Summary of Some Computational Aids for Obtaining a for Semantic Description of English

John Olney, Carter Revard, Paul ZiffCOLING 1967

TL;DR

This paper introduces computational approaches to automatically extract semantic information from English dictionaries, specifically Webster's Seventh New Collegiate Dictionary (W7) and the New Merriam-Webster Pocket Dictionary (MPD). The goal is to build a formal semantic description of English by generating lexical fields and semantic networks, aiding in the discovery of rules for word sense disambiguation and semantic component identification.

What Problem Does This Paper Address?

Creating a comprehensive formal semantic description of a language (FSDE) is a monumental task. It requires defining a set of lexical entries and rules that can accurately interpret the meaning of sentences and discourses. A key challenge is the sheer volume of information and the complex relationships between word senses. This paper tackles the problem of how to computationally process existing dictionary resources to accelerate the discovery of these lexical entries and semantic rules, moving towards a more automated approach to building an FSDE.

Methods

The core of the paper's methodology lies in processing machine-readable versions of two English dictionaries: Webster's Seventh New Collegiate Dictionary (W7) and the New Merriam-Webster Pocket Dictionary (MPD).

The authors are developing programs to:

  • Generate Derivational Lexical Fields: These fields group words and phrases that share a common base word or affix, displaying their senses as defined in W7. This helps in understanding how morphological changes affect meaning and syntactic function.
  • Generate Semantic Fields: These are represented as labeled networks where nodes are semantically related senses of a word, and labels denote the type of relationship (e.g., synonymy, antonymy). The construction of these fields considers word occurrences in definitions, syntactic features of defining phrases, stylistic regularities, and etymological information.
  • Identify Semantic Components and Primitives: The research aims to discover semantic components and rules for combining them. A procedure is outlined for identifying potential semantic primitives by iteratively finding sets of words whose definitions cross-reference each other.
  • Utilize MPD for Disambiguation: The smaller MPD is used to help disambiguate word senses found in W7 definitions. By identifying which W7 senses are abbreviated in MPD, they can provisionally eliminate less likely senses.

Datasets

  • Webster's Seventh New Collegiate Dictionary (W7): Transcribed into machine-usable form, aiming for completeness including typographic distinctions.
  • The New Merriam-Webster Pocket Dictionary (MPD): Also transcribed into machine-usable form.

The paper does not introduce new benchmark datasets but rather leverages existing dictionaries as the primary data source for their computational analysis.

Key Insights and Ideas

  • Dictionaries as Rich Sources for Semantic Analysis: The paper highlights that even traditional dictionaries, when processed computationally, can yield significant insights into semantic relationships and structures.
  • The Power of Derivational Fields: Manually constructing derivational fields has already proven useful in discerning how affixes shift meaning and syntactic function, suggesting this computational approach can automate this discovery.
  • Semantic Fields as Networks: Representing semantic relationships as labeled networks offers a structured way to visualize and analyze connections between word senses.
  • MPD as a Disambiguation Tool: The smaller MPD, despite its limited scope, serves as a valuable tool for disambiguating word senses within the more comprehensive W7, by acting as a filter for likely meanings.
  • Iterative Discovery of Semantic Primitives: The proposed procedure for finding semantic primitives through iterative cross-referencing suggests a systematic way to uncover fundamental semantic building blocks.

Key Results

The paper focuses on the development of computational aids and the discovery process rather than presenting definitive quantitative results of a fully built FSDE. However, it reports:

  • Successful transcription of W7 and MPD: Making these dictionaries available for computational processing.
  • Construction of derivational fields: This has been done manually for a considerable number of affixes, demonstrating the feasibility of the approach.
  • Preliminary indications from semantic field generation: Early results suggest that semantic fields can be constructed by considering word occurrences in definitions and other linguistic clues.
  • Identification of potential errors: The authors anticipate that errors in tracing semantic fields will arise from homographs used in different senses within definitions.

Limitations and Future Work

  • Limitations:
    • The discovery of rules and semantic components is acknowledged as a long-term effort requiring many man-years of work and collaboration.
  • The tracing of semantic fields will likely require close user monitoring, at least initially.
  • The process of identifying semantic primitives is described as a "gross approximation."
  • Future Work:
    • Continue the discovery of rules for obtaining senses from other senses and for combining semantic components.
  • Refine the generation of semantic fields and explore their use in discovering semantic components.
  • Develop precise terminology for describing semantic relationships through conceptual analysis of dictionary definitions.
  • Apply the routines to lexicography and other computational linguistics tasks not directly related to semantics.
  • Present preliminary results from the semantic primitive discovery procedure, stylistic analysis of W7, and derivational process studies.

Summary updated

BibTeX
@inproceedings{olney1967summary,
  author = {John Olney and Carter Revard and Paul Ziff},
  title = {{Summary of Some Computational Aids for Obtaining a for Semantic Description of English}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1967},
}
Figure