Applications of a Computer System for Transformational Grammar

Joyce FriedmanCOLING 1969

TL;DR

This paper introduces a computer system designed to assist linguists in writing transformational grammars. The system automates sentence generation, lexical insertion, and transformation application, allowing linguists to test and refine their grammars more efficiently and uncover subtle issues in their formalizations.

What Problem Does This Paper Address?

Writing a transformational grammar for even a small fragment of a natural language is a complex task. Linguists must ensure that individual grammar rules function correctly in isolation and, crucially, that they interact harmoniously to produce the desired linguistic output. The formal and notational details of grammar writing can be tedious, and the intricate ways rules interact can lead to unforeseen errors. This paper addresses the need for a tool that can help linguists manage this complexity, detect errors, and gain deeper insights into their grammars.

Methods

The paper describes a computer system written in Fortran IV for the IBM 360/67. This system embodies the theory of transformational grammar as outlined by Chomsky. It handles the three core components of a transformational grammar: phrase structure, lexicon, and transformations. The system can perform the full process of sentence generation, including generating phrase structures, inserting lexical items, and applying transformations. The system represents grammars as data, allowing for flexibility in input and modification. Transformations are formally defined by an identification, a structural description (SD), and a structural change (SC), with elementary operations used to define the structural changes to avoid ambiguity.

Datasets

Not discussed in the paper. The paper focuses on the system's functionality and its application to existing grammar fragments, rather than on training or evaluating on specific datasets.

Key Insights and Ideas

  • Automating Grammar Development: The core idea is that a computer can significantly aid linguists by automating the tedious aspects of grammar writing and testing, allowing them to focus on linguistic insights.
  • Error Detection and Refinement: The system helps identify errors in grammar formalization, ranging from simple typographical mistakes to more complex issues arising from the interaction of rules.
  • Forcing Precision: The need to translate linguistic rules into a formal computer-readable format forces linguists to be more precise in their statements, often revealing important problems they might otherwise overlook.
  • Testing Linguistic Models: The system can be used to test the adequacy of linguistic theories and their formalizations by seeing how well they can be implemented and used to generate language.
  • Instructional Tool: The system proves valuable in teaching linguistics, allowing students to experiment with grammars, make modifications, and observe the results.

Key Results

The paper demonstrates the system's capability to process and generate sentences based on input grammars. It presents examples of how the system can handle phrase structure rules, lexical insertion, and transformations. The authors highlight that the system can detect formal errors in grammars and provide feedback to the linguist. While not presenting quantitative performance metrics, the paper emphasizes the qualitative benefits: improved precision in grammar writing, easier error detection, and enhanced understanding of linguistic phenomena. The system has been used in various experiments and teaching contexts, confirming its utility.

Limitations and Future Work

  • Grammar Completeness: The system is designed to help write grammars, not necessarily to test fully completed grammars. Problems outside the scope of a given grammar are evident during testing.
  • Errors of Intent: The system cannot directly detect "errors of intent" where a grammar is syntactically correct but doesn't accurately describe the intended language. It relies on the linguist to interpret the output and identify such issues.
  • Linguistic Model Inadequacies: The system can reveal limitations in the underlying linguistic model itself, as seen in the discussion of the Core grammar's WH-attraction and Question-transformations.
  • Future Work: The paper mentions planned future experiments, including using the system to aid in writing a grammar of Japanese. The system's instructional use is also highlighted as a successful application.

Summary updated

BibTeX
@inproceedings{friedman1969applications,
  author = {Joyce Friedman},
  title = {{Applications of a Computer System for Transformational Grammar}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1969},
}
Figure