Powerful ideas in computational linquistics - Implications for problem solving, and education

Gerhard FischerACL 1979

TL;DR

This paper argues that computational linguistics (CL) offers a rich context for teaching powerful, transferable problem-solving skills, moving beyond traditional numerical or physics-based computer science education. By engaging students with CL problems and tools, the authors demonstrate how to foster deeper understanding of language, cognition, and programming itself.

What Problem Does This Paper Address?

The authors identify a gap in computer science education: a lack of focus on problem-solving in domains beyond traditional numerical and physics problems. They observe that while computational linguistics (CL) is a natural fit for exploring complex problem-solving, it's often overlooked due to a disconnect between linguistics and computer science communities, and inadequate programming tools. This paper proposes that CL can serve as a powerful pedagogical tool, offering "powerful ideas" that are universally applicable and can illuminate cognitive processes, thereby enhancing both problem-solving abilities and understanding of language and computation.

Methods

The paper describes an experimental "Language Laboratory" designed to immerse students in CL problem-solving. This lab encourages an active, exploratory learning style, contrasting with traditional Computer-Assisted Instruction (CAI). Students are encouraged to explore existing CL programs, modify them, and design new ones. The approach emphasizes understanding the "why" behind language phenomena and translating this understanding into programs, rather than just learning formal programming rules. The authors highlight the use of various formalisms like LOGO, LISP, ATNs, and semantic networks, and advocate for designing mini-languages tailored to specific problem domains. A key example of their methodology is the development of a pattern matcher, presented as a "glass-box" tool that students can readily understand and modify.

Datasets

Not discussed in the paper. The paper focuses on the pedagogical approach and the nature of problems rather than specific datasets.

Key Insights and Ideas

  • CL as a Pedagogical Powerhouse: CL problems offer a unique context for teaching general problem-solving skills because they are often ill-defined, require creative design rather than just coding, and emphasize the problem formulation phase.
  • "Powerful Ideas" as Transferable Skills: Concepts like "divide and conquer," understanding the difference between syntax and semantics, rule systems, and recursive control structures are presented as universally useful "powerful ideas" that can be learned and applied across disciplines.
  • Language as a Window into Cognition: Programming in CL can make our own cognitive functions explicit, providing metaphors for understanding mental processes like knowledge representation, inference, and communication.
  • The Value of "Glass-Box" Learning: Making programs transparent and modifiable (like the pattern matcher example) allows students to actively engage with the underlying mechanisms, fostering deeper understanding than black-box CAI approaches.
  • Bugs as Informative Clues: In CL, bugs are seen not just as errors but as valuable indicators of discrepancies between a model/theory and its implementation, offering insights into the nature of both.

Key Results

The paper doesn't present quantitative results in terms of accuracy or performance metrics. Instead, it focuses on qualitative findings:

  • Students enjoyed working in the Language Laboratory and gained a deeper understanding of language, general problem-solving, and programming.
  • Students with less interest in traditional mathematical problems were motivated by language-oriented applications.
  • The approach helped students move beyond a narrow view of computers as mere "adding machines" to seeing them as general information processing devices.
  • Students became familiar with general problem-solving ideas like representation, planning, and debugging, with bugs in language programs being highly visible and easier to diagnose.

Limitations and Future Work

  • Lack of Formal Evaluation: The authors acknowledge that they have not conducted formal evaluations but have relied on informal investigations, questionnaires, and think-aloud protocols.
  • Future Work: The paper implies that further documentation of problem-solving processes and more detailed empirical studies would be beneficial. The examples provided (pattern matching for infix-to-prefix translation and grammar parsing) suggest that extending these to more complex scenarios and exploring other "powerful ideas" are natural next steps. The authors also suggest that CL problems can serve as an "entry point" to problem-solving, programming, and mathematics, implying a broader integration of these concepts into education.

Summary updated

BibTeX
@inproceedings{fischer1979powerful,
  author = {Gerhard Fischer},
  title = {{Powerful ideas in computational linquistics - Implications for problem solving, and education}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {1979},
}
Figure