Semantics NLP Cognitive Linguistics Computational Linguistics

Contextual Polysemy

How meaning emerges, shifts, and resolves through linguistic environment

Contextual polysemy refers to the phenomenon wherein a single lexical item possesses multiple related meanings that are activated or disambiguated based on its immediate syntactic, semantic, and pragmatic context. Unlike homonymy, where meanings are etymologically distinct, polysemous senses share a conceptual core, with context serving as the primary mechanism for sense selection and interpretation.

Definition & Core Principles

Polysemy derives from the Greek poly (many) and sema (sign/meaning). A polysemous word maintains a unified lexical entry in the mental lexicon, but its interpretation is highly sensitive to contextual cues¹. Contextual polysemy emphasizes that meaning is not fixed at the word level but emerges dynamically during processing.

Key Distinction: Polysemy vs. Homonymy

Polysemous senses are conceptually related (e.g., head of a person / head of a department), whereas homonyms are coincidental duplicates with unrelated origins (e.g., bank for financial institution vs. bank of a river).

The resolution of polysemy in context relies on three interacting layers:

Syntactic context: Grammatical roles and collocations constrain possible senses.
Semantic context: Thematic relationships and selectional preferences activate relevant features.
Pragmatic context: World knowledge, discourse goals, and speaker intent guide final interpretation.

Cognitive & Linguistic Mechanisms

Human language processing handles polysemy through rapid, parallel activation of multiple senses, followed by context-driven inhibition of irrelevant meanings². Psycholinguistic studies using eye-tracking and ERP (event-related potentials) show that the brain pre-activates likely senses within 200ms of word onset, with full disambiguation occurring by 400ms.

Metaphorical Extension & Sense Networks

Most polysemous networks radiate from a concrete "basic sense" through metaphorical or metonymic pathways. For example, the verb run extends from physical motion to machinery operation, business management, and even data processing.

Sense Network: Run


Basic: Physical locomotion → 'She run to the store.'
Metaphorical: Machine operation → 'The engine run smoothly.'
Metonymic: Management → 'He run the project.'
Abstract: Data execution → 'The script run successfully.'

Contextual cues (collocates, syntactic frame, domain) suppress irrelevant branches and activate the target sense.

Computational Linguistics & NLP Applications

In natural language processing, contextual polysemy presents the Word Sense Disambiguation (WSD) problem. Traditional rule-based and dictionary-driven methods struggled with scale and ambiguity. The advent of contextual embeddings revolutionized this domain.

Transformer Architectures & Attention

Models like BERT, RoBERTa, and modern LLMs generate dynamic vector representations where a word's embedding shifts based on its surrounding tokens³. Self-attention mechanisms explicitly weight contextual dependencies, allowing the model to "select" the appropriate polysemous sense without explicit lexical resources.

Key computational approaches include:

Contextualized embeddings: Dynamic representations that encode sense information implicitly.
Zero-shot WSD: Leveraging cross-lingual and few-shot prompting to resolve novel polysemous contexts.
Sense induction vs. disambiguation: Clustering contextual vectors to discover emergent senses without predefined inventories.

Illustrative Examples

The following demonstrate how identical surface forms yield distinct interpretations through contextual framing:

Example 1: Light (Adjective/Noun)


Context A: 'The suitcase was surprisingly light.' → Low mass
Context B: 'Turn on the light.' → Illumination
Context C: 'She has a light touch when teaching.' → Gentle manner

Example 2: Cold (Medical/Thermodynamic/Relational)


Context A: 'She caught a cold.' → Viral illness
Context B: 'The room was cold.' → Low temperature
Context C: 'He gave her a cold shoulder.' → Emotional detachment

In each case, collocational patterns, syntactic position, and domain priors trigger distinct cognitive frames, resolving the polysemy without ambiguity for human readers.

References & Further Reading

Gorrell, P. (1997). Lexical Ambiguity and Polysemy in Computational Linguistics. Oxford University Press.
Kreiner, H., & Spalding, T. E. (2010). "The Time Course of Word-Sense Processing: Evidence from Masked Form- and Semantic Priming." Journal of Experimental Psychology: Learning, Memory, and Cognition, 36(5), 1379–1392.
Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding." NAACL.
Lipka, M., & Pilevneli, M. (2023). "Contextual Polysemy in Large Language Models: Emergent Sense Boundaries." Transactions of the ACL, 11, 402–418.
Aevum Encyclopedia Editorial Board. (2024). "Dynamic Semantics & Computational Lexicography." Aevum Knowledge Base.