DIALOGIC: A Core Natural-Language Processing System

Barbara Grosz, Norman Haas, Gary Hendrix, Jerry Hobbs, Paul Martin, Robert Moore, Jane Robinson, Stanley RosenscheinCOLING 1982

TL;DR

This paper introduces DIALOGIC, a modular natural-language processing system designed to translate English sentences into formal "logical forms" representing their literal meaning. Its core innovation lies in its flexible, modular architecture that facilitates transportability to new domains and integration into various applications, aiming to serve as a robust core for diverse NLP tasks.

What Problem Does This Paper Address?

The paper addresses the challenge of building robust and adaptable natural-language understanding (NLU) systems. Specifically, it tackles the need for a system that can accurately interpret the literal meaning of English sentences and represent it in a formal, machine-processable way. A key goal is to create a "core" processing component that can be reused across different applications, reducing the effort required to adapt NLU capabilities to new domains. This involves bridging the gap between the surface structure of natural language and its underlying semantic meaning, while ensuring the system's design is modular enough to be extended and integrated into larger projects.

Methods

DIALOGIC employs a modular architecture coordinated by an executive system called DIAMOND. The system comprises five key modules:

  • DIAGRAM Grammar: A comprehensive phrase-structure grammar of English, designed to handle a wide variety of sentence structures and linguistic phenomena. It's an augmented phrase-structure grammar where rules have associated constructors for admissibility testing and translators for producing logical forms.
  • Semantic Translators: These modules work in conjunction with the grammar to translate parsed phrases into logical form fragments. They define how the meaning of a phrase is composed from the meanings of its constituents.
  • Basic Semantic Functions: These functions attach logical-form fragments (lffs) and markers to the parse tree, insulating grammar changes from logical form changes. They handle the initial construction of semantic representations.
  • Scoping Algorithm: This algorithm collects the logical-form fragments and generates possible scopings for quantifiers and other sentence operators, ranking them based on a set of "critics" that evaluate various aspects of English scoping rules.
  • Basic Pragmatic Functions: These functions resolve indeterminacies in the logical form that require contextual or pragmatic information, such as interpreting prepositions, establishing relationships between noun phrases, and resolving pronoun references.

The system's design emphasizes a coupling of syntax and semantics, where translators define how logical form translation is derived from the attributes of words and phrases, aligning with theories that advocate for syntax and semantics working in tandem.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Modularity for Reusability: The core insight is that a modular design for natural-language processing, with clearly defined components, significantly enhances reusability and transportability across different applications and domains. Changes in one module (e.g., grammar) have minimal impact on others.
  • Bridging Syntax and Semantics: DIALOGIC explicitly couples syntactic parsing with semantic translation, ensuring that the grammatical structure directly informs the construction of the logical form. This is achieved through translators associated with grammar rules.
  • Compositional Semantics: The system aims to produce "logical forms" that are "purely formal" yet "as close as possible to the structure of natural language," emphasizing semantic compositionality for meaning-dependent processing.
  • Scoping as a Separate Problem: Quantifier scoping is treated as a distinct, post-parsing problem, handled by a specialized algorithm that generates and ranks potential scopings.
  • Pragmatics for Ambiguity Resolution: The inclusion of basic pragmatic functions highlights the necessity of incorporating contextual information to resolve ambiguities and derive a complete, context-specific interpretation.

Key Results

The paper describes the architecture and components of DIALOGIC, illustrating its functionality with an example. While specific quantitative results (e.g., accuracy metrics, processing speed) are not presented, the paper demonstrates how the system can parse an English sentence ("What SRI employees have children older than 15 years?") and construct a corresponding logical form. It shows how different modules contribute to this process, from syntactic parsing to semantic interpretation and quantifier scoping. The system is shown to be capable of handling complex noun phrases, relative clauses, and quantifiers, and it outlines the process of resolving ambiguities using pragmatic functions.

Limitations and Future Work

  • Limited Pragmatic Coverage: The paper acknowledges that only a "small core" of pragmatic functions is currently implemented, handling only a subset of the intended cases.
  • Incomplete Scoping: The scoping algorithm's set of critics is not exhaustive; it needs to address nonstandard quantifiers and generalized negative quantifiers.
  • Declarative Translation: Future work includes investigating a more declarative approach to specifying translation, where logical types are associated with phrasal categories, and translation is synthesized from constituent translations via functional application.
  • Grammar Coverage: The DIAGRAM grammar is continually being extended to cover more linguistic phenomena, and the formalism is being explored for better handling of linguistic generalizations and manageability.
  • Conjunction Handling: Conjunction processing is limited, pending further study.

Summary updated

BibTeX
@inproceedings{grosz1982dialogic,
  author = {Barbara Grosz and Norman Haas and Gary Hendrix and Jerry Hobbs and Paul Martin and Robert Moore and Jane Robinson and Stanley Rosenschein},
  title = {{DIALOGIC: A Core Natural-Language Processing System}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1982},
}
Figure