Proceedings of the Workshop on Computational Modeling of Polysynthetic Languages

Judith L. KlavansCOLING 2018

TL;DR

This paper presents the proceedings of a workshop focused on the computational modeling of polysynthetic languages, which are characterized by complex word structures. It highlights the challenges and opportunities in applying computational linguistics techniques to these languages, aiming to foster collaboration between researchers and practitioners for their documentation, preservation, and revitalization.

What Problem Does This Paper Address?

The paper addresses the significant challenges in computationally modeling polysynthetic languages. These languages, common among indigenous communities, feature intricate morphology where a single word can convey the meaning of an entire sentence in other languages. This complexity makes traditional Natural Language Processing (NLP) techniques, often developed for less morphologically rich languages, inadequate for tasks like documentation, corpus building, and language revitalization. The workshop aimed to bridge the gap between linguistic theory, computational methods, and practical applications, bringing together diverse communities to share knowledge and develop solutions.

Methods

The papers presented in these proceedings explore a range of methods for tackling the computational challenges of polysynthetic languages. These include:

  • Finite State Transducers (FSTs): Used for morphological analysis and generation, particularly for languages like Arapaho and Kwak'wala, demonstrating their effectiveness in capturing complex word structures.
  • Neural Morphological Analyzers: Investigating the application of deep learning models for morphological analysis, showing promise in learning patterns from data.
  • Language Teaching and Learning Software: Development of tools like "Kawennón:nis: the Wordmaker" for Kanyen'kéha (Mohawk), leveraging computational approaches to aid language education.
  • Machine Translation (MT): Analyzing information loss during MT between polysynthetic and fusional languages, highlighting the difficulties in preserving meaning and exploring strategies to mitigate these issues.
  • Automatic Glossing: Developing methods for automatically generating linguistic glosses for low-resource languages, crucial for documentation and analysis.

Datasets

The datasets mentioned or implied in the proceedings include:

  • Arapaho Verb Data: Used for training a neural morphological analyzer.
  • Kwak'wala Data: Utilized for developing a phonological approach to finite-state morphology.
  • Chukchi Data: Employed in the creation of a prototype finite-state morphological analyzer.
  • Kanyen'kéha (Mohawk) Data: Used for developing language teaching and learning software.
  • Nunavut Hansard Data: A significant dataset used for experiments in morphological analysis and machine translation, notable for its size and relevance to polysynthetic language documentation.
  • Mapudungun-Spanish Parallel Corpus: Mentioned in the context of Lori Levin's work, consisting of approximately 200,000 words of text and 120 hours of transcribed speech, used for language technology development.

Key Insights and Ideas

  • The "Word" is Different: Polysynthetic languages challenge the very notion of a "word" as understood in many European languages, requiring computational models to handle much longer and more complex units.
  • Bridging the Gap is Crucial: There's a critical need for collaboration between linguists, computational scientists, and language revitalization practitioners. Each group brings unique expertise, and their synergy is essential for progress.
  • Data is King, but Annotation is Hard: While large datasets like the Nunavut Hansard are valuable, the complex morphology of polysynthetic languages makes annotation a labor-intensive and specialized task.
  • Revitalization Needs Computational Support: Computational tools are not just for academic research; they are vital for the practical goals of language preservation and revitalization, enabling the creation of educational resources and documentation.
  • Information Loss in MT is a Major Hurdle: Translating from polysynthetic languages to others (and vice-versa) often results in a significant loss of nuanced meaning due to the dense information packed into single words.

Key Results

The proceedings showcase a variety of promising results:

  • Successful Morphological Analysis: Papers demonstrate the feasibility of building morphological analyzers for polysynthetic languages using FSTs and neural networks, achieving reasonable accuracy for specific languages.
  • Development of Language Learning Tools: The creation of software like "Kawennón:nis" indicates progress in applying computational linguistics to practical language education.
  • Insights into MT Challenges: Research highlights the extent of information loss in MT for these languages, providing a basis for future research into more robust translation systems.
  • Establishment of a Community: The workshop itself represents a key outcome – the formation of a community dedicated to addressing the computational challenges of polysynthetic languages.

Limitations and Future Work

  • Low-Resource Nature: Many polysynthetic languages are low-resource, meaning limited available data, which poses a significant challenge for data-driven NLP methods.
  • Generalizability: Models developed for one polysynthetic language may not easily transfer to others due to the diversity in their morphological systems.
  • Annotation Bottleneck: The manual effort required for annotating polysynthetic language data remains a major bottleneck for developing and evaluating computational models.
  • Future Work: The authors suggest continued efforts in developing shared tasks to foster collaboration, creating more robust MT systems, improving automatic glossing tools, and building comprehensive language learning resources. There is a strong emphasis on the need for interdisciplinary collaboration and the development of standardized evaluation metrics.

Summary updated

BibTeX
@inproceedings{klavans2018proceedings,
  author = {Judith L. Klavans},
  title = {{Proceedings of the Workshop on Computational Modeling of Polysynthetic Languages}},
  booktitle = {International Conference on Computational Linguistics},
  year = {2018},
}
Figure