The New Grammar Formalisms - A Tutorial Survey (Abstract)

TL;DR

This tutorial survey provides a concise overview of the burgeoning field of declarative grammar formalisms that emerged in the 1980s, contrasting them with the dominant transformational approaches of the 1970s. It highlights the shared core properties of these new formalisms, such as declarative semantics and unification, and traces their diverse inspirations from both linguistics and computational approaches like Prolog.

What Problem Does This Paper Address?

The paper addresses the shift in linguistic and computational grammar formalisms from the dominant transformational approaches of the 1970s to the emergence of declarative formalisms in the 1980s. In the 1970s, linguistics largely moved away from explicit rule formalisms, relying on informal notations for structural descriptions. Similarly, natural language processing (NLP) favored intertwined procedural approaches like augmented transition networks (ATNs). The paper aims to explain the rise of new formalisms that offered a more declarative and computationally tractable way to represent grammars, bridging the gap between linguistic theory and practical implementation.

Methods

The paper doesn't present a novel method or algorithm in the traditional sense. Instead, it acts as a tutorial survey, analyzing and categorizing existing declarative grammar formalisms. The "method" is essentially a comparative analysis based on shared core properties. These properties include:

  • Declarative Semantics: The ability to interpret the notation as a statement of facts rather than a procedure.
  • Type 2 Rule Format: Single-mother, unordered rules without explicit context sensitivity.
  • Node Admissibility: A preference for interpreting rules as governing node admissibility in a tree structure over string rewriting.
  • Recursive Tree/DAG Structure: A foundation based on recursively defined trees or directed acyclic graphs.
  • Unification: The primary operation for combining syntactic information.

The survey also traces the historical development and inspirations behind these formalisms, categorizing them based on their origins (linguistics vs. NLP) and key influences.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The "Acronym Explosion" Masks Commonality: The sheer number of new grammar formalisms (COG, CUG, DOG, etc.) might suggest radical differences, but the paper argues that they share fundamental characteristics, making them a cohesive group.
  • Declarative Semantics as a Game Changer: The availability of declarative semantics is a key differentiator from earlier, more procedural approaches, making grammars easier to understand and reason about.
  • Unification as a Unifying Principle: Unification, borrowed from logic programming, emerges as a crucial mechanism for combining information efficiently across different grammatical components.
  • Dual Origins: Linguistics and Prolog: The formalisms draw inspiration from both theoretical linguistics (e.g., Montague grammar, dependency grammar) and computational approaches like Prolog (leading to logic grammar formalisms).
  • Bridging Theory and Practice: Many of these formalisms are designed not necessarily to make strong linguistic claims, but to provide computationally tractable and linguistically natural ways to write grammars for NLP applications.

Key Results

The paper's "results" are conceptual rather than quantitative. The key outcome is the identification and characterization of a new class of grammar formalisms that dominated the 1980s. It establishes that these formalisms, despite their varied names and specific implementations, share a common set of core properties that distinguish them from previous paradigms. The survey provides a structured understanding of this landscape, highlighting the convergence of ideas from different research communities.

Limitations and Future Work

The paper acknowledges that the distinctions between formalisms, while sometimes subtle, can be significant and are not always captured by the shared properties alone. It also notes that a recent trend involves compiling formalisms designed for linguistic claims into those designed for computational tractability.

Future work, implied by the survey's nature, would involve deeper dives into specific formalisms, exploring their comparative strengths and weaknesses for particular linguistic phenomena or NLP tasks, and further investigating the interplay between linguistic theory and computational implementation. The paper sets the stage for such explorations by providing a foundational understanding of the declarative grammar landscape.

Summary updated

BibTeX
@inproceedings{1987new,
  title = {{The New Grammar Formalisms - A Tutorial Survey (Abstract)}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1987},
}
Figure