• Le 15 mars 2019
    De 15:00 à 17:30
    Campus Tertre
    Bâtiment Censive, Salle du LLING, C228

Vendredi 15 mars 2019, 15h
Responsable : Benjamin Spector (Institut Jean Nicod)

Titre: A decision-theoretic approach to Aristotle's square of opposition [Joint work with Emile Enguehard]

Résumé: English and other languages lexicalizs the quantifiers 'some', 'all' and 'no' but has no word for 'not all'. This asymmetry is also found in the temporal domain (no word for 'not always'). Horn (1973) suggested that the reason why 'not all' is not lexicalized is that that the word some carries the implicature 'not all', making 'not all' superfluous in some sense. This explanation is incomplete: if we found a language were 'not all' were lexicalized and 'some' were not, then we could 'explain' this à la Horn by saying that 'some' is superfluous as it is an implicature of 'not all'. Horn proposes that 'negative' quantifiers are inherently marked in order to make this proposal complete, so that there is an intrinsic preference to lexicalize 'some' rather than 'not all'. More recently Katzir & Singh (2013) proposed a somewhat similar account based on the view that quantifiers are built on the basis on some primitive concepts, with the result that the meaning of 'not all' is more complex than that of 'some'. 
In this work we offer a completely different line of explanation. Based on recent decision-theoretic models of pragmatics, we argue that a lexicon based on {no, some, all} is in a formally explicit sense more efficient for hearers and speakers than one based on {no, not-all, all}, given some independently motivated assumptions about the meaning of lexical predicates (namely, a lexical predicate such as dog tends to denote a minority  of the objects within a certain supercategory: there are less dogs than non-dog animals). The idea is that in the non-attested lexicon ({no, not-all, all}), on average speakers will use longer and less informative sentences than in the attested lexicon. This argument relies on a theory of how speakers choose their sentences and hearers interpret them, which is provided by the Rational Speech Act model of pragmatics. More generally, our approach provides an example of how reasoning about the expected utility of a toy language can shed light on certain typological generalizations.