Training Data Size Sensitivity in Unsupervised Rhyme Recognition

2026-04-09 · Source: Computation and Language · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, quick

Summary

RhymeTagger, a language-independent tool for unsupervised rhyme recognition, was evaluated across seven languages: Czech, German, English, French, Italian, Russian, and Slovene. The study investigated the impact of training data size and language differences on its accuracy. To establish a performance baseline, inter-annotator agreement was assessed on a manually annotated poem subset, revealing factors like phonetic similarity and word distance influencing expert disagreement. RhymeTagger's performance was also compared against three large language models (LLMs) using a one-shot learning approach. The research found that with sufficient training data, RhymeTagger consistently surpassed human agreement levels, whereas LLMs, due to their lack of phonetic representation, performed poorly on the task.

Key takeaway

For research scientists developing natural language processing tools for poetic analysis, this study indicates that RhymeTagger offers a robust, language-independent solution for rhyme recognition. You should consider integrating phonetic representations into your models, as LLMs without this capability significantly underperform. This approach can lead to more accurate and reliable automated literary analysis.

Key insights

RhymeTagger excels at unsupervised rhyme recognition, outperforming humans and LLMs when adequately trained.

Principles

• Rhyme classification is historically constructed and subjective.
• Phonetic representation is crucial for accurate rhyme recognition.

Method

RhymeTagger identifies rhymes by detecting repeating patterns in poetry corpora. Its performance was evaluated against human agreement and LLMs using one-shot learning across multiple languages.

In practice

• Use RhymeTagger for multilingual rhyme analysis.
• Prioritize phonetic features in rhyme recognition models.

Topics

• Unsupervised Rhyme Recognition
• RhymeTagger
• Training Data Sensitivity
• Multilingual Poetry Corpora
• Large Language Models

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• How I built a rhyme quality scoring algorithm for the French language — Quantifying rhyme quality requires phonemic analysis and articulatory proximity matrices, not just letter matching.
• TantraTagger: A Benchmark Dataset for Tantrayukti-Based Discourse Structure Labelling in Sanskrit Śāstra Texts — TantraTagger is a benchmark dataset for discourse structure labelling in Sanskrit Śāstra Texts.
• Detecting HIV-Related Stigma in Clinical Narratives Using Large Language Models — LLMs can effectively detect HIV-related stigma in clinical notes, with encoder models outperforming generative models in zero-shot settings.
• RUMLEM: A Dictionary-Based Lemmatizer for Romansh — A dictionary-based lemmatizer for Romansh addresses NLP challenges for less-resourced languages.
• JSPG: Dynamic Dictionary Filtering via Joint Semantic-Pinyin-Glyph Retrieval for Chinese Contextual ASR — Jointly integrating semantic, pinyin, and glyph features significantly improves Chinese contextual ASR keyword retrieval accuracy.
• A Good Part-of-Speech Tagger in about 200 Lines of Python — Practical NLP knowledge can be distilled into concise, confident, and accessible implementations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computation and Language.