Search

We describe the second IWPT task on end-to-end parsing from raw text to Enhanced Universal Dependencies. We provide details about the evaluation metrics and the datasets used for training and evaluation. We compare the approaches taken by participating teams and discuss the results of the shared task, also in comparison with the first edition of this task.

Modern computer-algebra programs are able to solve a wide range of mathematical calculations. However, they are not able to understand and solve math text problems in which the equation is described in terms of natural language instead of mathematical formulas. Interestingly, there are only few known approaches to solve math word problems algorithmically and most of employ models based on frames. To overcome problems with existing models, we propose a model based on augmented semantic networks to represent the mathematical structure behind word problems. This model is implemented in our Solver for Mathematical Text Problems (SoMaTePs) [1], where the math problem is extracted via natural language processing, transformed in mathematical equations and solved by a state-of-the-art computer-algebra program. SoMaTePs is able to understand and solve mathematical text problems from German primary school books and could be extended to other languages by exchanging the language model in the natural language processing module.

In this chapter, we describe our efforts in text-to-text generation within the IMOGEN project. In particular, we describe two focus areas of research to improve the quality of the answer: (a) graph-based content selection to improve the answer in terms of usefulness, and (b) sentence fusion to improve the answer in terms of formulation. We use sentence fusion to join together multiple sentences in order to eliminate overlapping parts, thereby reducing redundancy. The results of this work have been applied in the IMIX system. This system uses a question answering system to pinpoint fragments of text which are relevant to the information need expressed by the user. A content selection system then uses these fragments as entry points in the text to formulate a more complete answer. Sentence fusion is applied to manipulate the result in order to increase the fluency of the text.

This chapter describes the Vidiam project, which covered the development of a dialogue management system for multimodal question answering (QA) dialogues, as carried out in the IMIX project. The approach followed was datadriven, i.e., corpus-based. Since research in QA dialogue of multimodal information retrieval is still new, no suitable corpora were available to base a system on. This chapter reports on the collection and analysis of three QA dialogue corpora, involving textual follow-up utterances, multimodal follow-up questions, and speech dialogues. Based on the data, a dialogue act typology was created, which helps translate user utterances to practical interactive QA strategies.

The current endeavour focuses on the notion of positive versus negative polarity preference of verbs for their direct objects. This preference has to be distinguished from a verb's own prior polarity - for the same verb, these two properties might even be inverse. Polarity preferences of verbs are extracted on the basis of a large and dependency-parsed corpus by means of statistical measures. We observed verbs with a relatively clear positive or negative polarity preference, as well as cases of verbs where positive and negative polarity preference is balanced (we call these bipolar-preference verbs). Given clear-cut polarity preferences of a verb, nouns, whose polarity is yet unknown, can now be classified. We reached a lower bound of 81% precision in our experiments, whereas the upper bound goes up to 92%.