Scalar implicature as lexical nondefault inference

Abstract

Three experiments were conducted to determine the veracity of two separate approaches used to explain Scalar Implicature (SI) generation, namely, whether SIs arise as default inferences, or rather result from a pragmatic calculation based on context. Japanese second language (L2) learners of English were compared to first language (L1) English and Japanese speakers using the Truth Value Judgement Task (TVJT). While some previous observations showed L2 learners had no difficulty deriving SIs in TVJT, this study discovers L2 participants struggled more to generate SIs when confronted with quantifier most than when encountering quantifier some. This asymmetry of quantifier treatment was not observed in L1 participants. The asymmetry might not be surprising, given that, as often suggested in theoretical literature, most is semantically and morpho-syntactically more complex than some. The default lexical approach though still predicts L2 participants should have performed better at SI generation than our experiments found. The results also do not support the pragmatic approach, as there is no reason to believe under this view that L2 participants' performance generating SI should have differed between most and some. Through these results we suggest a third approach to SI generation, that is, SI is part of the lexical content attached to each quantifier but does not actually arise by default.