A Comprehensive Analysis of Constituent Coordination for Grammar Engineering

Agnieszka Patejuk, Adam PrzepiórkowskiCOLING 2012

TL;DR

This paper tackles the complex linguistic phenomenon of coordination, specifically how to formally model "unlike" and "very unlike" constituents in Polish within the Lexical Functional Grammar (LFG) framework. The authors propose novel LFG mechanisms, particularly leveraging off-path constraints, to accurately capture these challenging coordination patterns, paving the way for more robust and comprehensive natural language processing systems.

What Problem Does This Paper Address?

The paper addresses a significant challenge in computational linguistics and formal grammar: accurately modeling constituent coordination. While standard coordination of "like" categories (e.g., two noun phrases) is relatively straightforward, coordinating elements that are grammatically or semantically "unlike" (e.g., a noun phrase and a prepositional phrase, or even elements serving different grammatical functions) poses a considerable problem. Existing grammar implementations often struggle with these "non-standard" coordinations, which are, however, textually frequent. This limitation hinders the development of wide-coverage parsers and grammars, especially for morphologically rich languages like Polish, which exhibit complex interactions between coordination and phenomena like case assignment and agreement. The authors aim to provide a formal, linguistically motivated solution within LFG that can handle these complexities.

Methods

The paper proposes an LFG-based analysis for constituent coordination, focusing on two main areas:

  • Coordination of Unlike Categories: The authors extend the standard LFG approach to allow for coordination of elements that may not share the same grammatical function or category. They demonstrate how existing LFG mechanisms, particularly the handling of feature structures and the concept of distributive vs. non-distributive features, can be utilized.
  • Coordination of Distinct Grammatical Functions (Lexico-semantic Coordination): This is the more challenging aspect. The authors introduce and formalize the concept of "lexico-semantic coordination" in Polish, where coordinated elements might be dependents of different predicates. The key innovation here is the use of off-path constraints within LFG. These constraints allow for more flexible and nuanced checking of feature requirements across coordinated elements, effectively "smuggling" disjunctions into individual conjuncts before they are interpreted. This mechanism avoids the need to extend the core LFG formalism.

Datasets

The paper primarily uses attested examples from the National Corpus of Polish (NKJP) to illustrate and motivate the analysis. It also references the NKJP1M and NKJP250M subcorpora for statistical information on the frequency of coordination. No new datasets are introduced.

Key Insights and Ideas

  • Coordination is More Permissive than Assumed: The paper argues that coordination is subject to fewer constraints than previously thought, especially in languages like Polish. This challenges traditional assumptions in grammar engineering.
  • Off-Path Constraints are Key for Unlikes: The authors highlight the power of LFG's off-path constraints as a mechanism to elegantly handle the coordination of unlike constituents, particularly when those constituents might be dependents of different predicates. This is a significant contribution to LFG's expressive power for coordination.
  • Polish as a Rich Testbed: The paper emphasizes that Polish, with its complex case assignment and agreement system, serves as an excellent language for testing and developing robust coordination models. The interactions between coordination and these phenomena are crucial for a comprehensive grammar.
  • Valence Dictionaries Need "Coordinatability" Information: The authors point out that standard valence dictionaries lack information about whether arguments can be coordinated. They suggest that future dictionaries, like the developing "Walenty" for Polish, should explicitly encode this "coordinatability" information.

Key Results

The paper presents a formal LFG analysis that accounts for:

  • Coordination of unlike categories: This is achieved by leveraging existing LFG mechanisms for feature structures and distributive features.
  • Lexico-semantic coordination: This is handled through the novel application of off-path constraints, allowing for coordinated elements to be dependents of different predicates and to have distinct grammatical functions.
  • Interactions with case assignment: The proposed mechanisms are shown to integrate with Polish case assignment rules, demonstrating their practical applicability.

While the paper focuses on formal analysis rather than quantitative evaluation on a large dataset, it demonstrates the successful parsing of challenging constructed examples and claims that the implemented grammar correctly parses difficult instances of coordination encountered.

Limitations and Future Work

  • Space Constraints: The authors acknowledge that due to space limitations, many aspects of the implementation are omitted, particularly the agreement between verbs and coordinated subjects, and the phenomenon of single conjunct agreement.
  • Future Work: The authors suggest that the general mechanisms described are applicable to other languages. They also mention that the current grammar is being extended with more examples of unlike and lexico-semantic coordination for testing. The first stable version of the grammar will be made publicly available.

Summary updated

BibTeX
@inproceedings{patejuk2012comprehensive,
  author = {Agnieszka Patejuk and Adam Przepiórkowski},
  title = {{A Comprehensive Analysis of Constituent Coordination for Grammar Engineering}},
  booktitle = {International Conference on Computational Linguistics},
  year = {2012},
}

Loading related content...

Figure