Sometimes two expressions in a discourse can be about the same thing in a way that makes that very fact evident to the participants. Consider, for example, 'he' and 'John' in 'John went to the store and he bought some milk'. Let us call this 'de jure' coreference. Other times, coreference is 'de facto' as with 'Mark Twain' and 'Samuel Clemens' in a sincere use of 'Mark Twain is not Samuel Clemens'. Here, agents can understand the speech without (...) knowing that the names refer to the same person. After surveying many available linguistic and pragmatic tools (intentions to corefer, presuppositions, meanings, indexing, discourse referents, binding etc.) I conclude that we must posit a new semantic primitive to account for de jure coreference. (shrink)

In mathematical languages and in predicate logic coreferential terms can be interchanged in any sentence without altering the truth value of that sentence. Replacing 3 + 5 by 12 − 4 in any formula of arithmetic will never lead from truth to falsity or from falsity to truth. But natural languages are different in this respect. While in some contexts it is always allowed to interchange coreferential terms, other contexts do not admit this. An example of the first sort of (...) context is likes bananas: for any two coreferential noun phrases A and B the sentence A likes bananas is true if and only if B likes bananas is. A context that does not allow intersubstitution of coreferents is The Ancients knew that appears at dawn. If we fill the hole with the noun phrase the Morning Star we get the true (1a), while if we plug in the Evening Star we get the false (1b). Yet the Morning Star and the Evening Star both refer to the planet Venus and are thus coreferential. (shrink)

Coreference resolution is the task of finding all expressions that refer to the same entity in a text. It is one of the higher level NLP (Natural Language Processing) tasks. It allows, for example, to extract more information about medical products from larger texts. A product such as ‘ambidextrous gloves’ may appear in a text in many different forms. For example, they could be referred to by the pronoun ‘they’, such as in this sentence. The algorithm presented in this (...) paper finds pronouns and for each of them (except the pleonastic ‘it’) it creates a coreference candidate with entities that appeared earlier in the same sentence or in the previous sentence. Each candidate (pair of mentions) is described by 48 binary features which represent their grammatical and location properties. In the training set, each pair is marked as a coreference or not, based on which a decision tree classifier is trained. A classifier with a high precision of 0.94 and a decent recall of 0.61 were obtained on the training set, still with a good precision out of a sample of 0.64. (shrink)

Coherence in conversations and in texts can be partially characterized by a set of coherence relations, motivated ultimately by the speaker's or writer's need to be understood. In this paper, formal definitions are given for several coherence relations, based on the operations of an inference system; that is, the relations between successive portions of a discourse are characterized in terms of the inferences that can be drawn from each. In analyzing a discourse, it is frequently the case that we would (...) recognize it as coherent, in that it would satisfy the formal definition of some coherence relation, if only we could assume certain noun phrases to be coreferential. In such cases, we will simply assume the identity of the entities referred to, in what might be called a “petty conversational implicature,” thereby solving the coherence and coreference problems simultaneously. Three examples of different kinds of reference problems are presented. In each, it is shown how the coherence of the discourse can be recognized, and how the reference problems are solved, almost as a by‐product, by means of these petty conversational implicatures. (shrink)

Of course, although this view on meaning was the prevailing one for almost a century, many of the people who initiated the enterprise of logical semantics, including people like Frege and Wittgenstein, had an open eye for all that it did not catch. However, the logical means which Frege, Wittgenstein, Russell, and the generation that succeeded them, had at their disposal were those of classical mathematical logic and set-theory, and these indeed are not very suited for an analysis of other (...) aspects of meaning than those which the slogan covers. A real change in view then had to await the emergence of other concepts, which in due course became available mainly under the influence of developments in computer science and cognate disciplines such as artificial intelligence. And this is one of the reasons why it took almost a century before any serious and successful challenge of the view that meaning equals truthconditions from within logical semantics could emerge. (shrink)

The problem discussed here is to find a basis for a uniform treatment of the relation between pronouns and their antecedents, taking into account both linguists' and philosophers' approaches. The two main candidates would appear to be the linguists' notion of coreference and the philosophers' notion of pronouns as variables. The notion of coreference can be extended to many but not all cases where the antecedent is non-referential. The pronouns-as-variables approach appears to come closer to full generality, but (...) there are some examples of pronouns of laziness which appear to resist either of the two approaches. (shrink)

There seems to be a perfectly ordinary sense in which different speakers can use an empty name to talk about the same thing. Call this fictional coreference. It is a constraint on an adequate theory of empty names that it provide a satisfactory account of fictional coreference. The main claim of this paper is that the pretense theory of empty names does not respect this constraint.

Because reference is not transparent, coreference is not transparent either: it is possible for the subject to refer to the same individual twice without knowing that the two acts of reference target the same individual. That happens whenever the subject associates two distinct yet coreferential files with two token singular terms. The subject may not know that the two files corefer, so her ascribing contradictory properties to the same object does not threaten her rationality. But if the subject deploys (...) the same file twice, in association with both of the singular terms, she is bound to know that she is referring to the same entity twice. (shrink)

It seems that when the term "coreference" is used, whether in linguistics or in philosophy, there is often presumed to be a consensus about what it is, or at least about what it is in the context where the term is introduced. I don't think the term deserves to have much use at all, insofar as it disguises more interesting linguistic and pragmatic relations between nominal forms in natural language. My preoccupation with these relations issues in part from some (...) of the central concerns and distinctions introduced in Evans (1980), an essay that has had wide and, in some cases, a very deep influence on how a variety of reference issues have been addressed in modern linguistics. As a linguist, my interest is in the way in which natural language shapes what we know and from that perspective, I want to understand how natural language sets boundary conditions on how linguistic forms can be used to achieve readings that pick out the same entity more than once. (shrink)

Recent work has brought to prominence the idea that some utterances contain occurrences of noun phrases that not only corefer, but do so in a particularly guaranteed or explicit way—call such occurrences ‘de jure coreferential’. Studies of de jure coreference have considered both the characteristics of the relation, and its explanation. Pinillos (154(2):301–324, 2011) argues that de jure coreference is non-transitive, and uses this as part of his argument for a new semantic primitive explaining de jure coreference. (...) In this paper, I draw attention to a neglected distinction between de jure coreference and another kind of coreference with which it might easily be confused. I then consider Pinillos’s claim that de jure coreference is non-transitive, and conclude that even using Pinillos’s definition of de jure coreference, it is far from straightforward that he has demonstrated the non-transitivity of de jure coreference. I question whether Pinillos’s characterisation of de jure coreference is adequate, and offer a plausible alternative characterisation. I show that on this new characterisation, there is no evidence that the relation is non-transitive. (shrink)

The aim of this paper is to argue that update semantics is a natural framework for contextually restricted quantification, and to illustrate its use in the analysis of anaphoric definite descriptions and certain other anaphoric terms.

Update semantics1 embodies a radical view on the relation between context and interpretation. The meaning of a sentence is identified with its context change potential, where contexts are identified with information states. The recursive definition of semantic interpretation is stated in terms of a process of updating an information state with a sentence. Meanings of sentences, then, are update functions. In general, these are partial functions, since the possibility to update with a sentence may depend on the fulfillment of certain (...) constraints (presuppositions, the presence of antecedents for anaphora, etc.). It is important to note that context and interpretation are interdependent: the interpretation process depends on the context, `. (shrink)

A desirable quality of a coreference resolution system is the ability to handle transitivity constraints, such that even if it places high likelihood on a particular mention being coreferent with each of two other mentions, it will also consider the likelihood of those two mentions being coreferent when making a final assignment. This is exactly the kind of constraint that integer linear programming (ILP) is ideal for, but, surprisingly, previous work applying ILP to coreference resolution has not encoded (...) this type of constraint. We train a coreference classifier over pairs of mentions, and show how to encode this type of constraint on top of the probabilities output from our pairwise classifier to extract the most probable legal entity assignments. We present results on two commonly used datasets which show that enforcement of transitive closure consistently improves performance, including improvements of up to 3.6% using the b3 scorer, and up to 16.5% using cluster f-measure. (shrink)

The aim of this paper is to argue that update semantics is a natural framework for contextually restricted quantification, and to illustrate its use in the analysis of anaphoric definite descriptions and certain other anaphoric terms.

Update semantics1 embodies a radical view on the relation between context and interpretation. The meaning of a sentence is identified with its context change potential, where contexts are identified with information states. The recursive definition of semantic interpretation is stated in terms of a process of updating an information state with a sentence. Meanings of sentences, then, are update functions. In general, these are partial functions, since the possibility to update with a sentence may depend on the fulfillment of certain (...) constraints (presuppositions, the presence of antecedents for anaphora, etc.). It is important to note that context and interpretation are interdependent: the interpretation process depends on the context, `. (shrink)

For more than three decades, research into the psycholinguistics of pronoun interpretation has argued that hearers use various interpretation ‘preferences’ or ‘strategies’ that are associated with specific linguistic properties of antecedent expressions. This focus is a departure from the type of approach outlined in Hobbs , who argues that the mechanisms supporting pronoun interpretation are driven predominantly by semantics, world knowledge and inference, with particular attention to how these are used to establish the coherence of a discourse. On the basis (...) of three new experimental studies, we evaluate a coherence-driven analysis with respect to four previously proposed interpretation biases—based on grammatical role parallelism, thematic roles, implicit causality, and subjecthood—and argue that the coherence-driven analysis can explain the underlying source of the biases and predict in what contexts evidence for each will surface. The results further suggest that pronoun interpretation is incrementally influenced by probabilistic expectations that hearers have regarding what coherence relations are likely to ensue, together with their expectations about what entities will be mentioned next, which, crucially, are conditioned on those coherence relations. (shrink)

Coreference resolution is a challenging natural language processing task, and it is difficult to identify the correct mentions of an entity that can be any noun or noun phrase. In this article, a semisupervised, two-stage pattern-based bootstrapping approach is proposed for the coreference resolution task. During Stage 1, the possible mentions are identified using word-based features, and during Stage 2, the correct mentions are identified by filtering the non-coreferents of an entity using statistical measures and graph-based features. Whereas (...) the existing approaches use morphosyntactic and number/gender agreement features, the proposed approach uses semantic graph-based context-level semantics and nested noun phrases in the correct mentions identification. Moreover, mentions without the number/gender information are identified, using the context-based features of the semantic graph. The evaluation performed for the coreference resolution shows significant improvements, when compared with the word association-based bootstrapping systems. (shrink)

Consider the following sentence: “Mary meditated on the sentence ‘Bill is a good friend’ and concluded that he was a good friend.” It is standardly assumed that in sentences of this sort, containing so‐called “closed” quotations, the expressions occurring between quotation marks are mentioned and do not take their ordinary referents. The quoted NP “Bill” refers, if anything, to the name ‘Bill,’ not to the individual Bill. At the same time, the pronoun “he,” apparently anaphoric on quoted “Bill,” refers to (...) the individual Bill. The case seems thus to invalidate the intuitive principle that pronouns anaphoric on referential expressions inherit their reference from their antecedents. The paper formulates the argument, argues that sentences exhibiting the described pattern do not constitute evidence against the intuitive principle, and proposes an alternative account of the anaphoric relation involved. (shrink)

Following Kit Fine (2007), we can say that the de jure pair represent the referent as the same while the second one does not do so. There are roughly three ways of capturing this difference. One could say that de jure coreference between two expression occurrences happen because (a) the occurrences have identical meanings, (b) they have identical syntactic properties, or (c) they enter into a semantic relation not grounded in identity of meaning or syntax. In what follows, I (...) give some reason to think that de jure coreference is not a transitive relation. As a consequence, we can rule out (a) and (b) just on these grounds alone (since identity is a transitive relation). (c) then looks promising. I argue that this gives further support for a relationist semantics along the lines of what Kit Fine has proposed. (shrink)

De jure coreference in a discourse is typically understood as explicit coreference that speakers are required to recognize in order to count as having correctly understood the discourse. For example, in an utterance of the sentence ‘Tom went to the market because he needed soy milk’, the two underlined terms are typically coreferential in a way that appreciating their coreference is required to fully understand the utterance. Often, de jure coreference is understood as an equivalence relation, (...) so in particular it is thought of as a transitive relation. However, Pinillos (Philos Stud 154(2):301–324, 2011) provides examples that apparently challenge the transitivity of de jure coreference. In this paper, I argue for two claims. First, while it is inconclusive whether the relevant terms in Pinillos’s examples satisfy his third condition, it is much clearer that they fail to satisfy his first two conditions. Given that Pinillos’s conditions capture important characteristics of de jure coreference, his examples do not successfully show the non-transitivity of de jure coreference. Second, I present an alternative account of his examples, one that shows which representation the anaphoric pronoun in an example of the sort that he presents is de jure coreferential with. (shrink)

Emergentist approaches to language are burdened with two responsibilities in contemporary cognitive science. On the one hand, they must offer a different and better understanding of the well-known phenomena that appear to support traditional formal approaches to language. On the other hand, they must extend the search for alternative explanations beyond the familiar languages of Europe and East Asia. I pursue this joint endeavor here by outlining an emergentist account for constraints on local anaphora in English and Balinese, with a (...) view to showing that, despite numerous proposals to the contrary, the two languages manifest essentially the same system of coreference and that the system in question is shaped by processing pressures rather than grammatical principles. (shrink)

Among other virtues, Mental Files Theory provides a straightforward explanation of de jure coreference, i.e. identity of referent guaranteed by meaning alone: de jure coreference holds between terms when these are associated with the same mental file from which they inherit their reference. In this paper, I discuss an objection that Angel Pinillos raises against Mental Files Theory and other similar theories: the theory predicts that de jure coreference should be transitive, just like identity. Yet there are (...) cases, involving ‘slash-terms’, in which transitivity fails, or so it seems. In his book Mental Files, Recanati replies that the mental files theorist can accommodate Pinillos’ exceptions by offering a refined model of merging files, the ‘partial merging’ model. While agreeing with Recanati on the need for such a model, I contend that, pace Recanati and Pinillos, de jure coreference is transitive even in the presence of slash-terms. I will first show that paradoxical consequences ensue if slash-terms are said to de jure corefer with several basic terms at once. Then, building on two different accounts Recanati gives of referential confusion, I will show that on both views, de jure coreference cannot hold because of the behaviour of confused slash-terms. I will conclude that, in Mental Files Theory, a slash-term can, at most, de jure corefer with only one basic term per context. (shrink)

A model is presented that addresses both the distribution and comprehension of different forms of referring expressions in language. This model is expressed in a formalism (Kamp & Reyle, 1993) that uses interpretive rules to map syntactic representations onto representations of discourse. Basic interpretive rules are developed for names, pronouns, definite descriptions, and quantified descriptions. These rules are triggered by syntactic input and interact dynamically with representations of discourse to establish reference and coreference. This interaction determines the ease with (...) which coreference can be established for different linguistic forms given the existing discourse context. The performance of the model approximates that observed in studies of intuitive judgments of grammaticality and studies using online measures of language comprehension. The model uses the same basic interpretive mechanisms for coreference within and between sentences, thereby linking the domain traditionally studied by generative linguists to domains that have been of concern primarily to psychologists and computational linguists. (shrink)

Nous présentons un ensemble d’observations et de repères méthodologiques issus de la constitution et de l’annotation du corpus MC4 « Modélisation Contrastive et Computationnelle des Chaînes de Coréférences », corpus de (très) petite taille qui a permis des expérimentations, des études de faisabilité puis l’élaboration d’un corpus de beaucoup plus grande taille, celui du projet Democrat « Description et modélisation des chaînes de référence : outils pour l’annotation et le traitement automatique ». Nos remarques traitent notamment de la taille d’un (...) corpus annoté et du besoin de documentation complète, même pour un très petit corpus incompatible avec des exploitations ADT et TAL. (shrink)

0 Introduction 0.1 Variation in nominal coreference Nominal coreference has received much interest in the field of text linguistics as an essential strategy ...

Nous présentons une étude en cours sur la coréférence lorsque le référent est un individu ou un groupe d’individus. Avec l’objectif d’annoter manuellement les expressions référentielles et les chaînes de coréférence d’un texte écrit, nous décrivons les problèmes rencontrés et les solutions adoptées face à la nature sémantico-pragmatique des phénomènes retenus, à leur complexité, à la présence fréquente d’ambiguïtés et de sous-déterminations, et aux relations d’appartenance parfois floue d’un individu à des groupes. Nous montrons comment faire évoluer les outils d’annotation (...) pour optimiser les analyses linguistiques et leur fournir de nouveaux indicateurs. (shrink)

Nous présentons une étude en cours sur la coréférence lorsque le référent est un individu ou un groupe d’individus. Avec l’objectif d’annoter manuellement les expressions référentielles et les chaînes de coréférence d’un texte écrit, nous décrivons les problèmes rencontrés et les solutions adoptées face à la nature sémantico-pragmatique des phénomènes retenus, à leur complexité, à la présence fréquente d’ambiguïtés et de sous-déterminations, et aux relations d’appartenance parfois floue d’un individu à des groupes. Nous montrons comment faire évoluer les outils d’annotation (...) pour optimiser les analyses linguistiques et leur fournir de nouveaux indicateurs. (shrink)

On retrouve la notion de c-commande dans plusieurs contraintes linguistiques sur l’interprétation d’éléments qui sont dépendants pour leur interprétation référentielle. À bien y penser, il est assez étonnant d’avoir une relation syntaxique comme contrainte sur ce type d’interprétation : cela semble tout à fait accidentel, on n’en voit pas la motivation. Cette relation syntaxique correspond en fait le signifiant d’un signe combinatoire, et en regardant le côté signifié plutôt que signifiant de cette relation, sa présence dans une contrainte sur la (...) dépendance référentielle découle naturellement de propriétés sémantico-conceptuelles de la référence et la prédication. (shrink)

: The open-ended character of natural languages calls for the hypothesis that humans are endowed with a recursive procedure generating sentences which are hierarchically organized. Structural relations such as c-command, expressed on hierarchical sentential representations, determine all sorts of formal and interpretive properties of sentences. The relevant computational principles are well beyond the reach of conscious introspection, so that studying such properties requires the formulation of precise formal hypotheses, and empirically testing them. This article illustrates all these aspects of linguistic (...) research through the discussion of non-coreference effects. The article argues in favor of the formal linguistic approach based on hierarchical structures, and against alternatives based on vague notions of “analogical generalization”, and/or exploiting mere linear order. In the final part, the issue of cross-linguistic invariance and variation of non-coreference effects is addressed. Keywords: Linguistic Knowledge; Morphosyntactic Properties; Unconscious Computations; Coreference; Linguistic Representations Conoscenza linguistica e computazioni inconsce Riassunto: Il carattere aperto del linguaggio naturale avvalora l’ipotesi che gli esseri umani siano dotati di una procedura ricorsiva che genera frasi gerarchicamente organizzate. Relazioni strutturali come il c-comando, espresse su rappresentazioni frasali gerarchiche, determinano tutte le proprietà formali e interpretative delle frasi. I principi computazionali rilevanti sono totalmente al di fuori della portata della coscienza introspettiva e così lo studio di tali proprietà richiede la formulazione di precise ipotesi formali e la loro verifica sperimentale. Questo articolo illustra tutti questi aspetti della ricerca linguistica, esaminando gli effetti di non-coreferenza. Si argomenta in favore dell’approccio linguistico formale basato su strutture gerarchiche e contro alternative basate su vaghe nozioni di “generalizzazione analogica” e/o che impiegano il semplice ordine lineare. Nella parte finale si affronta il tema dell’invarianza e della variazione cross-linguistica degli effetti di non-coreferenza. Parole chiave: Conoscenza linguistica; Proprietà morfosintattiche; Computazioni inconsce; Coreferenzialità; Rappresentazioni linguistiche. (shrink)

In this article, I present different characteristics of de jure coreference, including an epistemological one and a logical one. Then, I present the semantic relation of coordination as what explains or grounds the phenomenon of de jure coreference. And I compare two theories of relational semantics, Fine’s (2007) and Taschek’s (1995a, 1995b, 1998), with respect to two famous puzzles in the philosophy of language, Frege’s puzzle and Kripke’s puzzle about belief. Lastly, I briefly discuss coordination between empty terms.

Covaluation is the generalization of coreference introduced by Tanya Reinhart. Covaluation distributes in patterns that are very similar yet not entirely identical to those of binding. On a widespread view, covaluation and binding distribute similarly because binding is defined in terms of covaluation. Yet on Reinhart's view, binding and covaluation are not related that way: binding pertains to syntax, covaluation does not. Naturally, the widespread view can easily explain the similarities between binding and covaluation, whereas Reinhart can easily explain (...) the dissimilarities. Reciprocally, the widespread view finds it harder to explain the dissimilarities, whereas Reinhart finds it harder to explain the similarities. Reinhart and others have proposed more than one explanation of the similarities, but as I argue, these explanations do not work. Hence although I adopt Reinhart's view, I propose a new explanation of the similarities and dissimilarities between binding and covaluation: While Reinhart has invoked semantic structure only to explain dissimilarities, I do so to explain both similarities and dissimilarities at once. Finally, I examine in light of this approach the topics of language acquisition, only-constructions, the identity predicate, the Partee/Bach/Higginbotham problem, the Dahl puzzle and its recent versions by Roelofsen. (shrink)