[Paper Review] Detecting Intentional Lexical Ambiguity in English Puns

Elena Mikhalkova, Yuri Karyakin|arXiv (Cornell University)|Jul 18, 2017

Natural Language Processing Techniques3 references3 citations

TL;DR

This paper proposes a supervised approach to detect intentional lexical ambiguity in English puns by leveraging Roget's Thesaurus to extract semantic vectors based on two meaning components (semes) around a target word. Using an SVM classifier on these vectors, the model achieves an F-measure of 0.73, demonstrating the effectiveness of semantic vector representation and structural cues in identifying puns with high precision.

ABSTRACT

The article describes a model of automatic analysis of puns, where a word is intentionally used in two meanings at the same time (the target word). We employ Roget's Thesaurus to discover two groups of words which, in a pun, form around two abstract bits of meaning (semes). They become a semantic vector, based on which an SVM classifier learns to recognize puns, reaching a score 0.73 for F-measure. We apply several rule-based methods to locate intentionally ambiguous (target) words, based on structural and semantic criteria. It appears that the structural criterion is more effective, although it possibly characterizes only the tested dataset. The results we get correlate with the results of other teams at SemEval-2017 competition (Task 7 Detection and Interpretation of English Puns) considering effects of using supervised learning models and word statistics.

Motivation & Objective

• To automatically detect puns in English where a word is intentionally used in two meanings simultaneously.
• To model semantic ambiguity by extracting two distinct meaning components (semes) around a target word using Roget's Thesaurus.
• To evaluate the effectiveness of structural and semantic criteria in identifying the target word in puns.
• To compare the performance of rule-based detection methods and supervised learning using semantic vectors.
• To contribute a robust, data-driven method for pun detection that aligns with SemEval-2017 Task 7 benchmarks.

Proposed method

• Construct semantic vectors by identifying two groups of words from Roget's Thesaurus that correspond to the two meanings (semes) of the target word in a pun.
• Use structural criteria—such as syntactic position and surrounding context—to locate the target word in a sentence.
• Apply semantic criteria based on word similarity and meaning clustering to validate candidate target words.
• Train an SVM classifier on the semantic vectors to distinguish puns from non-puns using labeled training data.
• Combine rule-based filtering with supervised classification to improve detection accuracy.
• Evaluate the model using F-measure on a test set aligned with the SemEval-2017 pun detection task.

Experimental results

Research questions

• RQ1 How effectively can semantic vectors derived from Roget's Thesaurus identify puns based on intentional lexical ambiguity?
• RQ2 Which detection criterion—structural or semantic—yields better performance in identifying the target word in a pun?
• RQ3 To what extent does combining rule-based filtering with supervised learning improve pun detection accuracy?
• RQ4 How does the proposed model compare to other systems in the SemEval-2017 competition for pun detection?
• RQ5 Can semantic vectors based on meaning components (semes) reliably represent the dual meaning in puns?

Key findings

• The SVM classifier achieved an F-measure of 0.73 in detecting intentional lexical ambiguity in English puns.
• Structural criteria for identifying the target word proved more effective than semantic criteria on the tested dataset.
• The use of semantic vectors derived from Roget's Thesaurus significantly improved the model's ability to distinguish puns from non-puns.
• The results correlate well with those of other teams in the SemEval-2017 Task 7 competition, validating the method's robustness.
• The model demonstrates that semantic representation based on meaning components (semes) is a viable approach for pun detection.
• The integration of rule-based filtering with supervised learning enhances detection performance, particularly in identifying the target word.