Why does a language acquisition model relying solely on distributional semantics for dictionary creation struggle with polysemous words, even with a large corpus?

💡 Explanation

Distributional semantics represents words by their context. A single vector for a polysemous word blurs the distinct senses because the mechanism of vector averaging merges different usages into one representation; therefore, the model fails to differentiate senses, rather than because of optimization failures.

• Why does speech recognition accuracy decline when input audio contains a poem recited with unnatural meter?
• Why does mapping the abstract 'argument' domain onto the 'war' conceptual frame often fail to achieve persuasion within a target audience?
• Why does formulaic language acquisition in second language learners often plateau despite continued explicit instruction?
• If speakers of a plantation creole are abruptly integrated into the dominant society, which consequence follows for the creole?
• A translator adapts a technical manual from Japanese to English for a US audience; which pragmatic effect dominates?
• Why does phonetic borrowing in signed languages occur more readily between geographically proximate signed languages than between a signed and spoken language?