EDITORS: Schwarz-Friesel, Monika; Consten, Manfred; Knees, Mareile Hillevi
TITLE: Anaphors in Text
SUBTITLE: Cognitive, formal and applied approaches to anaphoric reference
PUBLISHER: John Benjamins
Alessio S. Frenda, Centre for Language and Communication Studies, Trinity
The fifteen contributions to this book describe anaphoric reference as a
coherence-building device. There are three sections, corresponding to three
different approaches to anaphora: cognitive, text and discourse linguistics
(section 1), syntax and semantics of anaphors (section 2) and neurolinguistic
studies (section 3). The exposition draws on examples from the Germanic language
family, mainly German and English; Danish and French are also represented - all
in all, not quite the ''several Indo-European languages'' announced by the back cover.
In the course of this summary, I make a few evaluative remarks about specific
chapters where warranted.
I. The first chapter, ''Indirect anaphora in text: a cognitive account''
(pp.3-20), by Monika Schwarz-Friesel, opens the first section and provides a
fine-grained description of different types of indirect anaphora (IA) based on
naturally occurring data from German. IAs are definite noun phrases (NPs)
without explicit antecedents, whose definiteness is nonetheless perfectly
acceptable by virtue of their indirect reference to some discourse entity
(''anchor''). Schwarz-Friesel also defines a set of ''constraints on
associability'', i.e. conditions under which an IA may not be licensed. The
examples on which the argument is based are taken from German, and although I am
not a native speaker of the language, I was stricken by the oddity of some
alleged constraints: for instance, the author denies that _das Rot/die Röte_
('the (color) red') may be accepted as a valid IA for _viel Blut_ ('a lot of
blood') in the following example:
(I.1) Es floss viel Blut in diesem Kampf. *Das Rot/*Die Röte stach den Kriegern
in die Augen.
'A lot of blood was shed in the battle. *The red thrust the warriors in the
eyes.' (p.14, the author's translation)
However, _das Weiß_ may be accepted as an IA for the antecedent _Schnee_ ('snow'):
(I.2) Als wir erwachten, lag alles voller Schnee. Das Weiß blendete unsere Augen.
'When we got up, there was snow all over. The white was simply overwhelming.'
(p.15; the author's translation)
According to the author, _das Weiß_ is a valid IA for _Schnee_, while _das
Rot/die Röte_ are not for _viel Blut_, because in (I.2) ''we activate a
text-world model in which the spatial representation of the white snow is
dominant'' (p.15), which, it is assumed, we don't do in (I.1). Nevertheless, the
idea of redness can provide a valid IA for the mention of blood given the
appropriate context, as in (I.3):
(I.3) ... stach sie sich mit der Nadel in den Finger und es fielen drei Tropfen
Blut in den Schnee. Und weil das Rote im weißen Schnee so schön aussah ...
'... she pricked her finger with her needle and three drops of blood fell into
the snow. And as the red on the white snow looked so beautiful ...' (p.15)
The ''specific text-world model'' of (I.3) is that of a ''novel or fairy tale''
(p.15) and the claim is that this justifies the IA that was unacceptable in (I.1).
We are left wondering, however, whether (I.1) would be acceptable assuming the
context of a novel or fairy tale, but also whether the choice of the lexical
item (_das Rote_ as opposed to _das Rot/die Röte_) is also partially responsible
for the contrasting judgments.
II. In ''Indirect pronominal anaphora in English and French'' (pp.21-36), Francis
Cornish presents the findings of two experiments that had already appeared in
Cornish et al. (2005) to address the question whether or not unaccented
third-person pronouns constitute valid indirect anaphors, a question to which
conflicting answers have been offered by different scholars. The author attempts
to reconcile the two views by drawing a distinction between instances in which
the intended referent is central within the current discourse representation and
instances in which it is peripheral, thus accounting for the conflicting
judgments on (II.1) (acceptable) and (II.2) (unacceptable):
(II.1) Woman: ''Why didn't you write to me?''
Man: ''I did..., started to, but I always tore 'EM up.'' (p.23)
(II.2) Harry drove to London. *IT broke down half-way (p.26)
_(th)em_ in (II.1) refers to an intended referent (letters) which is central to
the current discourse representation, and is thus acceptable, while _it_ in
(II.2) cannot be used with the intended meaning (the car) because the intended
referent is not central to the discourse representation. Central antecedents are
defined as those which correspond to nuclear arguments of the predicate, while
peripheral antecedents correspond to non-nuclear arguments.
This appears problematic to me: if LETTER is a possible nuclear argument of
WRITE, why is CAR not a possible nuclear argument of DRIVE? The author argues
that the intended referent in (II.2) is evoked as ''the means by which the
situation described was (partially) realised'' (p.27), although some kind of
vehicle would certainly seem to be as central to the representation of (II.2) as
some kind of document is central to the representation of (II.1). For instance,
one would question the acceptability of (II.3):
(II.3) a. Harry wrote to Harriet. She never read ?them/?it (= the letters/the
b. Harry wrote to Harriet. ?They/?It (= the letters/the letter) never reached her.
We are thus left with the impression that factors other than argument centrality
might have a non-marginal role in this.
III. ''Lexical anaphors in Danish and French'' (pp.37-47), by Lita Lundquist,
discusses so-called ''unfaithful anaphors'' and their distribution in French and
Danish. Unfaithful anaphors are characterized by their referential ambiguity -
they are NPs which may be interpreted either anaphorically or as introducing a
new referent, as _Dominique de Villepin_ and _Le ministre de l'Intérieur_ in
(III.1) DOMINIQUE DE VILLEPIN a présenté, mercredi, le programme de stabilité
des finances publiques françaises. LE MINISTRE DE L'INTERIEUR a fixé jeudi un
objectif de baisse de 3% de la délinquance en 2006.
'Wednesday DOMINIQUE DE VILLEPIN presented the stability programme for the
French public finances. Thursday, THE MINISTER OF INTERNAL AFFAIRS set a goal of
reducing delinquency with [sic] 3% in 2006.' (pp.39f, the author's translation)
This type of anaphor is claimed to be far more common in Romance than in
Germanic languages, where lexical repetitions or pronominal anaphors seem to be
preferred (Korzen and Lundquist 2003). This would seem to reflect the different
lexicalization strategies preferred by the two language groups: Romance
languages, it is claimed, tend to ''lexicalise more semantic features in nouns
than Germanic languages'' (pp.40f and Herslund and Baron 2003); semantically
richer nouns correlate with semantically richer anaphors, semantically poorer
nouns with semantically poorer anaphors.
I would have liked further elaboration of the concept that unfaithful anaphora
constitutes a form of grammaticalization of rhetorical relations, introduced as
a final suggestion. As a final remark, it should be noted that the usage
described in Lundquist's chapter is based on written data only, mostly of a
journalistic nature, and generalizations should therefore be taken cautiously.
IV. ''Referential collaboration with computers: do we treat computer addressees
like humans?'' (by A. Maes, P. Marcelis and F. Verheyen, pp.49-68) introduces the
issue of human-computer interaction. Referential expressions are chosen based on
assumptions, made and continuously updated by the speaker, about the hearer's
background knowledge and can thus be taken as an indicator of human attitude
towards their computer interlocutors. The authors present the findings of an
experiment in which simulated human-computer interaction (instruction-giving)
was compared to computer-mediated human-human interaction, suggesting that, in
situations in which human instructors do not receive any feedback from their
addressees, they will tend to make a greater referential effort when they assume
a human addressee than when they assume a computer addressee.
V. In ''Reflexivity and temporality in discourse deixis'' (pp.69-80) Friedrich
Lenz addresses the debated distinction between deixis and anaphora, focusing on
discourse deixis, which he sees as a particular instance of time deixis rather
than as an independent deictic dimension; its function is to refer to temporal
entities within the ''communicative context of the participants'' (p.68). In this
sense, discourse deixis is defined as ''reflexive language use'' (p.70), i.e.
portions of discourse referring to other portions of the same discourse.
Lenz's insightful treatment is conceptually very dense and it would have
probably benefited from a few more concrete examples.
VI. ''The function of complex anaphors in texts: evidence from corpus studies and
ontological considerations'' (by M. Consten, M. Knees and M. Schwarz-Friesel,
pp.81-102) presents a corpus-based analysis of complex anaphors in German.
Complex anaphors are defined as nominal expressions with a propositionally
structured antecedent and a conceptually complex referent (a second- or
third-level entity in terms of Lyons 1977, 1989). The analysis looks at both
their function (constructing textual coherence) and constraints restricting the
way they can be used to change the ontological status of their referent (through
a process of abstraction). A resolution model for complex anaphors is also
provided which combines Discourse Representation Theory (DRT) and Text World
Models (see Schwarz 2000, 2001).
To my mind, a problem seems to arise with the notion of anchor in example (34),
reproduced below as (VI.1):
(VI.1) Auf mein Drängen hin entschied sich Bethe, nach Los Alamos zu kommen ...
Währenddessen rief Oppenheimer an und lud uns zu sich nach Princeton ein. Ich
sagte zu Bethe: ''Nach DIESEM GESPRÄCH werden Sie nicht mehr kommen.''
'Bethe decided to come to Los Alamos at my urging ... Meanwhile, Oppenheimer
called and invited us to stay with him at Princeton. I told Bethe: ''After THIS
INTERLOCUTION you won't be willing to come any more.''' (p.98, the authors'
The authors argue that _diesem Gespräch_ is an indirect complex anaphor based on
the INVITATION scheme (p.99), which licenses its interpretation as roughly ''the
conversation occurred between Bethe, Oppenheimer and the narrator as they were
in Princeton at Oppenheimer’s invite''. The problem is that _diesem Gespräch_ is
contained in a direct-speech segment, which reproduces something the narrator
said to Bethe in the past. In this context, the NP was deictic, as it referred
to the conversation he and Bethe had previously had with Oppenheimer. In this
sense, the NP is not anaphoric, as it would have been if the narrator had for
instance said ''I thought that after THAT CONVERSATION Bethe wouldn't be willing
to come any more''. Of course, in the here-and-now context of the narration, the
reader is able to understand what _diesem Gespräch_ refers to, inferentially,
through the INVITATION scheme as suggested by the authors. But it should be
recognized that the status of the NP in question, as it appears in this example,
is ambiguous in this respect and deserving of further investigation in its own
VII. In ''Metaphorical anaphors: a phenomenon of the semantics-pragmatics
interface'' (pp.103-119), Helge Skirl describes a very interesting and
little-studied phenomenon, which occurs when two semantically incongruous NPs
are intended as coreferential and the anaphoric resolution is possible only
through metaphor. The analysis is conducted in the theoretical framework of
Schwarz's (2000) Text World Models.
VIII. Klaus von Heusinger's ''Accessibility and definite noun phrases''
(pp.123-144) opens Section 2 of the book. Here, the author combines DRT and
Centering Theory approaches to deal with the way definite NPs can modify the
accessibility status of a discourse item. Contrary to the view that definite NPs
are licensed by current accessibility structure, by which they are ''passively''
bestowed with definite status, the author argues that definite NPs themselves
have the potential to change the accessibility structure through what he terms
''salience spreading''- a process whereby a discourse entity is promoted, upon its
mention, to the status of most salient discourse entity not only within its own
ontological category but also within relevant supersets, thus licensing
referential chains in which generic terms refer to specific entities.
IX. In ''The non-subject bias of German demonstrative pronouns'' (pp.145-164), P.
Bosch, G. Katz and C. Umbach consider two German anaphoric series, the
demonstrative _der (die, das)_ and the personal pronoun _er (sie, es)_, and the
complementary distribution effect resulting from what appears to be the
preference of the former for syntactical object antecedents and of the latter
for syntactical subject ones. The investigation is conducted through corpus
analysis and two psycholinguistic studies. The results seem to confirm the
authors' initial intuition and are coherent with analogous results obtained by
other researchers for Dutch, Finnish and Italian.
Only subject and object are considered: it would be interesting to repeat this
study including other non-subject arguments (e.g. dative), as well as NP
X. ''Anaphoric properties of German right dislocation'' (by Maria
Averintseva-Klisch, pp.165-182) considers German _Rechtsversetzung_ ('right
dislocation') and argues against the traditional view, attributed to Altmann
(1981), that its function is simply to ''repair'' a potentially ambiguous
anaphoric reference. Starting from the observation that this structure is
commonly found, and not only in German, even with perfectly unambiguous
anaphors, the authors argues that there are in fact two distinct structures,
''right dislocation proper'' (RD) and ''afterthought'' (AT), which are prosodically,
syntactically and semantically differentiated. While AT is employed to resolve
referential ambiguity and is not syntactically integrated in the sentence to its
left, RD is used to establish a discourse-old referent as the topic of the
following discourse segment.
I should remark that, since this chapter crucially relies on examples drawn from
German, the latter ought to always be translated and not just glossed. This is
not always the case, even when the meaning of a sentence is not so easily
guessed from the glosses, as in example (14.a), presented here as (X.1):
(X.1) ... Die schmecken aber leider nicht, die Äpfel.
They taste but unfortunately not, the apples (p.172)
A reader who is not familiar with the German construction won't be able to
easily recover the intended meaning ('The apples don't taste good,
unfortunately') from the glosses.
XI. In ''Antecedents of diverse types: an investigation of the syntactic and
semantic relationships in a wh-relative construction'' (pp.183-206), Anke Holler
deals with the German wh-relative clause, arguing against the ''ordinary
assumption'' that it ''always relates to a sentential or propositional antecedent''
(p.184). Instead, she aims to show that the syntactic and semantic relationships
that wh-relatives entertain with their antecedents should be kept distinct:
while the syntactic antecedent is always propositional, the semantic antecedent
can be an abstract object of various types. The analysis is conducted in the
theoretical framework of DRT.
One point could be raised apropos of the syntactic status the author envisages
for the wh-clause when she concludes that ''a _wh_-relative clause is not
integrated into the syntactic structure of the matrix clause, but ties in with
the complete matrix clause'' (p.196): we are not told how exactly the distinction
between ''integrating'' and ''tying in'' should be defined, a point of no small
importance given the space dedicated to the syntactic difference between RD and AT.
XII. In ''Corpus-based and machine learning approaches to anaphora resolution: a
critical assessment'' (pp.207-222), Michael Strube provides an extensive review
of current research in the implementation of anaphora resolution in the field of
natural language processing. The author explains the nature of the problem
addressed and presents state-of-the-art approaches to its resolution based on
corpora and machine learning.
XIII. The third and last section of the book, on the neurolinguistic approach to
the study of anaphora, is opened by E. C. Ferstl and F. Th. Siebörger's chapter
on ''Neuroimaging studies of coherence processes'' (pp.225-240). (It should be
remarked well in advance that it is beyond the writer's capabilities to attempt
a critical evaluation of the technical issues contained in this section,
especially as far as neuroanatomy and related experimental settings are concerned.)
Ferstl and Siebörger introduce the neuropsychological approach to coherence
building and present the results of two neuroimaging studies conducted by the
authors using functional magnetic resonance imaging methods, aimed to obtain
spatial information about the brain activity (functional neuroanatomy) in order
to establish ''qualitative dissociations between cognitive processes'' (pp.227).
XIV. In ''Reference assignment in the absence of sufficient semantic content''
(pp.241-258), Petra Burkhardt discusses experimental evidence regarding the
establishment of referential dependencies for pronominal entities using
event-related brain potentials (ERPs), an analytic method that permits the
investigation of the neurophysiological correlates of some cognitive processes.
The experiment conducted by the author focuses on a class of pronouns (termed
''logophors'') which are morphologically but not semantically reflexive, i.e.
which do not express coreferentiality between different arguments of the same
predicate like proper reflexive pronouns (''coargument reflexives''). The
experimental evidence suggests that while coreference with coargument reflexives
is resolved at the syntactic level at a lower processing cost, coreference with
logophors and non-reflexive pronouns is established at the discourse level and
at a higher cost. From a theoretical point of view, these results provide
neurophysiological evidence for considering syntax and discourse as two
different language processing levels.
XV. In ''Resolving complex anaphors: evidence from online comprehension''
(pp.259-277), K. Marx, I. Bornkessel-Schlesewsky and M. Schlesewsky outline the
functioning of the ERP method, which they use to describe the neurophysiological
correlate of the complexation process posited by theories of complex anaphora
such as Schwarz (2000), Schwarz-Friesel, Consten, and Marx (2004). (Also see
Consten, Knees, and Schwarz-Friesel, this vol.) The hypothesis the authors set
about to investigate (for German) is that complex anaphor resolution requires a
greater cognitive effort than the resolution of direct NP anaphors (intended as
anaphors with a NP antecedent), which the data seem to confirm in a
statistically significant way only when the antecedent of the NP anaphor is
syntactically a subject.
Besides specific points which I raised within the respective chapters, there is
something to be said as regards the publishing standard of this book, which
suffers as a whole from what seems to have been a rather poor proofreading process.
Minor oversights include banal typos like the use of 'German inverted commas',
or easily recognizable German abbreviations like AKK for ACC(usative), FOK for
FOC(us) (pp.167, 177); but we also find abbreviations which would only be
familiar to a German speaker, like _Mio_ for ''million'' (p.152), as well as the
odd German word that was ''lost in translation'' and is found stranded in the
English text, like _anapher_ for _anaphor_ (pp.129, 131 et passim).
In other instances, however, the oversights threaten to subvert the logical
consequentiality of a passage. In the following example, the negative term shown
in brackets is missing from the text and its absence yields a self-contradictory
''It is generally assumed that ... definite noun phrases are interpreted as
static terms, i.e. as terms that [do not/cannot] change the accessibility. In
contrast ... it was shown that definite noun phrases dynamically change the
accessibility structure'' (p.142).
An approximate count of miscellaneous oversights (ranging from banal typos to
significant inaccuracies) in the first 200 pages yields some 118 occurrences,
which would suggest a statistical distribution of at least one occurrence per
page - a rate that is certainly surprising to find in the output of a leading
On the whole, the book addresses a wide range of issues in a number of specific
disciplines (corpus linguistics, neurolinguistics, syntactic and semantic
theories). It is therefore a commendable attempt to give a fairly comprehensive
take on a single linguistic phenomenon, anaphora. It will therefore be welcomed
by all those interested in an overall view that takes into account its many facets.
On account of its multidisciplinary approach this book is not destined for a
specialist reader, yet it is somewhat surprising to find a number of key
concepts and theory-specific constructs often taken for granted. For instance,
not only is familiarity with the tenets of Discourse Representation Theory
largely assumed, but the theory itself is introduced for the first time solely
by the acronym, DRT. Similarly, when linguistic examples are crucial to the
point being made, it is often the case that no gloss or translation is presented
to the reader. In the case of the neurolinguistics section, perhaps one of the
most interesting of the book, specialist knowledge is clearly required in order
to evaluate the results presented. There would be little to object to if a given
article were found by the interested reader in a sector-specific journal, but a
greater overall readability would probably be expected from an edited book like
Altmann, H. 1981. _Formen der Herausstellung im Deutschen: Rechtsversetzung,
Linksversetzung, Freies Thema und verwandte Konstruktionen_. Tübingen: Niemeyer.
Cornish, Francis, Alan Garnham, H. Wind Cowles, Marion Fossard, and Virginie
Andre. 2005. Indirect anaphora in English and French: A cross-linguistic study
of pronoun resolution. _Journal of Memory and Language_ 52, no. 3:363-376.
Herslund, Michael, and Irène Baron. 2003. Language as World View: endocentric
and exocentric representations of reality. In _Language and culture_, Ed. Irène
Baron, 168, Forlaget Samfundslitteratur.
Korzen, Iørn, and Lita Lundquist, eds. 2003. _Sprogtypologi og oversættelse:
Endocentriske og exocentriske sprog_. HHK: Samfundslitteratur.
Lyons, J. 1977. _Semantics_. Cambridge: Cambridge University Press.
Lyons, J. 1989. Semantic ascent: A neglected aspect of syntactic typology. In
_Essays on grammatical theory and universal grammar_, ed. D. Arnold. Oxford:
Schwarz, M. 2000. _Indirekte Anaphern in Texten. Studien zur domänen-gebunden
Referenz und Kohärenz im Deutschen_. Tübingen: Niemeyer.
Schwarz, M. 2001. Establishing coherence in text: Conceptual continuity and
text-world models. _Logos and Language_ II, no. 1:15-24.
Schwarz-Friesel, Monika, Manfred Consten, and K. Marx. 2004. Semantische und
konzeptuelle Prozesse bei der Verarbeitung von Komplex-Anaphern. In
_Flexibilität und Stabilität_, ed. I. Pohl. Frankfurt am Main: Peter Lang.
ABOUT THE REVIEWER
Alessio Frenda is a PhD candidate at the Centre for Language and Communication
Studies of Trinity College, Dublin. He is interested in historical linguistics
and morphosemantics and is currently investigating the evolution of grammatical
gender in Irish and Welsh from a cognitive and functional point of view.