Bulgarian Inflectional Morphology in Universal Networking Language

Velislava StoykovaCOLING 2012

TL;DR

This paper presents a web-based application that models Bulgarian inflectional morphology using semantic networks and the Universal Networking Language (UNL). It tackles the complexity of Bulgarian's rich inflectional system by developing a formal representation within UNL, demonstrating its potential for underresourced languages.

What Problem Does This Paper Address?

Modeling inflectional morphology, especially for languages with complex systems like Bulgarian, is a significant challenge in Natural Language Processing (NLP). Bulgarian exhibits a rich inflectional system with numerous forms for nouns, verbs, and other parts of speech, often involving sound alternations and irregular patterns. Existing NLP approaches struggle to adequately represent this complexity, particularly when aiming for language-independent formalisms. This paper addresses the need for a robust method to encode Bulgarian inflectional morphology in a way that is both computationally tractable and can be integrated into universal language processing frameworks.

Methods

The authors propose a web-based application that leverages the Universal Networking Language (UNL) to model Bulgarian inflectional morphology. UNL represents information as a hypergraph of directed binary labeled links ("relations") between "Universal Words" (UWs), which represent universal concepts.

Key methodological aspects include:

  • Semantic Networks: The core of the approach uses semantic networks to represent the relationships between linguistic elements and their grammatical features.
  • Universal Networking Language (UNL): UNL is employed as the formalism for encoding this information. UNL's structure allows for language-independent representation of concepts and their relationships.
  • A-rules (Affixation Rules): The paper focuses on UNL's "A-rules" for handling inflection. These rules are designed to add morphemes (prefixes, suffixes, infixes) to base forms to generate inflected word forms. They distinguish between simple A-rules (single action) and complex A-rules (multiple actions).
  • Handling Sound Alternations: The approach aims to interpret sound alternations and irregularities within the definition of the main inflectional rules, rather than requiring separate mechanisms.
  • Lexical Representation: The system utilizes a UNL knowledge representation scheme that allows for semantic hierarchical representation, similar to WordNet, enabling synonym representation and translation.

Datasets

The paper mentions that the lexical resources used for this application were developed within the "The Little Prince Project" of the UNDL Foundation. This project aimed to create UNL grammar and lexical resources for several European languages based on the book "The Little Prince." Consequently, the lexicon is limited to the text of this book.

Key Insights and Ideas

  • UNL's Expressive Power for Inflection: The study demonstrates that UNL, with its A-rules, can effectively represent and generate various inflected forms of Bulgarian words, including handling sound alternations.
  • Unified Approach to Morphology: By using A-rules, the system provides a unified way to encode inflectional morphology, differentiating between inflectional, semantic, and lexemic hierarchies.
  • Potential for Underresourced Languages: The successful application to Bulgarian, an underresourced language in many NLP contexts, highlights UNL's potential for building NLP resources for such languages.
  • Integration of Lexical and Morphological Information: The UNL framework allows for the integration of lexical information (like synonyms) with morphological rules, facilitating richer language processing.

Key Results

The paper demonstrates the application of UNL to model Bulgarian inflectional morphology for nouns, adjectives, numerals, pronouns, and verbs.

  • Nouns: 74 word inflectional types were defined, each with specific rules for generating inflected forms based on number and definiteness. The example of the word "ezik" (language) is presented, showing the generation of its inflected forms.
  • Adjectives: 14 word inflectional types were used, covering gender, number, and definiteness.
  • Numerals and Pronouns: 5 and 6 word inflectional types, respectively.
  • Verbs: Represented by 48 inflectional types.
  • Syntactic Account: The UNL interpretation also includes 21 syntactic rules for subcategorization frames, linearization, and defining semantic relations.

Limitations and Future Work

  • Lexicon Limitation: The lexicon is currently limited to the text of "The Little Prince," which restricts the breadth of vocabulary and inflectional patterns covered.
  • Further Improvement: The authors state that the application is open for further improvement and development. This includes introducing additional grammar rules and enlarging the database for use in different projects.
  • Database Expansion: A key future direction is to expand the UNL grammar and lexical resources to cover a wider range of Bulgarian linguistic phenomena and vocabulary.

Summary updated

BibTeX
@inproceedings{stoykova2012bulgarian,
  author = {Velislava Stoykova},
  title = {{Bulgarian Inflectional Morphology in Universal Networking Language}},
  booktitle = {International Conference on Computational Linguistics},
  year = {2012},
}
Figure