Problemes Syntaxiques De L’indexation Automatique De Documents

Andree Borillo, Jacques VirbelCOLING 1967

TL;DR

This paper proposes a novel approach to automatic document indexing by focusing on syntactic analysis to build a more robust representation of document content. By combining syntactic and semantic information, the authors aim to overcome limitations of purely lexical or semantic indexing methods, leading to more accurate and comprehensive document representations.

What Problem Does This Paper Address?

Traditional automatic document indexing often relies on identifying keywords (lexical analysis) or understanding the meaning of concepts (semantic analysis). However, these approaches can struggle to capture the nuanced relationships between concepts within a document. This paper addresses the challenge of representing these relationships accurately, particularly in scientific abstracts, by introducing a method that leverages syntactic structures to build a more complete and interconnected representation of document content. The authors highlight that simply identifying keywords isn't enough; understanding how they relate to each other syntactically is crucial for effective information retrieval.

Methods

The proposed method involves a two-phase process:

  • Lexical Phase: This initial phase, assumed to be completed, involves identifying concepts in the natural language text and translating them into corresponding descriptors using an automatic dictionary.
  • Syntactic Phase: This is the core of the paper's contribution. It focuses on transforming the list of descriptors into a structured representation. This phase employs:
  • Syntactic Analysis: The text is segmented into units, and grammatical and functional information is attributed to each word. This involves identifying word categories (noun, verb, adjective, etc.) and their roles within a sentence.
  • Syntagme Construction: The paper defines three types of relations between descriptors: "consecutive" (dynamic interdependence), "comparative" (formal rapprochement), and "associative" (static relationship). Syntagmes are formed by pairs of descriptors united by these relations.
  • "Mots-outils" (Keywords/Connectors): These are identified as crucial markers for consecutive and comparative relations. The paper categorizes these based on grammatical type, semantic meaning, and syntactic structure.
  • Syntactic Schemas: These are defined as sequences of grammatical and functional codes that represent possible syntactic constructions around a "mot-outil."
    • Semantic Verification: For ambiguous cases or when syntactic information is insufficient, a "réseau notionnel" (notional network) is used to verify the semantic validity and orientation of relationships. This network represents semantic affinities between concepts.

Datasets

Not discussed in the paper. The authors mention using scientific abstracts, specifically from the "Bulletin signalétique du C.N.R.S." in the field of Psycho-Physiology, for their examples.

Key Insights and Ideas

  • Syntactic Structure is Key: The paper strongly advocates for the importance of syntactic analysis in document indexing, arguing that it provides a richer representation than purely lexical or semantic approaches.
  • "Mots-outils" as Anchors: Identifying and categorizing "mots-outils" is presented as a systematic way to discover and build consecutive and comparative relationships.
  • Syntactic Schemas for Generalization: The concept of "syntactic schemas" allows for the generalization of syntactic patterns, making the analysis more robust.
  • Semantic Verification as a Crucial Filter: The "réseau notionnel" acts as a vital semantic filter, resolving ambiguities and ensuring the validity of relationships that are syntactically plausible but semantically incorrect.
  • Associative Relations as Residual: The paper treats associative relations as a residual category, identified when other relation types are not applicable, emphasizing their importance in capturing broader conceptual links.

Key Results

The paper describes the process and methodology for achieving a structured document representation rather than presenting quantitative results. The primary outcome is the generation of a "documentary representation" that aims to be a connected graph, free of isolated descriptors. The authors illustrate their method with examples of how texts are analyzed and transformed into these structured representations, demonstrating the identification of consecutive, comparative, and associative relations.

Limitations and Future Work

  • Ambiguity of Associative Relations: The paper acknowledges the difficulty in using "mots-outils" for associative relations due to their numerous and heterogeneous nature. This leads to a different approach for associative relations, focusing on descriptors themselves.
  • Semantic Verification Necessity: The need for semantic verification in cases of syntactic ambiguity or insufficient information is highlighted, indicating that syntactic analysis alone is not always sufficient.
  • Incomplete Syntagmes: The paper discusses the occurrence of "incomplete syntagmes" due to missing correlanda, the presence of natural language terms without SYNTOL equivalents, or pronouns.
  • Future Work: The authors state that practical preparation for a new computer experiment is underway to determine the validity and limitations of their proposed method.

Summary updated

BibTeX
@inproceedings{borillo1967problemes,
  author = {Andree Borillo and Jacques Virbel},
  title = {{Problemes Syntaxiques De L’indexation Automatique De Documents}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1967},
}
Figure