Presentation is loading. Please wait.

Presentation is loading. Please wait.

Learning Subjective Adjectives from Corpora Janyce M. Wiebe Presenter: Gabriel Nicolae.

Similar presentations


Presentation on theme: "Learning Subjective Adjectives from Corpora Janyce M. Wiebe Presenter: Gabriel Nicolae."— Presentation transcript:

1 Learning Subjective Adjectives from Corpora Janyce M. Wiebe Presenter: Gabriel Nicolae

2 Introduction Subjectivity in natural language refers to aspects of language used to express opinions and evaluations (Banfield 1982; Wiebe 1994) Subjectivity tagging is distinguishing sentences used to present opinions and other forms of subjectivity (subjective sentences) from sentences used to objectively present factual information (objective sentences).

3 Why do we need subjectivity tagging? Because: Apart from the subject of the document, additional components influence its relevance: the evidential status of the material presented attitudes adopted (Hatzivassiloglou) The task is especially relevant for: News reporting Internet forums – recognizing flames Applications for which it is relevant: Information extraction Information retrieval

4 Subjectivity – Examples (1/2) Simple subjective sentence: Simple objective sentence: At several different layers, it’s a fascinating tale. Bell Industries Inc. increased its quarterly to 10 cents from 7 cents a share.

5 Subjectivity – Examples (2/2) Subjective sentence about a speech event: Objective sentence about a speech event: “The cost of health care is eroding our standard of living and sapping industrial strength,” complains Walter Maher, a Chrysler health-and- benefits specialist. Northwest Airlines settled the remaining lawsuits filed on behalf of 156 people killed in a 1987 crash, but claims against the jetliner’s maker are being pursued, a federal judge said.

6 Aspects of subjectivity expressions (1/5) There are expressions subjective in all contexts: ! But many are subjective depending on the context: sapping, eroding A potential subjective element is a linguistic element that may be used to express subjectivity. A subjective element is an instance of a potential subjective element, in a particular context, that is indeed subjective in that context. (Wiebe 1994)

7 Aspects of subjectivity expressions (2/5) There are different types of subjectivity, and the work focuses on three: positive evaluation (e.g. fascinating) negative evaluation (e.g. terrible) speculation (e.g. probably)

8 Aspects of subjectivity expressions (3/5) A subjective element expresses the subjectivity of a source. Source = writer or someone mentioned in text. At several different layers, it’s a fascinating tale. Source = writer. “The cost of health care is eroding our standard of living and sapping industrial strength,” complains Walter Maher, a Chrysler health-and- benefits specialist. Source = Maher.

9 Aspects of subjectivity expressions (4/5) A subjective element also has a target. Target = what the subjectivity is all about or directed toward. At several different layers, it’s a fascinating tale. Target = a tale. “The cost of health care is eroding our standard of living and sapping industrial strength,” complains Walter Maher, a Chrysler health-and- benefits specialist. Target = the cost of health care.

10 Aspects of subjectivity expressions (5/5) The former examples have object-centric subjectivity. Other examples: Subjectivity may also be addressee-oriented (directed towards the listener and reader) I love this project. The software is horrible. You are an idiot.

11 Experiments Corpus: 1,001 sentences of the Wall Street Journal Treebank Corpus (Marcu et al. 1993) manually annotated with subjectivity classifications in addition: subjective elements & strength of elements (on a scale of 1 to 3)

12 Improving Adjective Features Using Distributional Similarity (1/2) Intuition: words correlated with many of the same things in text are more similar. Challenging test: 10-fold cross validation 1/10 training 9/10 testing For each training set i Extract all adjectives from subjective elements of strength 3 For each adjective Identify top 20 entries in a similarity thesaurus (Lin 1994) These are the seed sets for fold i. Evaluate these seed sets on the remaining 9/10 of the corpus.

13 Improving Adjective Features Using Distributional Similarity (2/2) Baseline: the precision of a simple adjective feature (= the conditional probability that a sentence is subjective, given that at least one adjective appears). Average precision: 55.8% Above-mentioned process: Average precision: 61.2%  Increase: 5.4% Repeat experiment with similarities from WordNet. Average precision: 62.0%  Slight increase, but lower coverage.

14 Refinements with Polarity and Gradability (1/2) Polarity: presented in previous paper. Gradability: the semantic property that enables a word to participate in comparative constructs and to accept modifying expressions that act as intensifiers and diminishers. Gradable adjectives express properties in varying degrees of strength, relative to a norm (explicitly or implicitly supplied by the modified noun) (Hatzivassiloglou) list of 73 adverbs and NPs that are frequently used as grading modifiers. a small planet – a large house a little, exceedingly, somewhat, very

15 Refinements with Polarity and Gradability (2/2) The work uses samples of adjectives identified as: having positive polarity having negative polarity being gradable Samples were determined using a new corpus from the Wall Street Journal.

16 Results and Discussion Experiments for automatic/manual identification of polarity +, -, +-, and gradable adjectives. Promising results: In all cases, the average improvement over the baseline of the intersection btw. seed sets and gradability/polarity sets is at least 9%. The gradability/polarity sets and the seed sets are more precise together than alone. Excellent individual results: gradability/automatic and polarity-/automatic sets intersected with the seed sets. Future work: Some of the data that is currently part of the test set could be used to filter the sets (3/10 of data used for training and 1/3 of training data used for seeding, and 2/3 for filtering)


Download ppt "Learning Subjective Adjectives from Corpora Janyce M. Wiebe Presenter: Gabriel Nicolae."

Similar presentations


Ads by Google