Effects of adjective orientation and gradability on sentence subjectivity

Effects of adjective orientation and gradability on sentence subjectivity

Abstract

Subjectivity is a pragmatic, sentence-level feature that has important implications for text processing applications such as information extraction and information retrieval. We study the effects of dynamic adjectives, semantically oriented adjectives, and gradable adjectives on a simple subjectivity classifier, and establish that they are strong predictors of subjectivity. A novel trainable method that statistically combines two indicators of gradability is presented and evaluated, complementing existing automatic techniques for assigning orientation labels.

(Vasileios Hatzivassiloglou, Janyce M. Wiebe)

https://dl.acm.org/doi/10.3115/990820.990864

https://dl.acm.org/doi/pdf/10.3115/990820.990864

https://www.semanticscholar.org/paper/Effects-of-Adjective-Orientation-and-Gradability-on-Hatzivassiloglou-Wiebe/3a8d4fd2c30e5031a574bc25363c8639912b3bbd


The paper "Effects of Adjective Orientation and Gradability on Sentence Subjectivity" by Vasileios Hatzivassiloglou and Janyce M. Wiebe examines how certain properties of adjectives—specifically, their semantic orientation (positive or negative connotation) and gradability (ability to express varying degrees)—influence the subjectivity of sentences. Understanding these relationships is crucial for enhancing text processing applications, such as information extraction and retrieval.

Key Findings:

  1. Semantic Orientation:

    • Adjectives with a clear positive or negative orientation (e.g., "beautiful" vs. "ugly") are strong indicators of subjectivity. The authors developed a method to automatically classify adjectives based on their orientation by analyzing conjunctions (e.g., "and," "but") in large corpora. This method achieved an accuracy of up to 92.37% when sufficient conjunction data was available. ACL Anthology
  2. Gradability:

    • Gradable adjectives, which can express properties in varying degrees (e.g., "cold," "colder," "coldest"), are also strong predictors of subjectivity. The authors introduced a trainable statistical model that combines indicators such as inflection for degree and modification by grading words (e.g., "very," "somewhat") to determine an adjective's gradability. This model achieved an overall accuracy of approximately 88% in classifying adjectives as gradable or non-gradable. ACL Anthology
  3. Dynamic Adjectives:

    • Adjectives that denote qualities subject to control by the possessor (termed "dynamic adjectives," e.g., "careful," "kind") were found to be significant indicators of subjectivity. The presence of such adjectives in a sentence increased the likelihood of the sentence being subjective. ACL Anthology

Implications for Text Processing:

By incorporating knowledge of an adjective's orientation and gradability, computational models can more accurately predict the subjectivity of sentences. This enhanced ability is beneficial for various applications, including:

  • Information Extraction: Distinguishing between factual statements and opinions to extract reliable data.

  • Information Retrieval: Improving search results by considering the subjective or objective nature of content.

  • Sentiment Analysis: Assessing the sentiment expressed in texts, which is valuable for tasks like product review analysis.

The study demonstrates that automatically identified features related to adjective orientation and gradability can be as effective as manually determined ones, suggesting the feasibility of applying these methods across different genres and domains.