Modals and lexically-regulated saturation

Depraetere, Ilse

doi:10.1016/j.pragma.2014.08.003

Cited by 18 publications

(6 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…On the other hand, human language processing has often been conjectured to have both context-dependent and context-independent properties (Barsalou, 1982;Rubio-Fernández, 2008;Depraetere, 2014Depraetere, , 2019. Given this divide, our approach may provide an alternative mechanism for clarifying how these two properties interact in the computational setting from both an interpretability standpoint (i.e.…”

Section: Discussion and Open Problemsmentioning

confidence: 99%

Interpreting Pretrained Contextualized Representations via Reductions to Static Embeddings

Bommasani¹,

Davis²,

Cardie³

2020

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

109

144

View full text Add to dashboard Cite

Contextualized representations (e.g. ELMo, BERT) have become the default pretrained representations for downstream NLP applications. In some settings, this transition has rendered their static embedding predecessors (e.g. Word2Vec, GloVe) obsolete. As a side-effect, we observe that older interpretability methods for static embeddings -while more mature than those available for their dynamic counterparts -are underutilized in studying newer contextualized representations. Consequently, we introduce simple and fully general methods for converting from contextualized representations to static lookup-table embeddings which we apply to 5 popular pretrained models and 9 sets of pretrained weights. Our analysis of the resulting static embeddings notably reveals that pooling over many contexts significantly improves representational quality under intrinsic evaluation. Complementary to analyzing representational quality, we consider social biases encoded in pretrained representations with respect to gender, race/ethnicity, and religion and find that bias is encoded disparately across pretrained models and internal layers even for models that share the same training data. Concerningly, we find dramatic inconsistencies between social bias estimators for word embeddings.

show abstract

Section: Discussion and Open Problemsmentioning

confidence: 99%

Interpreting Pretrained Contextualized Representations via Reductions to Static Embeddings

Bommasani¹,

Davis²,

Cardie³

2020

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics

109

144

View full text Add to dashboard Cite

show abstract

“…6 5 Given the aim of the chapter is not possible to address in detail the defining criteria of the modal meanings. See Depraetere and Reed (2011) and Depraetere (2014) for a discussion of the modal taxonomy adopted in this chapter. The tables below summarize the defining features that underlie the meaning distinctions: There are no examples of an affirmative declarative can immediately followed by a perfect infinitive in the 10,000,000-word BNC or in the 1,000,000-word ICE-GB.…”

Section: Modal Meaningsmentioning

confidence: 99%

“…These are meaning components that were not addressed in Grice's theory of conversation. (Depraetere 2014) Saturation is an obligatory process that is linguistically mandated (see Carston 2004: 49). There is a linguistic variable that needs to be specified in order for the sentence to communicate a complete (minimally truth-evaluable) proposition.…”

Section: The Meaning Of Have and The Semantics-pragmatics Interfacementioning

confidence: 99%

The Meanings of have and the Semantics/Pragmatics Interface

Depraetere

2017

Logic, Argumentation &Amp; Reasoning

Self Cite

View full text Add to dashboard Cite

In this chapter, I analyse the contribution made by have to the meanings expressed in sentences with a (central) modal verb followed by a perfect infinitive. Four meanings are discussed: (a) have as a marker that locates the possibility or necessity in the past; (b) have as a marker that establishes a relation of anteriority between the modal meaning (of possibility or necessity) and the situation referred to, (c) have as a marker of counterfactual meaning, (d) have as a marker of actualization. In the first part of the chapter an empirical overview is given. Drawing on the discussion in Chap. 2, I then address the question of how the role of the context should be captured, more specifically, whether it is saturation or free pragmatic enrichment that plays a role in bringing about the different meanings of non-finite have when it combines with a modal. KeywordsNon-finite have • Modal verbs • Saturation * I am very grateful to Susan Reed, for insightful discussion of a draft of this paper, and for her critical observations on the notion of 'factual have' (Depraetere 2009), which was relabelled in terms of the more accurate 'actualisation have' on her suggestion. AimsThis chapter offers an analysis of the contribution of have to the meanings expressed in sentences with a central modal auxiliary of necessity or possibility followed by a perfect infinitive. 1 Even though the meaning of the perfect in sentences with modals has been addressed in previous research (

show abstract

“…For instance, the modal in the common expression if I could just say... is standardly understood as expressing permission, which is why it sounds funny when this string of words ends up being used to express another kind of root possibility, such as ability (e.g., Homer Simpson: "If I could just say a few words… I'd be a better public speaker"). 2 So, while there is a semantic core of 'possibility' meaning (referred to as the 'context-independent semantic layer' in Depraetere 2014), in context, this meaning is made more specific and the hearer chooses between one of several limited options (a process which constitutes 'the context-dependent semantic layer' in Depraetere 2014). That is, assigning to a modal a particular meaning such as 'have permission to' rather than 'have the ability to', 'have the opportunity to', etc.…”

Section: Introduction: Modal Constructions Lexically Restricted Satumentioning

confidence: 99%

“…In the case of If I could just say…, what is meant is more than just 'If I had permission to say…'; this sequence of words is a standard way of asking for permission (i.e., the expression functions as a request) and of expressing that the speaker assumes that this permission is in fact granted (i.e., the expression also functions as a floor-taker) (Cheng 2007: 206;Cappelle & Depraetere 2016a: 27). In Depraetere's (2014) multi-layered approach to modal meaning, this layer is called the 'context-dependent, pragmatic layer'.…”

Section: Introduction: Modal Constructions Lexically Restricted Satumentioning

confidence: 99%

The necessity modalshave to,must,need to, andshould

Cappelle

Depraetere

Lesuisse

2019

Self Cite

View full text Add to dashboard Cite

When an ambiguous lexical item appears within a familiar string of words, it can instantly receive an appropriate interpretation from this context, thus being saturated by it. Such a context may also short-circuit illocutionary and other pragmatic aspects of interpretation. We here extract from the British National Corpus over 500 internally highly collocating and high-frequency lexical n-grams up to 5 words containing have to, must, need to, and/or should. These contexts-as-constructions go some way toward allowing us to group these four necessity modals into clusters with similar semantic and pragmatic properties and to determine which of them is semantico-pragmatically most unlike the others. It appears that have to and need to cluster most closely together thanks to their shared environments (e.g., you may have/need to…, expressing contingent, mitigated necessity), while should has the largest share of unique n-grams (e.g., rhetorical Why shouldn’t I…?, used as a defiant self-exhortation).

show abstract

Modals and lexically-regulated saturation

Cited by 18 publications

References 28 publications

Interpreting Pretrained Contextualized Representations via Reductions to Static Embeddings

Interpreting Pretrained Contextualized Representations via Reductions to Static Embeddings

The Meanings of have and the Semantics/Pragmatics Interface

The necessity modalshave to,must,need to, andshould

Contact Info

Product

Resources

About