Questioning Indefinites in Dialogues

The standard analysis of indefinites (e.g. Russell or more recently in the Discourse Representation Theory) views them as introducing a new discourse referent that is considered as independent from both the textual and extra-textual context. Still it can be observed that in many situations such as dialogue there are cases where sequences of indefinites are explicitly referring to the same entity, when for instance its properties have to be refined. On the basis of a corpus study of task-oriented dialogue transcriptions, we show that indefinites can only be coreferential if the events that bear them as arguments are coreferential as well. We thus identify the basic cues that may lead to the automatic determination of such situations (e.g. adverbials, sub-categorising statements etc.) and apply them to analyze our corpus examples. We also show that the configuration of dialogue acts (re-statements, question-answer pairs) directly influence the subsequent use of indefinite following an initial statement. 1. Standard Analysis of Indefinites The standard analysis of indefinite noun phrases is based on Russell, viewing them as introducing a quantifier. In traditional first order logic, existential and universal quantification states a relation between the denotations of two predicates, and is opposed to designation which consists in naming individuals by constants. For a sentence like (1), the standard translation for a is the following : (1) A man is walking : ∃x man(x) ∧ walking(x) The advantage of this analysis is to isolate noun phrases which do not refer to a particular individual, and to give a rather good approximation for linguistic observations like those well known as "quantifier scope ambiguities". But an important limitation of this view is that it is unable to give a correct account of phenomena like cross-sentential anaphora and "donkey sentences" (Kamp, 1981). This limitation was one of the criteria for considering that indefinites in a restricted sense – a N, Ns, two Ns – are not to be treated as quantifiers like at least one N, every N, exactly two Ns. It leads to a separate treatment of quantified NPs and indefinites in the DRT framework (Kamp and Reyle, 1993). Whereras quantified NPs introduce a tripartite structure, where the quantifier expresses a particular relation between a restrictor and a scope, indefinite NPs introduce just a new variable for an individual. In the following, we concentrate on such indefinites. The fact that indefinites introduce a new variable is intended to reflect the idea that they are used to introduce new discourse referents, whereas definite descriptions, demonstratives and pronouns are used to refer to already known entities. Such a correlation between linguistic 1 Other criteria and a more detailed overview can be found in Corblin (1994). forms and assumed familiarity or cognitive status has been proposed in particular by Prince's Taxonomy (1981) and refined by the Givenness Hierarchy (Gundel et al.,1993). For both, the use of indefinite descriptions is optimally associated to "brand new" entities (Prince, 1981), for which the hearer can only access a representation of the type described by N (Gundel et al., 1993). But following Gundel et al., a cognitive status of the Givenness Hierarchy entails all lower statuses, and a particular form can be often replaced by forms which require a lower status. Since the status "type-identifiable" is entailed by all other statuses, all other statuses meet necessary conditions for use of the indefinite article. Nevertheless, as shown in data from different languages, it is not the case. Therefore, Gundel et al (1993) invoke supplementary conversational implicatures, stating that the use of an indefinite noun phrase implicates that the referent is not uniquely identifiable, and hence not familiar. Related to this cognitive aspect of the use of indefinites – non familiarity – is a linguistic one, which is context independance. From a linguistic point of view, Corblin (1987) opposes the category of indefinites to the categories of definites and demonstratives on the basis of their relations to the context. He considers the category of indefinites as a typically context independent category, taking this feature as a consequence of the basic referential operations associated with the interpretation of indefinites : enumeration and extraction. This means, that a sentence of the type "n N X" (Two dogs bark) an always be interpreted as an extraction of n individual values from the the class of N, based on information given by X. It follows from this principle that each possible extraction is strictly independent from all the others, since the conditions for an extraction are fully defined by a given class N, a number n and predicative information X. This view on the functioning of indefinites excludes any connection to the context or to previous extractions. The condition of non familiarity and the interpretation principle of context independance justify the standard view on indefinites, which is a non coreferential one. in ria -0 00 99 39 0, v er si on 1 24 J an 2 00 9 Author manuscript, published in "DAARC (2000) 9 p"