[Paper Review] Learning Unification-Based Natural Language Grammars

Miles Osborne|arXiv (Cornell University)|Feb 3, 1995

Natural Language Processing Techniques89 references3 citations

TL;DR

This paper proposes a hybrid learning approach that combines data-driven and model-based methods to correct undergeneration in unification-based natural language grammars. By integrating data from the Spoken English Corpus with a linguistic model, the system reduces both undergeneration and overgeneration, producing grammars that generate linguistically plausible parses with improved coverage and soundness.

ABSTRACT

When parsing unrestricted language, wide-covering grammars often undergenerate. Undergeneration can be tackled either by sentence correction, or by grammar correction. This thesis concentrates upon automatic grammar correction (or machine learning of grammar) as a solution to the problem of undergeneration. Broadly speaking, grammar correction approaches can be classified as being either {\it data-driven}, or {\it model-based}. Data-driven learners use data-intensive methods to acquire grammar. They typically use grammar formalisms unsuited to the needs of practical text processing and cannot guarantee that the resulting grammar is adequate for subsequent semantic interpretation. That is, data-driven learners acquire grammars that generate strings that humans would judge to be grammatically ill-formed (they {\it overgenerate}) and fail to assign linguistically plausible parses. Model-based learners are knowledge-intensive and are reliant for success upon the completeness of a {\it model of grammaticality}. But in practice, the model will be incomplete. Given that in this thesis we deal with undergeneration by learning, we hypothesise that the combined use of data-driven and model-based learning would allow data-driven learning to compensate for model-based learning's incompleteness, whilst model-based learning would compensate for data-driven learning's unsoundness. We describe a system that we have used to test the hypothesis empirically. The system combines data-driven and model-based learning to acquire unification-based grammars that are more suitable for practical text parsing. Using the Spoken English Corpus as data, and by quantitatively measuring undergeneration, overgeneration and parse plausibility, we show that this hypothesis is correct.

Motivation & Objective

• Address undergeneration in wide-covering grammars, where grammars fail to parse valid sentences.
• Overcome limitations of purely data-driven learners, which often overgenerate ill-formed sentences.
• Mitigate incompleteness in model-based learners by using data to fill gaps in linguistic knowledge.
• Develop a system that balances data-driven learning (for coverage) and model-based learning (for soundness).
• Create a grammar learning framework suitable for practical text processing with reliable semantic interpretation.

Proposed method

• Integrate data-driven learning using the Spoken English Corpus to identify missing syntactic patterns.
• Apply model-based learning with a predefined linguistic model of grammaticality to constrain grammar rules.
• Use unification-based formalism to represent grammatical constraints and feature structures.
• Combine both approaches in a hybrid learning architecture where data corrects model gaps and the model filters noisy data.
• Employ quantitative metrics to evaluate undergeneration, overgeneration, and parse plausibility.
• Iteratively refine the grammar using feedback from parsing results and linguistic plausibility judgments.

Experimental results

Research questions

• RQ1 Can combining data-driven and model-based learning reduce undergeneration in unification-based grammars?
• RQ2 Does the hybrid approach prevent overgeneration compared to purely data-driven methods?
• RQ3 To what extent does the model-based component improve the linguistic plausibility of parses?
• RQ4 How effective is the system in achieving both broad coverage and grammatical soundness?
• RQ5 Can the system be empirically validated using real-world corpus data and quantitative metrics?

Key findings

• The hybrid system significantly reduces undergeneration compared to model-based learning alone.
• Overgeneration is minimized because the model-based component filters out ungrammatical structures identified by data-driven learning.
• Parse plausibility improves, as the resulting grammars produce linguistically acceptable analyses for a broader range of sentences.
• The system achieves better balance between coverage and soundness than either data-driven or model-based methods used in isolation.
• Quantitative evaluation using the Spoken English Corpus confirms that the hybrid approach outperforms baseline methods on all key metrics.
• The approach demonstrates feasibility for practical text parsing systems requiring both wide coverage and linguistic accuracy.