Though state-of-the-art sentence representation models can perform tasks requiring significant knowledge of grammar, it is an open question how best to evaluate their grammatical knowledge. We explore five experimental methods inspired by prior work evaluating pretrained sentence representation models. We use a single linguistic phenomenon, negative polarity item (NPI) licensing in English, as a case study for our experiments. NPIs like any are grammatical only if they appear in a licensing environment like negation (Sue doesn't have any cats vs. *Sue has any cats). This phenomenon is challenging because of the variety of NPI licensing environments that exist. We introduce an artificially generated dataset that manipulates key features of NPI licensing for the experiments. We find that BERT has significant knowledge of these features, but its success varies widely across different experimental methods. We conclude that a variety of methods is necessary to reveal all relevant aspects of a model's grammatical knowledge in a given domain. 1 Other prominent theories of NPI licensing are based on notions of non-veridicality (

Natural language inference (NLI) is an increasingly important task for natural language understanding, which requires one to infer whether a sentence entails another. However, the ability of NLI models to make pragmatic inferences remains understudied. We create an IMPlicature and PRESupposition diagnostic dataset (IMPPRES), consisting of >25k semiautomatically generated sentence pairs illustrating well-studied pragmatic inference types. We use IMPPRES to evaluate whether BERT, InferSent, and BOW NLI models trained on MultiNLI (Williams et al., 2018) learn to make pragmatic inferences. Although MultiNLI appears to contain very few pairs illustrating these inference types, we find that BERT learns to draw pragmatic inferences. It reliably treats scalar implicatures triggered by "some" as entailments. For some presupposition triggers like only, BERT reliably recognizes the presupposition as an entailment, even when the trigger is embedded under an entailment canceling operator like negation. BOW and InferSent show weaker evidence of pragmatic reasoning. We conclude that NLI training encourages models to learn some, but not all, pragmatic inferences.

I give a new account of neg-raising with belief predicates as scaleless implicatures, inferences that are predicted by grammatical theories of scalar implicatures, when a quantifier projects subdomain alternatives but no scalar alternative. I argue that the neg-raising inference should be treated as an implicature because of parallels observed in its distribution and that of other known cases of implicatures (namely typical scalar implicatures, free choice effects, and other reported cases of scaleless implicatures). Furthermore, the scaleless implicature account of neg-raising is preferred over the previously proposed analysis by Romoli (2013) of neg-raising as a scalar implicature, because Romoli has to make the ad hoc assumption that 'think' has an excluded middle alternative, while in this new account, the lack of a scalar alternative is predicted by its absence in the lexicon of English, and the presence of subdomain alternatives for quantifiers has been assumed in a variety of other work.

No abstract

No abstract