Multi-task learning by using contextualized word representations for syntactic parsing of a morphologically rich...

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Urdu Language Processing

Background:

• Syntactic parsing for morphologically rich languages like Urdu presents significant challenges.
• Existing Urdu treebanks require conversion for dependency parsing tasks.
• Developing effective parsing frameworks for Urdu is crucial for advancing NLP research.

Purpose of the Study:

• To achieve state-of-the-art results in both constituency and dependency parsing for Urdu.
• To introduce a novel sequence labeling scheme for a unified parsing representation.
• To explore the benefits of single-task and multi-task learning paradigms for Urdu parsing.

Main Methods:

• Conversion of the CLE-UTB phrase structure treebank into a dependency treebank using language-specific mapping rules.
• Development of a novel sequence labeling scheme to unify parsing tasks.
• Training contextualized word representations on a large 220 million token Urdu corpus.
• Implementation of a parsing framework utilizing single-task and multi-task learning.

Main Results:

• Achieved an F1 score of 91.39 for constituency parsing, an improvement of 3.29 points.
• Obtained a labeled attachment score of 85.69 for dependency parsing, an improvement of 1.49 points.
• Demonstrated significant performance enhancement through multi-task learning.

Conclusions:

• The proposed sequence labeling scheme and multi-task learning approach effectively improve Urdu syntactic parsing.
• Learning cross-task representations offers measurable benefits for parsing morphologically rich languages.
• This work advances the state-of-the-art in Urdu constituency and dependency parsing.