Deverbal Compound Noun Analysis Based on Lexical Conceptual Structure

Koichi Takeuchi, Kyo Kageura, Teruo KoyamaACL 2003

TL;DR

This paper introduces a novel approach to analyzing the semantic relationships within compound nouns, particularly those derived from verbs. By leveraging a refined Lexical Conceptual Structure (LCS) framework, the authors provide a systematic method to disambiguate how the meaning of a modifier noun relates to its deverbal head, achieving high accuracy in both Japanese and English. This work offers a principled way to understand the underlying mechanisms of compound noun formation, moving beyond simple categorization.

What Problem Does This Paper Address?

Compound nouns, like "computer science" or "car repair," are incredibly productive in language. However, precisely defining the semantic relationship between their constituent parts (e.g., is it "science about computers" or "science done by computers"?) is a significant challenge. Existing methods often rely on broad semantic categories or probabilistic models that struggle with the nuanced interplay of lexical meaning and context. This paper aims to bridge this gap by proposing a principled, theory-driven approach that clarifies how the inherent semantic properties of words, specifically deverbal heads, govern these relationships. The authors highlight the difficulty in identifying the complete set of lexical factors needed for accurate analysis and seek to establish a systematic mechanism for understanding these complex nominal structures.

Methods

The core of this paper's contribution lies in its proposed Lexical Conceptual Structure (LCS)-based Disambiguation Model. The authors develop their own version of LCS, termed TLCS (Theoretical Lexical Conceptual Structure), which provides a clear set of basic predicates and LCS types designed to capture the semantic roles of arguments in deverbal heads.

The approach involves several key steps:

  1. Categorization of Deverbal Heads: They classify deverbal heads based on their TLCS types. Table 1 in the paper lists 12 such types, using predicates like [x ACT ON y] for "operate" or [x CONTROL [BECOME [y BE AT z]]] for "translate."
  2. Categorization of Modifier Nouns: Modifier nouns are categorized based on their behavior with respect to these TLCS types. This includes distinctions like:
  3. Accusativity: Whether a modifier can take an accusative case (+ACC vs. -ACC), which is particularly relevant for distinguishing adjuncts.
  4. Basic Components of TLCS: Categorizing modifiers based on their compatibility with specific TLCS predicates (e.g., +ON, -ON, +EC, -EC).
  5. Restriction Rules for Identifying Relations: A set of procedural rules is established to disambiguate the relationship between the modifier and the deverbal head. These rules prioritize certain categories (like -ACC modifiers or specific TLCS types for the head) to identify adjunct relations, and then use the interaction of lexical meanings and modifier categories to determine if the relation is an internal argument or an adjunct.

Datasets

The authors evaluated their method on two main sets of data:

  • Japanese Compound Nouns: 1223 two-constituent compound nouns with deverbal heads. This set was comprised of 809 from a dictionary of technical terms and 414 from newspaper articles.
  • English Compound Nouns: 200 compound nouns of technical terms, randomly extracted from a dictionary.

The paper does not introduce new datasets but rather uses existing collections of technical terms and newspaper articles for evaluation.

Key Insights and Ideas

  • LCS as a Unifying Framework: The research demonstrates that a refined LCS framework (TLCS) can systematically explain the semantic relations in compound nouns, offering a principled alternative to purely categorical or statistical methods.
  • Modifier-Noun Categorization is Crucial: The way modifier nouns are categorized (e.g., by their accusative case behavior or their semantic compatibility with TLCS predicates) is a vital component in disambiguating their role (argument vs. adjunct).
  • Cross-Lingual Applicability: The TLCS framework, originally developed with Japanese data, proves effective for analyzing English compound nouns as well, suggesting a degree of universality in the underlying semantic mechanisms.
  • The Power of Lexical Factors: The study strongly emphasizes that lexical factors, specifically the semantic structure of the deverbal head and the properties of the modifier, are key to disambiguating compound noun meanings.

Key Results

The proposed LCS-based approach achieved remarkably high accuracy:

  • Japanese: 99.3% (1215 out of 1223) of compound nouns were analyzed correctly.
  • English: 97% (194 out of 200) of compound nouns were analyzed correctly.

The authors also note that the -ACC modifier category appears particularly effective in disambiguating relations, especially when comparing Japanese and English, where English modifiers often take suffixes that signal their adjectival function.

Limitations and Future Work

The authors acknowledge a few limitations:

  • Ambiguity of Words: The model can struggle with words that have multiple meanings (e.g., "right" as an adjective vs. a noun), leading to misinterpretations.
  • Future Work: They propose to address word ambiguity by treating different senses of a word as distinct entries (e.g., 'right_adj', 'right_noun'). They also suggest further exploration into how their LCS-based approach can be integrated into syntactic analysis.

Summary updated

BibTeX
@inproceedings{takeuchi2003deverbal,
  author = {Koichi Takeuchi and Kyo Kageura and Teruo Koyama},
  title = {{Deverbal Compound Noun Analysis Based on Lexical Conceptual Structure}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2003},
}
Figure