Verbal Reduplication in Eastern Armenian
Unpacking event structure through reduplication and morphosyntactic form.
Overview
Reduplication in Eastern Armenian is more than repetition—it encodes aspectual nuance, event plurality, and distributive force. This project investigates how verbal reduplication (e.g., կտրել "to cut" vs. կտրտել "to chop repeatedly") reflects rich semantic distinctions that are often ignored or flattened in NLP systems.
By analyzing the morphosyntactic patterns and semantic effects of verbal reduplication, this research aims to contribute to linguistic theory while also enhancing the way AI systems handle lesser-studied morphological processes in underrepresented languages.
Methodology
Our approach combines theoretical linguistics, corpus analysis, and computational modeling:
- Data Collection: Gathering examples from contemporary Eastern Armenian texts, interviews, and media, including literary and colloquial usage
- Semantic Typology: Classifying reduplication types (iterative, distributive, delimitative, etc.) and mapping to event structures
- Syntactic Analysis: Identifying productivity patterns, affix ordering, and morphophonological constraints
- Annotation: Creating a small, high-quality annotated corpus for event structure and reduplication
- Computational Modeling: Designing rule-based and machine learning models to detect and classify reduplicated verbs, with implications for machine translation, morphological analyzers, and LLM fine-tuning
Preliminary Results
- Reduplication interacts with transitivity, Aktionsart, and animacy—showing scalar variation that challenges binary feature classification
- Certain reduplication patterns are lexicalized, while others are productive, requiring context-sensitive disambiguation
- Cross-linguistic parallels (e.g., in Persian or Turkish) suggest potential for a fractal morphosyntax approach—where similar reduplicative principles operate at different levels of granularity
- Existing Armenian NLP tools underperform on reduplicated forms due to lack of training data and morphological segmentation challenges
Use Case
- Morphological analyzers and POS taggers for Armenian benefit from explicit modeling of reduplication
- Machine translation systems can reduce ambiguity and improve fluency when properly handling event plurality
- Language learning tools gain explanatory power when highlighting event-level semantic distinctions
- Supports broader goals in fractal linguistics, typology, and low-resource language modeling
Team
Karine Megerdoomian
Latest publication or presentation
(if available)