
serve an effect of stylistic markedness: The ref-
erent is represented, as it were, as something put 3.1 Demonstratives and definite articles
at a distance. A similar effect is seen in English
Demonstrative uses of der and its morphological
when, e.g., a salient female person is referred to
variants are formally identical with definite arti-
as "that woman" or the current head of govern-
cle uses – except for the genitive and the dative
ment as "that prime minister".
plural forms (cf. tables 1 and 2). Their gender
For the purposes of this paper we summarize
and number is determined either by the semantic
this hypothesis about the function of demonstra-
classification (animacy, sex) of the intended ref-
tives in the following form:
erent or by the gender and number of a (virtual or
actual) antecedent (cf. Bosch 1987:72-73).
Complementarity Hypothesis
Definite determiner occurrences always start
Anaphoric personal pronouns prefer referents
an NP, i.e. they are followed by an (attributive)
that are established as discourse topics, while
adjective, adverb, numeral, or noun, etc., while
demonstratives prefer non-topical referents.
demonstrative pronouns are themselves of the
category NP. Some unclarity may arise though in
Clearly, this is a working hypothesis that needs
cases that could look like cases of elided nouns,
further elaboration, in particular with regard to
as in (5) above, repeated here.
the notions of salience and topic. But it is suffi-
cient for a first validation with respect to corpus
data, provided we can make it operational.
In line with most research in the area, guided sing plur
in particular by the Keenan and Comrie's (1977) masc fem neut m/f/n
Accessibility Hierarchy, Givòn's (1984:139) nom der die das die
Topicality Hierarchy, and Centering Theory gen dessen deren dessen deren/
(Grosz e.a. 1995), we decided that noun phrases derer5
that occur in the nominative are thereby ceteris dat dem der dem denen
paribus very likely to establish their referent as a acc den die das die
topic for the following sentence, and that noun Table 1: Forms of the demonstrative pronoun
phrases that are not in the nominative are less
likely to establish topics for the following sen-
tence. We disregard in this study all other pa- sing plur
rameters that may influence salience or topic- masc fem neut m/f/n
hood, such as definiteness, referent type, voice,
nom der die das die
or embeddedness.
Before we can check our hypothesis on corpus gen des der des der
data, however, we have to deal with a number of dat dem der dem den
smaller problems related to the identification of acc den die das die
demonstratives. Table 2: Forms of the definite article

5
derer only when immediately followed by a relative clause, or as
a relative pronoun when preceded by a preposition. But cf.
Bærentzen (1995) for details.
(5) Die Gelder sollen nicht aus dem Etat des the NEGRA 6 corpus of written German, with
Umweltministeriums, sondern aus dem des regard to their frequency and with regard to the
Entwicklungsministeriums fließen. classification of their antecedents.
[The money should not come out of the
budget of the Ministry of Environment, but 4.1 Frequency results
from [the budget] of the Ministry of Devel-
opment.] The first observation concerns relative fre-
quency. Demonstratives are here much rarer than
One might want to argue that dem is really ellip- personal pronouns: We counted 1436 instances
tical for dem Etat; and indeed the insertion of of personal pronouns and only 180 demonstra-
Etat would lead to a fully acceptable paraphrase. tives. Perhaps this proportion cannot straightfor-
Still, dem can here only be analysed as a demon- wardly be generalized but is typical in any case
strative pronoun. The evidence comes from a for written discourses and those that typically
plural variant: show topic continuity, as newspaper articles do.

(5') Die Gelder sollen nicht aus den Mitteln des 4.2 Preferred antecedents
Umweltministeriums, sondern aus {denen/
*den} des Entwicklungsministeriums fließen. The main result we can report concerns anaphor-
[The money should not come out of the antecedent relations for demonstratives and per-
means of the Ministry of Environment, but sonal pronouns. The relevant figures for the
out of those of the Ministry of Develop- NEGRA corpus are given in tables 3 and 4, using
ment.] the break-up discussed in Section 2 above7.
Here the article form den is ungrammatical and
only the demonstrative denen is acceptable. Our antecedent position
claim that dem in the entirely parallel (5) is also a
preceding sentence
demonstrative is thus born out. earlier in same
non- discourse sentence
nominative
nominative
3.2 Demonstrative and relative pronouns
14.5% 46.7% 8.9% 30.0%
The distinction between occurrences of the Table 3 Position of antecedents of demonstratives
demonstrative and occurrences of the relative
pronoun is trickier because the inventory of
forms is exactly the same. In principle there antecedent position
should be a difference in their distribution preceding sentence
earlier in same
though, because German relative clauses have non- discourse sentence
nominative
sub-clause, i.e. verb-final, word order. Unfortu- nominative
nately, there still remains an area of overlap that 48.0% 7.3% 17.7% 27.2%
is controversial on theoretic grounds (cf. Gärtner Table 4 Position of antecedents of personal pro-
2001). We have here decided to count all and nouns
only those pronouns from the paradigm as rela-
tives that are unambiguously distinguished as The possibly most striking difference between
relatives by V-final word order, plus those from personal and demonstrative pronouns is that the
syntactically ambiguous structures with a pro- antecedent of a demonstrative in over 90 % of all
noun plus a finite verb. All other occurrences are cases is found in the same or the preceding sen-
regarded as demonstratives.
6
NEGRA is a POS-tagged and syntactically annotated corpus of
4 Empirical results German of 355,000 words, a subset of the Frankfurter Rundschau
corpus, which is also available as a tree bank. See http:// www. coli.
uni-sb.de/sfb378 /negra-corpus/negra-corpus.html.
We compared the occurrences of the demonstra- 7
The counts include all singular and plural masculine and feminine
tive with those of the personal pronoun forms in forms. Neuter forms were excluded because of the additional prob-
lems they cause with respect to referent type and non-NP antece-
dents.
tence8, while personal pronouns refer to objects ferent for personal and demonstrative pronouns.
that are introduced earlier in discourse about Hence this class should be discarded for a quanti-
twice as often (17.7% vs. 8.9%). This figure on tative comparison and should be treated sepa-
its own already gives initial support to the hy- rately.
pothesis that personal pronouns more likely The other set of figures we should ignore in a
than demonstratives pick up referents that have quantitative comparison are those that have to
the status of a topic: Topics are usually main- do with sentence-internal pronouns. Here the
tained over more than one sentence. point is that personal pronouns frequently do not
Tables 3 and 4 also show that there is an function referentially when their antecedent is in
about equal proportion of occurrences for both the same sentence, while demonstrative pronouns
pronoun classes that have a sentence-internal are referential in all their occurrences. Personal
antecedent. And we can also clearly see where pronouns, next to their referential occurrences,
antecedents are preferentially located: for per- also have "bound" or "syntactic" occurrences (cf.
sonal pronouns in nominative NPs in the preced- Reinhart 1983; Bosch 1983) as in the following
ing sentence and for demonstratives in non- sentences
nominatives. This would confirm our hypothesis
regarding the complementarity of personal and (9) Nobody thought he would make it.
demonstrative pronouns.
The figures in tables 3 and 4 are not ideal for (10) Fred felt sick, when he returned.
a comparison though. For two reasons: (a) they
include demonstratives that have their antece- or their German equivalents
dents neither in the same sentence nor in the pre-
ceding sentence, and (b) they include personal (9') Niemand dachte, er würde es schaffen.
pronouns that have their antecedents in the same
sentence. – The first case poses a problem with (10') Fred fühlte sich krank, als er zurückkam.
regard to antecedents further back in discourse,
of which there are practically none. The figure The personal pronouns in these sentences (on the
almost exclusively rests on cases where actually intended interpretation) are not independently
no antecedent could not be found in the last three referential items and hence their job obviously
preceding sentences; and these are demonstra- cannot be done by (referential) demonstrative
tives that accommodate their referents. An typi- pronouns:
cal example is the following:
(9'') Niemand dachte, der würde es schaffen.
(8) Jetzt, wo hier auf einmal an jedem Feldweg
jemand Blumen verkauft, wollen die uns ver- (10'') Fred fühlte sich krank, als der zurück-
treiben. kam.
[Now, where there is suddenly someone sell-
ing flowers along each country lane, they [the The occurrences of der here cannot be inter-
authorities] want to chase us away.] preted the same way as the personal pronouns
before, but must refer to a person introduced
Although this kind of accommodating use is pos- elsewhere in discourse. We find a similar effect
sible also for personal pronoun forms, the domi- in English with stressed personal pronouns:
nant case among pronouns referring to an
antecedent "earlier in discourse" is just straight- (9''') Nobody thought HE would make it.
forward anaphora. But this means that the pa-
rameters governing the class of references to (10''') Fred felt sick, when HE returned.
antecedents "earlier in discourse" are quite dif-
Syntactic occurrences of personal pronouns
8
are in a relation of congruence or agreement with
We must add that we counted cases of accommodated referents
and occurrences for which no antecedent could be recovered among an NP that c-commands them and not in a rela-
those where the antecedent was not in the same or the preceding tion of co-reference (cf. Bosch 1983). Although
sentence.
not all personal pronouns with an explicit ante- topic of the preceding sentence, which is accom-
cedent in the same sentence are syntactic pro- panied by a certain stylistic effect. These are
nouns, a large proportion is. And since this mode cases where there is only one grammatically
of operation is unavailable for demonstratives, suitable antecedent. Thirdly, there are factors that
we should not include sentence-internal pro- would make personal pronouns prefer non-topics
nouns when we are interested in comparing op- as referents, e.g. strategies of parallel interpreta-
tions for referent choice and had better focus on tion. And, finally, although this is comparatively
the area where personal pronouns and demon- rare in newspaper corpora, we find cases in
stratives actually compete: when they have ante- which world knowledge simply dictates an inter-
cedents in the previous sentence. pretation that runs counter to what the grammar
If we discount antecedents earlier in discourse of the pronouns make us expect, i.e. badly writ-
and in the same sentence and take the uses with ten text. – All of these factors require further in-
an antecedent in the previous sentence as 100%, vestigation.
the figures look as in tables 5 and 6.
5 Conclusion and further work
antecedent in preceding sentence
From the data presented it would appear that our
nominative non-nominative Complementarity Hypothesis is well supported:
23.6% 76.4% while personal pronouns prefer topic referents,
Table 5 Classification of antecedents of demonstra- demonstrative pronouns prefer referents that are
tive pronouns when in previous sentence not topical. This raises a number of issues that
need further investigation. Among them the
question about the interaction of the case pa-
antecedent in preceding sentence rameter for salience with other parameters. This
nominative non-nominative issue is investigated in psycholinguistic experi-
ments we are now running. Further there is a
86.7% 13.2%
great need to study the intrasentential division of
Table 6 Classification of antecedents of personal labour between personal and demonstrative pro-
pronouns when in previous sentence
nouns, which involves, in particular, closer atten-
tion to the relation between relatives and
These tables show the complementary function
demonstratives. A large area of investigation, for
of personal and demonstrative pronouns clearer
which we do not currently have suitable corpora,
than tables 3 and 4. Personal pronouns have an-
is the question of how the regularity we have
tecedents in the nominative in the overwhelming
majority of cases, while demonstratives rather found for newspaper texts carry over to other
discourse types, in particular spoken discourse.
look for non-nominatives. – If our assumption is
In this context also the parameter of stress needs
right that the nominative signals topicality of the
referent of an NP, or a high rank in the atten- further research, which plays (nearly) no role in
written text, but influences reference options for
tional structure of discourse, and non-nomina-
German demonstratives just as it does for per-
tives signal non-topicality, or a lower rank in the
sonal pronouns in English and German. – An-
attentional structure, then these results have a
other, and somewhat different, line of
clear theoretical interpretation that follows our
development is the application of results from
discussion in Section 2.
this line of research to computational reference
The complementarity documented in tables 5
resolution. We believe that the contrastive inves-
and 6 is not perfect, however. What are the dis-
tigation of reference options of different lexical
turbing factors? First, we should recall that we
forms, as here demonstratives vs personal pro-
used a pretty coarse operationalization of the no-
nouns, may be particularly useful in developing
tions of salience and topic that ignores a number
further constraints also for the reference resolu-
of potentially relevant parameters, such as defi-
tion of the vastly more popular personal pro-
niteness, voice, embeddedness, or reference type.
nouns, even when reference resolution for
Then, as already mentioned above, there is the
use under which a demonstrative refers to the
demonstratives themselves is not seen as repre- Edward Keenan and Bernard Comrie. 1977.
senting a particularly urgent task. Noun phrase accessibilty and universal gram-
mar. Linguistic Inquiry, 8:63-99.
Acknowledgements George Lakoff (1971) Presupposition and rela-
tive well-formedness. In: Danny Steinberg &
The work reported here is being carried out in a Jakobovits (eds.): Semantics. CUP. Cam-
study project of the Cognitive Science Interna- bridge. 329-344.
tional Master Programme at the University of Tanya Reinhart. 1983. Anaphora and semantic
Osnabrück. Project participants are Phil Cum- interpretation. Croom Helm, London.
mins, Boris Gutbrod, Kyoung-ho Park, Tom Ro- Mats Rooth. 1992. A theory of focus interpreta-
zario, Yufan Zhao and as staff advisers Peter tion. Natural Language Semantics 1:75-116.
Bosch, Graham Katz, Carla Umbach, and Jaque- Helmut Schmid. 1994. Probabilistic Part-of-
line Griego. The authors gratefully acknowledge Speech Tagging Using Decision Trees. Pro-
the contributions of all participants and advisers. ceedings of International Conference on New
Methods in Language Processing. September
References 1994
Adrian Akmajian & Ray Jackendoff. 1970. Co- Bonnie Lynn Webber. 1991. Structure and osten-
referentiality and stress. Linguistic Inquiry, 1. sion in the interpretation of discourse deixis.
Per Bærentzen. 1995. Zum Gebrauch der Pro- Language and Cognitive Processes 6:107-
nominalformen 'deren' und 'derer' im heutigen 135.
Deutsch. Beiträge zur Geschichte der
deutschen Sprache und Literatur, 117: 299-
217.
Peter Bosch. 1983. Agreement and Anaphora.
Academic Press, London.
Peter Bosch. 1987. Pronouns under control.
Journal of Semantics, 5:65-78.
Peter Bosch. 1988. Representing and accessing
focussed referents. Language and Cognitive
Processes, 3: 207-231.
Philip Cummins et al. in preparation. German
demonstratives. Institute of Cognitive Science,
University of Osnabrück.
Miriam Eckert & Michael Strube. 2000. Dia-
logue acts, synchronizing units, and anaphora
resolution. Journal of Semantics 17:51-89
Hans-Martin Gärtner. 2001. Are there V2 rela-
tive clauses in German? Journal of Compara-
tive Germanic Linguistics, 3: 97-141.
Talmy Givòn. Syntax. A Functional-Typological
Introduction. 1984. Benjamins, Amsterdam.
Barbara Grosz, Aravind Joshi, Scott Weinstein.
1995. Centering: A framework for modeling
the local coherence of discourse. Computa-
tional Linguistics 21:67-90
Megumi Kameyama. 1999. Stressed and un-
stressed pronouns. In: Peter Bosch & Rob van
der Sandt. eds..: Focus. CUP, Cambridge.
306-321.