Computational Data Analysis for Syntax

Ludmila Uhlihova, Iva Nebeska, Jan KralikCOLING 1982

TL;DR

This paper presents a computational approach to analyzing the syntactic structures of present-day standard Czech. By developing and applying a suite of computer programs to a large corpus, the researchers aimed to uncover quantitative linguistic principles and the functioning of language in different communicative spheres, particularly focusing on dependency syntax.

What Problem Does This Paper Address?

Understanding the intricate workings of language, especially its syntactic structures, has traditionally relied heavily on qualitative analysis. While this provides deep insights, it can be time-consuming and may miss broader statistical patterns. The authors address the challenge of systematically analyzing a large volume of linguistic data to uncover quantitative relationships and tendencies within the Czech language. They aim to move beyond anecdotal observations to a more rigorous, data-driven understanding of how syntactic elements function and interact in real-world communication. This is part of a larger project focused on the quantitative analysis of present-day standard Czech across various styles.

Methods

The core of this research lies in a sophisticated computational methodology for analyzing syntactic structures. The authors developed a suite of computer programs designed to process a large corpus of text. Key aspects of their approach include:

  • Corpus Creation: A substantial corpus of 540,000 word-forms was compiled from non-narrative texts (newspaper, administrative, and scientific). This corpus was meticulously prepared by encoding each word-form with detailed morphological and syntactic information.
  • Syntactic Coding: A digital code was assigned to each word, capturing its part of speech, morphological categories (case, gender, number, tense, mood, voice), and syntactic categories (subject, object, predicate, attribute, adverbials). Special characters were used for specific relationships (e.g., dependency direction) and constructions.
  • Lemmatization: Each word-form was also assigned a basic lexical form (lemma) for frequency analysis.
  • Dependency Syntax Focus: The analysis specifically focused on dependency syntax, mapping relationships between words and their governing elements.
  • Hierarchical Program Structure: The computer programs were structured hierarchically, starting with simple frequency counts of syntactic categories and features, progressing to more complex analyses of relationships between elements, and finally to detailed verbal contexts.
  • Data Input and Processing: The encoded data was fed into external computer memory via punched cards and magnetic tapes. Processing was carried out on a TESLA 200 computer.

Datasets

The primary dataset used in this study is a corpus of 540,000 word-forms (occurrences) derived from non-narrative texts, specifically:

  • Newspaper texts
  • Administrative texts
  • Scientific texts

This corpus was divided into 180 samples, each consisting of 3000 running word-forms. The paper does not mention the introduction of new benchmark datasets.

Key Insights and Ideas

  • Quantitative Syntax is Feasible: The research demonstrates that a large-scale, statistically-driven analysis of syntax is not only possible but also yields valuable insights into language structure and function.
  • Dependency Syntax Reveals Patterns: Focusing on dependency relationships allowed for a detailed mapping of how words and clauses are connected, revealing patterns not easily discernible through other methods.
  • Hierarchical Analysis Unlocks Complexity: The staged approach of the computer programs, moving from simple frequencies to complex interrelationships, effectively managed the complexity of syntactic analysis.
  • Syntax as a Bridge: The study highlights how syntactic analysis can bridge the gap between lexical and other linguistic levels (morphology, semantics), offering a deeper understanding of language as a system.
  • Sentence Patterns are Limited but Powerful: While sentence patterns are structurally abstract and limited in number, their frequent repetition in texts contributes to their stylistic value and communicative efficiency.

Key Results

The paper outlines results obtained through different sets of programs:

  • Simple Computational Characteristics: This yielded basic frequency data on sentence elements, sentence types (simple, complex, compound), syntagm types, and word/clause order. It also provided insights into how sentence elements are expressed (e.g., by nominal phrases vs. dependent clauses) and the frequency of various syntactic connectors. Length distributions of clauses and sentences were also determined.
  • Compound Computational Characteristics: This set of programs focused on relationships between syntactic elements and their part-of-speech, the syntactic relevance of morphological categories (like noun cases), the correlation between sentence length and complexity, and the linear position of subordinate clauses.
  • Types of Verbal Contexts Searched by Computer: This enabled the study of concrete lexical manifestations within syntactic structures. Findings included tendencies in the placement of certain adjective types, the relationship between semantics of coordinated elements and their function, correlations between subordinate clause frequencies and governing predicate semantics, and links between infinitive morphology and modality semantics. Specific attention was paid to syntactic structures involving verbs "to be" and "to have."

Limitations and Future Work

  • Limitations: The paper acknowledges that the project is not yet completely finished. While the computer-aided analysis is presented as a valuable counterpart to qualitative research, the full extent of its contribution to understanding the "functioning of language means in communication" is still being explored. The paper implies that the statistical data obtained confirms some intuitive expectations while leading to deeper insights in others, suggesting that the full implications of all findings may still be unfolding.
  • Future Work: The authors suggest that the obtained statistical data provides a rich material for further steps of analysis. The project's ongoing nature implies continued exploration of quantitative linguistic principles, relationships, and general laws governing the Czech language. The focus on understanding the "functional load of syntactic elements and units in texts from various styles and from various communicative spheres" indicates a direction for future detailed investigations.

Summary updated

BibTeX
@inproceedings{uhlihova1982computational,
  author = {Ludmila Uhlihova and Iva Nebeska and Jan Kralik},
  title = {{Computational Data Analysis for Syntax}},
  booktitle = {International Conference on Computational Linguistics},
  year = {1982},
}
Figure