We describe our system (DT Team) submitted at SemEval-2017 Task 1, Semantic Textual Similarity (STS) challenge for English (Track 5). We developed three different models with various features including similarity scores calculated using word and chunk alignments, word/sentence embeddings, and Gaussian Mixture Model(GMM). The correlation between our system’s output and the human judgments were up to 0.8536, which is more than 10% above baseline, and almost as good as the best performing system which was at 0.8547 correlation (the difference is just about 0.1%). Also, our system produced leading results when evaluated with a separate STS benchmark dataset. The word alignment and sentence embeddings based features were found to be very effective.

This paper introduces a ruled-based method and software tool, called SemAligner, for aligning chunks across texts in a given pair of short English texts. The tool, based on the top performing method at the Interpretable Short Text Similarity shared task at SemEval 2015, where it was used with human annotated (gold) chunks, can now additionally process plain text-pairs using two powerful chunkers we developed, e.g. using Conditional Random Fields. Besides aligning chunks, the tool automatically assigns semantic relations to the aligned chunks (such as EQUI for equivalent and OPPO for opposite) and semantic similarity scores that measure the strength of the semantic relation between the aligned chunks. Experiments show that SemAligner performs competitively for system generated chunks and that these results are also comparable to results obtained on gold chunks. SemAligner has other capabilities such as handling various input formats and chunkers as well as extending lookup resources.

Negation is often found more frequent in dialogue than commonly written texts, such as literary texts. Furthermore, the scope and focus of negation depends on context in dialogues than other forms of texts. Existing negation datasets have focused on non-dialogue texts such as literary texts where the scope and focus of negation is normally present within the same sentence where the negation is located and therefore are not the most appropriate to inform the development of negation handling algorithms for dialogue-based systems. In this paper, we present DT -Neg corpus (DeepTutor Negation corpus) which contains texts extracted from tutorial dialogues where students interacted with an Intelligent Tutoring System (ITS) to solve conceptual physics problems. The DT -Neg corpus contains annotated negations in student responses with scope and focus marked based on the context of the dialogue. Our dataset contains 1,088 instances and is available for research purposes at http://language.memphis.edu/dt-neg.

We analyze in this paper a number of data sets proposed over the last decade or so for the task of paraphrase identification. The goal of the analysis is to identify the advantages as well as shortcomings of the previously proposed data sets. Based on the analysis, we then make recommendations about how to improve the process of creating and using such data sets for evaluating in the future approaches to the task of paraphrase identification or the more general task of semantic similarity. The recommendations are meant to improve our understanding of what a paraphrase is, offer a more fair ground for comparing approaches, increase the diversity of actual linguistic phenomena that future data sets will cover, and offer ways to improve our understanding of the contributions of various modules or approaches proposed for solving the task of paraphrase identification or similar tasks.

We describe the DARE corpus, an annotated data set focusing on pronoun resolution in tutorial dialogue. Although data sets for general purpose anaphora resolution exist, they are not suitable for dialogue based Intelligent Tutoring Systems. To the best of our knowledge, no data set is currently available for pronoun resolution in dialogue based intelligent tutoring systems. The described DARE corpus consists of 1,000 annotated pronoun instances collected from conversations between high-school students and the intelligent tutoring system DeepTutor. The data set is publicly available.

This paper introduces a collection of freely available Latent Semantic Analysis models built on the entire English Wikipedia and the TASA corpus. The models differ not only on their source, Wikipedia versus TASA, but also on the linguistic items they focus on: all words, content-words, nouns-verbs, and main concepts. Generating such models from large datasets (e.g. Wikipedia), that can provide a large coverage for the actual vocabulary in use, is computationally challenging, which is the reason why large LSA models are rarely available. Our experiments show that for the task of word-to-word similarity, the scores assigned by these models are strongly correlated with human judgment, outperforming many other frequently used measures, and comparable to the state of the art.