Intelligent Tutor to Support Teaching and Learning of Tatar

Alsu Zakirova, Jue Hou, Anisia Katinskaia, Anh-Duc Vu, Roman YangarberACL 2024

TL;DR

This paper introduces Revita, a web-based intelligent tutoring system adapted for teaching and learning the Tatar language. By leveraging natural language processing and construct-centered learning, Revita automatically generates personalized exercises from authentic texts, aiming to improve fluency for both heritage speakers and L2 learners of this low-resource language.

What Problem Does This Paper Address?

Learning a new language, especially a low-resource one like Tatar, presents significant challenges. While many existing tools offer basic introductions or focus on rote memorization, there's a notable gap in interactive, personalized learning platforms that adapt to individual interests and proficiency levels. Tatar, spoken by over seven million people, is technologically underserved, limiting its online presence and overall vitality. This paper addresses the need for advanced L2 learning tools for Tatar, particularly for learners beyond the beginner stage, by adapting an existing intelligent tutoring system (Revita) to its specific linguistic features.

Methods

The core of Revita's approach is "construct-centered learning." Instead of focusing on isolated words or grammar rules, it breaks down language into "constructs" – linguistic units like phrases, clauses, or grammatical patterns. The system uses Natural Language Processing (NLP) to analyze authentic texts uploaded by users (teachers or learners). Based on this analysis, Revita automatically generates various types of exercises (e.g., fill-in-the-gap, multiple-choice, listening) tailored to the identified constructs within the text. A key feature is its adaptive assessment, which continuously monitors learner performance to adjust exercise difficulty and provide personalized, graduated feedback. The system also utilizes a "heatmap" visualization to show learners and teachers which constructs have been practiced and how well they are mastered.

Datasets

The paper does not explicitly mention specific datasets used for training or evaluation. However, it emphasizes the use of "arbitrary authentic texts" chosen by the learners or teachers as the primary learning material. The system's constructs are based on existing linguistic descriptions of the Tatar language, including grammars and rule inventories.

Key Insights and Ideas

  • Construct-Centered Learning for Low-Resource Languages: The paper demonstrates the feasibility of applying a sophisticated L2 learning methodology (construct-centered) to a low-resource language like Tatar, which is crucial for language preservation and revitalization.
  • Personalized Learning from Authentic Content: Allowing learners to choose their own texts makes the learning process more engaging and relevant, directly addressing their interests and needs.
  • Automated Exercise Generation: The system's ability to automatically generate diverse exercises from any text significantly reduces the burden on educators and provides a scalable solution for language learning.
  • Graduated Feedback for Deeper Understanding: Providing hints that progressively reveal the correct answer, rather than just stating it, helps learners understand why they made a mistake, fostering deeper comprehension.
  • Visualizing Progress with Heatmaps: The heatmap feature offers a unique and intuitive way for both learners and teachers to track progress and identify areas needing more attention.

Key Results

The paper details the implementation of various Tatar constructs within the Revita system, including:

  • Basic Constructs: Declension of nouns, degrees of comparison of adjectives, and tenses of verbs.
  • Complex Constructions: Post-positional constructions (67 implemented), verbal constructions (344 implemented), and modality constructs (18 implemented). The system has been adapted to analyze Tatar texts and generate exercises based on these constructs. While quantitative results on learner improvement are not presented, the paper emphasizes the system's functionality and its potential to enhance Tatar language learning for intermediate to advanced learners.

Limitations and Future Work

The authors acknowledge several limitations and outline future directions:

  • Morphological Analysis Quality: The current reliance on the Apertium morphological analyzer for Tatar is noted as a bottleneck. While it partially meets needs, its ongoing development is crucial for improving exercise quality.
  • Lack of Foundational Resources: The scarcity of foundational NLP resources for Turkic languages, including Tatar, is a significant challenge that limits the quality of downstream applications.
  • Syntactic Dependency Parser: The absence of a robust syntactic dependency parser for Tatar is a major limitation. Its development is seen as a key step to significantly enhance the system's analysis capabilities and exercise generation.
  • Expanding Text Corpus: The ongoing work aims to build a more comprehensive public library of Tatar texts to provide learners with more choices and to facilitate the training of AI models.
  • More Complex Syntactic Constructs: Future development will focus on incorporating more complex syntactic structures, particularly synthetic subordinate clauses.
  • Deployment and Global Impact: The ultimate goal is to deploy the Tatar Revita platform for widespread use in teaching and learning, contributing to the preservation and global interest in the Tatar language.

Summary updated

BibTeX
@inproceedings{zakirova2024intelligent,
  author = {Alsu Zakirova and Jue Hou and Anisia Katinskaia and Anh-Duc Vu and Roman Yangarber},
  title = {{Intelligent Tutor to Support Teaching and Learning of Tatar}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2024},
}
Figure