[Paper Review] Identifying Metaphoric Antonyms in a Corpus Analysis of Finance Articles

Aaron Gerow, Mark T. Keane|arXiv (Cornell University)|Dec 13, 2012

Language, Metaphor, and Cognition10 references3 citations

TL;DR

This study investigates metaphoric antonyms in financial news by analyzing corpus data of 17,000+ articles (10M+ words), using distributional similarity of verb arguments to predict human-identified antonyms for 'UP' and 'DOWN' verbs. It finds that cosine similarity in argument distributions predicts the most frequent antonym 87% of the time, supporting distributional models for metaphor comprehension.

ABSTRACT

Using a corpus of 17,000+ financial news reports (involving over 10M words), we perform an analysis of the argument-distributions of the UP and DOWN verbs used to describe movements of indices, stocks and shares. In Study 1 participants identified antonyms of these verbs in a free-response task and a matching task from which the most commonly identified antonyms were compiled. In Study 2, we determined whether the argument-distributions for the verbs in these antonym-pairs were sufficiently similar to predict the most frequently-identified antonym. Cosine similarity correlates moderately with the proportions of antonym-pairs identified by people (r = 0.31). More impressively, 87% of the time the most frequently-identified antonym is either the first- or second-most similar pair in the set of alternatives. The implications of these results for distributional approaches to determining metaphoric knowledge are discussed.

Motivation & Objective

• To investigate how people identify metaphoric antonyms in financial language, particularly for verbs like 'UP' and 'DOWN' describing market movements.
• To examine whether distributional similarity in argument structures can predict human-identified antonym pairs.
• To evaluate the effectiveness of distributional semantics in modeling metaphorical knowledge in domain-specific language.
• To contribute empirical evidence on the cognitive plausibility of distributional approaches to metaphor.

Proposed method

• Analyzed a corpus of 17,000+ financial news articles containing over 10 million words to extract argument distributions of 'UP' and 'DOWN' verbs.
• Conducted two studies: a free-response and matching task with participants to collect human-identified antonyms.
• Calculated cosine similarity between argument distributions of verb pairs to assess semantic similarity.
• Used the similarity scores to rank potential antonym pairs and compare predictions with human responses.
• Evaluated the predictive power of distributional similarity by measuring correlation and accuracy of top-ranked antonyms.
• Applied the results to assess the viability of distributional models for capturing metaphorical knowledge in language.

Experimental results

Research questions

• RQ1 Which verbs are most commonly identified by humans as metaphoric antonyms of 'UP' and 'DOWN' in financial contexts?
• RQ2 To what extent can distributional similarity in argument structures predict human-identified metaphoric antonym pairs?
• RQ3 How well does cosine similarity between verb argument distributions correlate with the frequency of human-identified antonyms?
• RQ4 Can distributional models reliably predict the most frequently selected antonym among multiple alternatives?

Key findings

• Cosine similarity between argument distributions showed a moderate correlation (r = 0.31) with the proportion of human-identified antonym pairs.
• In 87% of cases, the most frequently identified antonym was among the first or second most similar verb pairs based on argument distribution.
• The most commonly identified antonyms were consistently predicted by the top-ranked distributional similarities, indicating strong predictive power.
• The results support the use of distributional semantics as a viable approach for modeling metaphorical knowledge in domain-specific language.
• The study demonstrates that argument distribution similarity is a robust indicator of metaphorical antonymy in financial discourse.