Beata Beigman Klebanov

Different Texts, Same Metaphors: Unigrams and Beyond

Current approaches to supervised learning of metaphor tend to use sophisticated features and restrict their attention to constructions and contexts where these features apply. In this paper, we describe the development of a supervised learning system to classify all content words in a running text as either being used metaphorically or not. We start by examining the performance of a simple unigram baseline that achieves surprisingly good results for some of the datasets. We then show how the recall of the system can be improved over this strong baseline.

Supervised Word-Level Metaphor Detection: Experiments with Concreteness and Reweighting of Examples

We present a supervised machine learning system for word-level classification of all content words in a running text as being metaphorical or non-metaphorical. The system provides a substantial improvement upon a previously published baseline, using re-weighting of the training examples and using features derived from a concreteness database. We observe that while the first manipulation was very effective, the second was only slightly so. Possible reasons for these observations are discussed.

Sentiment profiles of multiword expressions in test-taker essays: The case of noun-noun compounds

The property of idiomaticity vs. compositionality of a multiword expression traditionally pertains to the semantic interpretation of the expression. In this article, we consider this property as it applies to the expressions sentiment profile (relative degree of positivity, negativity, and neutrality). Thus, while heart attack is idiomatic in terms of semantic interpretation, the sentiment profile of the expression (strongly negative) can, in fact, be determined from the strongly negative profile of the head word. In this article, we (1) propose a way to measure compositionality of a multiword expressions sentiment profile, and perform the measurement on noun-noun compounds; (2) evaluate the utility of using sentiment profiles of noun-noun compounds in a sentence-level sentiment classification task. We find that the sentiment profiles of noun-noun compounds in test-taker essays tend to be highly compositional and that their incorporation improves the performance of a sentiment classification system.

Advancing the validity argument for standardized writing tests using quantitative rhetorical analysis

Essay writing is a common type of constructed-response task used frequently in standardized writing assessments. However, the impromptu timed nature of the essay writing tests has drawn increasing criticism for the lack of authenticity for real-world writing in classroom and workplace settings. The goal of this paper is to contribute evidence to a validity argument for standardized writing tests. Using measurements of distances between rhetorical profiles in the corpora of interest, we examined connections between argumentative writing on standardized assessments and in external writing situations; namely, opinionated writing in academic and real-life settings. The results show that test corpora, focusing on argumentation in two standardized tests, are rhetorically similar to academic argumentative writing in a graduate-school setting, and about as similar as a corpus of civic writing in the same genre. The proximity between the test corpora and corpora representing external criteria of interest support the assessment use argument. The argumentative writing skills employed on the test are similar to the skills employed in academic and civic settings, despite the differences in the nature of the settings under which the writing samples for these different corpora are produced.

A Left Turn: Automated Feedback and Activity Generation for Student Writers

Writing is key to educational and workplace success. The majority of computer-based writing support applications target K-12 students and offer standard feedback concerning the U.S.centric, 5-paragraph, expository essay genre, which is typically taught to developing writers and used on standardized tests. This writing demo takes a “left turn”. By “left turn”, we mean that this demo is illustrating a new writing support concept. In contrast to most systems that provide only feedback, this demo illustrates how guided activities can be automatically generated as a complement to relevant writing feedback driven by natural language processing (NLP) methods. Further, the demo is intended to handle a wide range of writing genres in postsecondary academic settings. The demo feedback and activities illustrated in this paper are ideas that are intended to promote postsecondary students’ engagement with writing assignments, and address writing domain knowledge that is critical for producing higher quality postsecondary writing assignments across writing genres and academic disciplines.