Many speakers, many worlds: Interannotator variations in the quantification of feature norms

Linguistic Issues in Language Technology – LiLT Volume 13, Issue 2 May 2016 Many speakers, many worlds Interannotator variations in the quantification of feature norms Aur´elie Herbelot Eva Maria Vecchi Published by CSLI Publications LiLT volume 13, issue 2 May 2016 Many speakers, many worlds Interannotator variations in the quantification of feature norms Aur´elie Herbelot, University of Trento Eva Maria Vecchi, University of Cambridge 1 Introduction Quantification (see e.g. Peters and Westerst˚ahl, 2006) is probably one of the most extensively studied phenomena in formal semantics. But because of the specific representation of meaning assumed by model- theoretic semantics (one where a true model of the world is a priori available), research in the area has primarily focused on one question: given a model, what does it mean for a speaker to utter a statement of the form Qx[P(x)], where Q is a natural language quantifier such as no, few, some, many, most, all, at least 3... (or even a null quantifier ∅)? What is the relation of a quantifier to the truth value of a sentence? In contrast, relatively little has been said about the way the underlying model comes about, and its relation to individual speakers’ conceptual knowledge. Consider for instance the simple model in Fig.1. Given this state-of- affairs, where the set of cats and the set of black things overlap, the sentences some cats are black and at least one cat is black can be said to be logically true, while all cats are black is logically false. Expanding on this purely logical interpretation, researchers (e.g. Huang and Snedeker (2009), Grodner et al. (2010) and Degen and Tanenhaus (2015)) have 1 LiLT Volume 13, Issue 2, May 2016. Many speakers, many worlds. Copyright c ⃝ 2016, CSLI Publications. 2 / LiLT volume 13, issue 2 May 2016 cat black FIGURE 1 The set of cats and the set of black things overlap. further shown how quantifier preferences depend on a variety of factors ranging from the cardinality of the set involved to the ‘Question Under Discussion’ in the discourse. That is, depending on the number of cats involved in the state-of-affairs under consideration, humans may prefer the description two cats are black to some cats are black, despite both sentences being true. What semantics and pragmatics have so far failed to provide is an account of models themselves: how is it that in the first place, a speaker might model the world in a way that the set of cats and the set of black things overlap? In grounded situations, perceptual input arguably constrains the beliefs of the speaker about the observed state-of-affairs. If we stand in a room with five cats, two of which are black, it seems intuitive to infer that our model of the state-of-affairs includes a set of five cats, and the overlap between the set cats and the set of black things has cardinality 2. It is much less clear how the model comes about in non-grounded situations (which constitute the majority …

Content truncated to 500 words