A5029729156- Natural Language Processing Techniques
- Syntax, Semantics, Linguistic Variation
- Topic Modeling
- Linguistics and Discourse Analysis
- Speech and dialogue systems
- Text Readability and Simplification
- Linguistic Variation and Morphology
- Spanish Linguistics and Language Studies
- Language Development and Disorders
- Language, Discourse, Communication Strategies
- Logic, Reasoning, and Knowledge
- Epistemology, Ethics, and Metaphysics
- Historical Linguistics and Language Studies
Leibniz-Centre General Linguistics
2022-2023
New York University
2019-2020
Alex Warstadt, Yu Cao, Ioana Grosu, Wei Peng, Hagen Blix, Yining Nie, Anna Alsop, Shikha Bordia, Haokun Liu, Alicia Parrish, Sheng-Fu Wang, Jason Phang, Anhad Mohananey, Phu Mon Htut, Paloma Jeretic, Samuel R. Bowman. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.
Natural language inference (NLI) is an increasingly important task for natural understanding, which requires one to infer whether a sentence entails another. However, the ability of NLI models make pragmatic inferences remains understudied. We create IMPlicature and PRESupposition diagnostic dataset (IMPPRES), consisting 32K semi-automatically generated pairs illustrating well-studied types. use IMPPRES evaluate BERT, InferSent, BOW trained on MultiNLI (Williams et al., 2018) learn...
Vision-language models (VLMs), which process image and text inputs, are increasingly integrated into chat assistants other consumer AI applications. Without proper safeguards, however, VLMs may give harmful advice (e.g. how to self-harm) or encourage unsafe behaviours consume drugs). Despite these clear hazards, little work so far has evaluated VLM safety the novel risks created by multimodal inputs. To address this gap, we introduce MSTS, a Multimodal Safety Test Suite for VLMs. MSTS...
Abstract This paper presents a novel account for the optionality of negative concord observed with neither...nor coordinations, in an otherwise strict language—Turkish. I argue that apparently optional is surface phenomenon can be traced back to structural ambiguity arising from type-flexibility coordination operators. The semantic type affects its syntactic position relative Turkish NegP, which results different facts: whenever generalized quantifier type, it originates vP below akin other...
Children acquiring a non-negative concord language like English or German have been found to consistently interpret sentences with two negative elements in manner as conveying single semantic negation. Corpus-based investigations for and show that children also produce but only negation meaning. As any approach indefinites needs account both the typological variation child data, we revisit three most current syntactic Agree-based analyses, well movement-based they either difficulties data...
This article addresses the question of how root necessity modals are able to take scope over negative operators. Previous work has argued that wide readings derived by syntactic movement modal negation. We argue against this view. Reviewing facts from a number different languages, we show availability is not conditioned clause structure in which embedded, and deriving full range configurations where they found requires complications for rules involved. discuss other issues correct...
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 models. use a single linguistic phenomenon, negative polarity item (NPI) licensing in English, as case study for our experiments. NPIs like "any" are only if they appear environment negation ("Sue doesn't have any cats" vs....
We claim that weak necessity modals like English "should" are referential expressions denote pluralities of worlds, against the standard analysis, according to which all modal auxiliaries quantifiers. Weak pattern plural definites when tested for homogeneity effects (Löbner 2000, Križ 2016): they have scopeless readings under negation, tolerate exceptions in certain discourse contexts, and exhibit other properties characteristic homogeneous definite plurals. also discuss how extend this...
I give a new account of neg-raising with belief predicates as scaleless implicatures, inferences that are predicted by grammatical theories scalar when quantifier projects subdomain alternatives but no alternative. argue the inference should be treated an implicature because parallels observed in its distribution and other known cases implicatures (namely typical free choice effects, reported implicatures). Furthermore, is preferred over previously proposed analysis Romoli (2013)...
Natural language inference (NLI) is an increasingly important task for natural understanding, which requires one to infer whether a sentence entails another. However, the ability of NLI models make pragmatic inferences remains understudied. We create IMPlicature and PRESupposition diagnostic dataset (IMPPRES), consisting >25k semiautomatically generated pairs illustrating well-studied types. use IMPPRES evaluate BERT, InferSent, BOW trained on MultiNLI (Williams et al., 2018) learn...
The French necessity modals falloir and devoir can produce wide scope interpretations with respect to negation. However, when they are marked by the perfective aspect, this ability disappears, must take narrow scope. In paper, I present a novel analysis of these modals’ interpretation, as ‘scaleless implicature’ – strengthening phenomenon arising from lack possibility scalemate. This strengthening, argue, is blocked actuality entailment triggered aspect.