Formal Properties of Rule Orderings in Linguistics

Francis Jeffry PelletierCOLING 1980

TL;DR

This paper delves into the formal properties of how linguistic rules are applied, moving beyond psychological explanations to analyze their generative power. The authors introduce a framework for comparing different rule-ordering theories, demonstrating that some seemingly distinct approaches are equivalent in their ability to generate linguistic structures.

What Problem Does This Paper Address?

In the late 1960s, linguists realized that the generative power of standard linguistic theories (like Chomsky's Aspects) was equivalent to unrestricted rewrite rules, a very powerful formal system. This led to a search for "stronger linguistic metatheories" – ways to constrain linguistic theories. Two main viewpoints emerged: the "L-view" (Language view), which focuses on restricting the types of rules allowed to limit generative power, and the "G-view" (Grammar view), which focuses on restricting rule application to simplify the language learning process. This paper focuses on the latter, specifically on how rules are ordered within a derivation. While many proposals for rule ordering exist, their formal implications for what grammars can generate (their generative capacity) are not always clear. The authors aim to formally compare these different rule-ordering theories to understand their true generative power and equivalence.

Methods

The paper adopts the "L-view" of constraints, focusing on the formal generative capacity of different rule-ordering theories. They analyze eight proposed rule-ordering mechanisms: Total Ordering, Partial Ordering, Semi Ordering, Semi Partial Ordering, Unorderings, Quasi Ordering, Random Ordering, and Simultaneous Application. The core of their method involves comparing these theories in terms of their "strong generative capacity," which refers to their ability to generate the same set of empirical claims (like ambiguities and paraphrases). They prove a series of theorems to establish relationships of superiority, equivalence, or non-comparability between these theories.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Formalizing Rule Orderings: The paper provides a rigorous, formal framework for analyzing the generative power of various rule-ordering mechanisms in linguistics, moving beyond purely psychological motivations.
  • L-view vs. G-view: It clearly distinguishes between the "L-view" (focus on generative power) and the "G-view" (focus on psychological learnability) of linguistic constraints, arguing for the former's importance in formal linguistic theory.
  • Equivalence of Theories: Surprisingly, some seemingly distinct rule-ordering theories are shown to be equivalent in their generative capacity. For instance, "Unorderings" (applying rules in any order, but only once) and "Semi Partial Orderings" are found to be equivalent to "Total Orderings" under certain conditions.
  • Hierarchy of Power: The paper establishes a hierarchy of generative power among the rule-ordering theories, revealing which theories are more expressive than others.
  • "Random Ordering" Nuance: The paper clarifies that what some linguists mean by "random ordering" often implicitly includes "universal principles" that constrain derivations, making it distinct from true "random ordering" where any order is permissible.

Key Results

The paper establishes a set of relationships between different rule-ordering theories based on their strong generative capacity. These relationships are visually represented in Figure 1. For example:

  • Total Orderings are shown to be stronger than Semi Partial Orderings.
  • Semi Partial Orderings are stronger than Unorderings.
  • Unorderings are stronger than Random Orderings.
  • Simultaneous Application is equivalent to Total Orderings (under the "non-contradictory" condition).
  • The paper also investigates specific "universal principles" that can be added to "random ordering," finding that some are equivalent to total orderings, while others fall in between total and partial orderings.

Limitations and Future Work

The authors acknowledge that their analysis is primarily focused on the "L-view" (generative power) and that the "G-view" (psychological plausibility) is not their main concern, though they briefly canvass arguments for it and find them inconclusive. They suggest that further work could explore the precise nature of these psychological claims. The paper also closes with a reflection on the role of mathematical linguistics for the "ordinary working linguist," implying a need for clearer communication and application of formal results.

Summary updated

BibTeX
@inproceedings{pelletier1980formal,
  author = {Francis Jeffry Pelletier},
  title = {{Formal Properties of Rule Orderings in Linguistics}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1980},
}
Figure