Where Do Aspectual Variants of Light Verb Constructions Belong?

Aggeliki Fotopoulou, Eric Laporte, Takuya NakamuraACL 2021

TL;DR

This paper tackles the classification challenge of "aspectual variants" of light verb constructions (LVCs) in French, which are often difficult to categorize as verbal idioms (VI), LVCs, or fully compositional phrases (CP). The authors propose a refined set of features and a modified decision tree to better distinguish these expressions, arguing for their inclusion as a subtype of LVC.

What Problem Does This Paper Address?

In linguistics, "light verb constructions" (LVCs) are phrases where a verb with minimal semantic content (like "take" or "have") combines with a noun to form a predicate (e.g., "take a decision"). However, a common phenomenon involves verbs that are aspectual variants of these LVCs. For instance, "take on debt" expresses the beginning of the state of "having debt." These aspectual variants, like "prendre conscience" (take awareness) for "avoir conscience" (have awareness), are frequent but pose a classification problem. They don't fit neatly into existing categories of verbal idioms (VI), standard LVCs, or fully compositional phrases (CP) because they add a temporal aspect (beginning, ending, duration) to the core meaning of the LVC. The paper aims to establish clearer criteria for classifying these expressions, proposing a more consistent approach than current guidelines.

Methods

The authors leverage the PARSEME guidelines for annotating verbal multiword expressions (MWEs) as a starting point. They survey existing features used to distinguish between VIs, LVCs, and CPs, focusing on those relevant to aspectual variants. Key features examined include:

  • Semantic contribution of the verb: Whether the verb adds only morphological meaning or a distinct aspectual meaning (e.g., beginning, cessation).
  • Equivalence with a part of the phrase: Whether a subphrase retains the predicative meaning.
  • Distributional constraints: How components of the phrase behave when substituted with related words.
  • Inflectional constraints: How changes in verb tense or aspect affect the meaning.

Based on their analysis, they propose a modification to the decision tree used for classification. Specifically, they suggest creating a new category, LVC.asp, for expressions that are negative to the LVC.3 test (verb adds aspectual meaning) and have a close relationship with a standard LVC. They also extend this to aspectual variants of prepositional phrase (PP) idioms, suggesting they can be analyzed as LVCs when combined with a copular verb.

Datasets

The paper primarily refers to the PARSEME corpus and its guidelines for annotating verbal MWEs. No new datasets were introduced.

Key Insights and Ideas

  • Aspectual variants are a distinct phenomenon: These expressions add a temporal dimension (beginning, duration, etc.) to the core meaning of an LVC, making them different from standard LVCs or VIs.
  • Existing guidelines are inconsistent: The PARSEME guidelines, while comprehensive, struggle to consistently classify these aspectual variants, leading to ambiguity and potential mislabeling.
  • A new category is needed: The authors argue for a dedicated category (LVC.asp) to house these aspectual variants, treating them as a subtype of LVC rather than forcing them into VI or CP categories.
  • Distributional similarity is key: The close relationship in terms of arguments and distributional constraints between aspectual variants and their corresponding LVCs suggests they should be treated similarly.
  • PP idioms also have aspectual variants: Similar to noun-based LVCs, prepositional phrase idioms can also have aspectual variants (e.g., "enter into force" from "in force"), which can be analyzed within the LVC framework using copular verbs.

Key Results

The paper doesn't present quantitative results in terms of accuracy scores. Instead, its key results are conceptual and methodological:

  • Identification of two sets of expressions: The authors identify "stricto sensu aspectual variants of LVC" (noun-based) and "aspectual variants of prepositional-phrase idioms."
  • Proposal for a refined classification: They advocate for a modified decision tree that incorporates a new category (LVC.asp) for these expressions, treating them as a subtype of LVC.
  • Justification for the proposed classification: The arguments are based on analyzing distributional constraints, semantic contributions, and relationships with existing LVCs and VIs, demonstrating that these aspectual variants share more characteristics with LVCs than with VIs or CPs.

Limitations and Future Work

  • Subjectivity in feature testing: The authors acknowledge that some tests, particularly those assessing the "meaning added by the verb" (LVC.3), can be subjective and difficult to apply reproducibly, leading to potential inconsistencies in annotation.
  • Scope of the study: The paper focuses primarily on French examples, although it notes that similar phenomena exist in other Romance languages and potentially English.
  • Future work: The authors suggest adapting the decision tree to include LVC.asp and exploring how these findings can be integrated into computational models for MWE processing and lexical databases. They also highlight the need for further investigation into the reliability of certain classification tests.

Summary updated

BibTeX
@inproceedings{fotopoulou2021where,
  author = {Aggeliki Fotopoulou and Eric Laporte and Takuya Nakamura},
  title = {{Where Do Aspectual Variants of Light Verb Constructions Belong?}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2021},
}
Figure