Quantifying the cross-linguistic effects of syncretism on agreement attraction

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Social Sciences & Behavioral Studies, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

A study investigates agreement attraction errors, where verbs incorrectly agree with an intervening noun instead of the grammatical head, and how these errors are influenced by morphological syncretism. This phenomenon amplifies errors in languages like English, German, and Russian, but not in Turkish or Armenian, a pattern previously lacking a principled explanation. Researchers utilized surprisal and attention entropy derived from large language models (LLMs) as proxies for language processing to analyze this cross-linguistic variation across four languages. The LLM-derived measures successfully replicated known behavioral findings in English and German, showing syncretism modulates attraction. They also aligned with the null results observed in Turkish, indicating no modulation, and partially captured the patterns found in Russian.

Key takeaway

For research scientists modeling psycholinguistic phenomena, this work demonstrates that LLM-derived metrics like surprisal and attention entropy offer a robust tool for understanding complex cross-linguistic variations. You should consider integrating these computational proxies to investigate how morphological features influence human language processing, potentially revealing underlying mechanisms that explain observed behavioral differences across languages. This approach can enhance the predictive power of your models.

Key insights

Large language models' surprisal and attention entropy can model cross-linguistic agreement attraction modulated by morphological syncretism.

Principles

• Morphological syncretism amplifies agreement attraction errors in some languages.
• LLM-derived metrics can serve as proxies for human language processing.
• Cross-linguistic patterns in agreement attraction vary significantly.

Method

The study used surprisal and attention entropy from large language models to investigate how morphological syncretism affects agreement attraction errors across English, German, Turkish, and Russian.

In practice

• Apply LLM surprisal to predict human processing difficulties.
• Use attention entropy for cross-linguistic linguistic analysis.

Topics

• Agreement Attraction
• Morphological Syncretism
• Large Language Models
• Surprisal
• Attention Entropy
• Psycholinguistics
• Cross-linguistic Analysis

Best for: NLP Engineer, AI Scientist, Research Scientist

Related on AIssential

• Probing Semantic Alignment, Lexical Invariance, and Syntactic Influence in LLM Metaphor Processing — LLMs struggle with deep metaphor comprehension, often misinterpreting concepts and relying on lexical associations over context or full syntactic understanding.
• Agent-based models for the evolution of morphological alternation patterns — Agent-based simulations, evaluated by an LLM-driven "AI Historical Linguist," model the emergence and spread of morphological alternations in language.
• Probing for Reading Times — Early language model layers predict initial human reading effort better than scalar surprisal, which excels at later stages.
• Differences in Typological Alignment in Language Models' Treatment of Differential Argument Marking — LMs show human-like markedness preferences but not object preference in differential argument marking.
• Fully Automated Identification of Lexical Alignment and Preference-Stage Shifts in Large Language Models — Automated metrics can identify and quantify lexical overuse and preference learning shifts in LLMs without manual curation.
• Do Language Models Know Theo Has a Wife? Investigating the Proviso Problem — LLMs align with human judgments on the proviso problem but use shallow pattern matching, not deep pragmatic reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.