Analyse et classification automatique en chinois

Maurice CoyaudCOLING 1967

TL;DR

This paper explores the automatic analysis and classification of Chinese texts, focusing on the role of "numeral" particles (classifiers) in nominal classification. It reviews existing research and proposes a machine-assisted approach to build a comprehensive dictionary of numeral-noun combinations, crucial for both linguistic pedagogy and computational linguistics.

What Problem Does This Paper Address?

The automatic processing of Chinese text presents unique challenges due to its character-based nature and distinct grammatical structures. Specifically, this paper addresses the problem of accurately classifying Chinese nouns. This classification is heavily reliant on "numeral" particles, which function as classifiers and are essential for understanding noun phrases. The authors highlight that existing research has primarily focused on dependent analysis (translation-oriented) or has been limited in scope. There's a need for a more comprehensive and independent analysis that can be leveraged for machine translation and other computational linguistic tasks. The paper also points out the pedagogical importance of understanding these numeral-noun combinations, which is currently lacking in readily available resources.

Methods

The paper outlines a two-pronged approach:

  • Review of Existing Research: The authors begin by surveying Soviet and American research on automatic Chinese text analysis, categorizing it into dependent (translation-focused) and independent (language-agnostic) analysis. They emphasize the growing interest in independent analysis for its broader applicability.
  • Project for Nominal Classification: The core of the paper details a project focused on classifying Chinese nouns using numeral particles. This involves:
  • Building a Comprehensive Dictionary: The goal is to create a detailed dictionary that maps nouns to their corresponding numeral particles. This is a significant undertaking, as there are approximately 100 numeral particles in contemporary Chinese, and their usage is not always straightforward.
  • Machine Assistance: Given the scale of the task (classifying around 3,000 nouns), the authors propose using machine assistance to achieve accurate and error-free results. This is contrasted with their manual work on approximately 500 nouns.
  • Diachronic Analysis: The project also aims to study the evolution of the numeral system from ancient Chinese (CA) to contemporary Chinese (CC), identifying changes and continuities in numeral usage and classification.

Datasets

The paper does not explicitly list or introduce new datasets for training or evaluation. Instead, it refers to:

  • Corpora: The authors mention the use of corpora for analyzing ancient Chinese (CA) and contemporary Chinese (CC).
  • Existing Inventories: They reference previous work that has compiled lists of numeral particles, such as Herrfurth (1964) for Quoyu and Liu Shiru's work on ancient Chinese.
  • Manual Classification: The authors have manually classified around 500 nouns as part of their preliminary work.

Key Insights and Ideas

  • Numeral Particles as Key to Nominal Classification: The paper strongly emphasizes that numeral particles are not merely grammatical markers but are fundamental to understanding and classifying nouns in Chinese. Their correct identification and association with nouns are crucial for accurate linguistic analysis.
  • The Importance of Independent Analysis: Moving beyond translation-specific analysis, the paper advocates for an independent analysis of Chinese grammar that can be applied to various computational tasks, including classification.
  • Diachronic Evolution of Numeral Particles: The study highlights that the numeral system in Chinese has evolved significantly from ancient to contemporary times. Understanding this evolution is key to a complete picture of Chinese nominal classification.
  • Machine Assistance for Scalability: The sheer volume of Chinese nouns and the complexity of numeral particle usage necessitate machine assistance for comprehensive and accurate classification. Manual methods are too time-consuming and prone to error for large-scale analysis.
  • Pedagogical and Computational Value: The research has dual benefits: it can inform the teaching of Chinese by clarifying numeral-noun combinations and provide essential data for computational linguistics applications like machine translation and information retrieval.

Key Results

The paper presents preliminary findings from their manual classification efforts and theoretical framework:

  • Number of Numeral Particles: Contemporary Chinese (CC) has about 100 numeral particles (84 in a restricted sense and 16 indicating ordinary usage). Ancient Chinese (CA) had 141 proper numerals.
  • Intersection of Classes: The study identifies numerous intersections between nominal classes formed by numeral particles, indicating a complex web of relationships rather than simple disjoint sets.
  • Diachronic Changes: The paper notes that many numeral particles from ancient Chinese are not present in contemporary Chinese, and vice versa. For example, out of 141 CA numerals, 80 do not appear in CC, and out of 84 CC numerals, 24 were not attested in the CA corpora studied.
  • The Role of "ge": The numeral particle "ge" (gehua) is highlighted as a general classifier whose prevalence has increased from CA to CC, though the exact impact is still being evaluated.

Limitations and Future Work

  • Limited Manual Classification: The authors acknowledge that their manual classification of 500 nouns is incomplete and serves as a preliminary study.
  • Need for Machine Assistance: The primary limitation is the manual nature of the current work, which is why they propose machine assistance for a more robust and comprehensive analysis of approximately 3,000 nouns.
  • Generalizability: The authors express interest in extending their classification framework to other languages with numeral particles (e.g., Laotian, Vietnamese, Thai, Malay, Japanese, Maya) to enable precise cross-linguistic comparisons.
  • Further Diachronic Analysis: The evolution of the numeral system, particularly the "gehua," is an ongoing area of interest.
  • Exploration of Semantic Affinities: The paper suggests a future direction of using nominal classification to identify semantic affinities, which could help in choosing between grammatically acceptable interpretations.

Summary updated

BibTeX
@inproceedings{coyaud1967analyse,
  author = {Maurice Coyaud},
  title = {{Analyse et classification automatique en chinois}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1967},
}
Figure