Morpheme Induction for Emergent Language

2025-10-03

AI Summary

This paper addresses the unsupervised morpheme induction problem in emerging languages. We propose CSAR, an unsupervised algorithm that leverages parallel semantic–utterance pairs and employs a mutual information-weighted greedy decomposition framework. CSAR iteratively performs counting, candidate selection, and corpus reduction to decouple form–meaning mapping units. To our knowledge, this is the first application of mutual information-driven greedy decomposition to morpheme discovery in emerging languages—offering both interpretability and computational efficiency. Evaluated on synthetic data and real-world emerging language corpora, CSAR significantly outperforms existing baselines. It successfully quantifies core linguistic phenomena—including synonymy and polysemy—demonstrating its capacity to uncover latent semantic structure. Our approach establishes a novel paradigm for structured modeling of low-resource languages, advancing morphological analysis where annotated data are scarce or unavailable. (138 words)

Technology Category

Natural Language Processing: Lexical Semantics and MorphologyMachine Learning: Unsupervised & Self-Supervised LearningCognitive Modeling & Cognitive Systems: Analogy

Application Category

Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semanticsGraph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphsSearch and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

We introduce CSAR, an algorithm for inducing morphemes from emergent language corpora of parallel utterances and meanings. It is a greedy algorithm that (1) weights morphemes based on mutual information between forms and meanings, (2) selects the highest-weighted pair, (3) removes it from the corpus, and (4) repeats the process to induce further morphemes (i.e., Count, Select, Ablate, Repeat). The effectiveness of CSAR is first validated on procedurally generated datasets and compared against baselines for related tasks. Second, we validate CSAR's performance on human language data to show that the algorithm makes reasonable predictions in adjacent domains. Finally, we analyze a handful of emergent languages, quantifying linguistic characteristics like degree of synonymy and polysemy.

Problem

• Inducing morphemes from emergent language corpora
• Validating algorithm performance on human language data
• Analyzing linguistic characteristics like synonymy and polysemy

Innovation

• Greedy algorithm weights morphemes via form-meaning mutual information
• Iteratively selects and removes highest-weighted morpheme pairs
• Validated on procedural datasets and human language data