Ivandré Paraboni

Computational Processing of the Portuguese Language

This paper reports findings from an analysis of errors made by an automatic speech recogniser trained and tested with 3-10-year-old European Portuguese childrens speech. We expected and were able to identify frequent pronunciation error patterns in the childrens speech. Furthermore, we were able to correlate some of these pronunciation error patterns and automatic speech recognition errors. The findings reported in this paper are of phonetic interest but will also be useful for improving the performance of automatic speech recognisers aimed at children representing the target population of the study.This book constitutes the refereed proceedings of the 11th International Workshop on Computational Processing of the Portuguese Language, PROPOR 2014, held in Sao Carlos, Brazil, in October 2014. The 14 full papers and 19 short papers presented in this volume were carefully reviewed and selected from 63 submissions. The papers are organized in topical sections named: speech language processing and applications; linguistic description, syntax and parsing; ontologies, semantics and lexicography; corpora and language resources and natural language processing, tools and applications.

Generating Referring Expressions: Making Referents Easy to Identify

It is often desirable that referring expressions be chosen in such a way that their referents are easy to identify. This article focuses on referring expressions in hierarchically structured domains, exploring the hypothesis that referring expressions can be improved by including logically redundant information in them if this leads to a significant reduction in the amount of search that is needed to identify the referent. Generation algorithms are presented that implement this idea by including logically redundant information into the generated expression, in certain well-circumscribed situations. To test our hypotheses, and to assess the performance of our algorithms, two controlled experiments with human subjects were conducted. The first experiment confirms that human judges have a preference for logically redundant expressions in the cases where our model predicts this to be the case. The second experiment suggests that readers benefit from the kind of logical redundancy that our algorithms produce, as measured in terms of the effort needed to identify the referent of the expression.

Reference and the facilitation of search in spatial domains

Earlier work has suggested that, in hierarchically ordered domains (e.g., a document divided into sections and subsections), referring expressions that are judiciously over-specified to a higher extent than is achieved by existing generation algorithms can make it considerably easier for a hearer to find the referent of the referring expression. The present paper investigates over-specification in spatial domains, which plays an important role in daily life. We report an experiment whose aim is (1) to find out whether over-specification plays a similar role in spatial domains as in hierarchically ordered domains, (2) to obtain a better understanding of the reasons why over-specification can be helpful to hearers, and (3) to propose an algorithmic model of reference production that takes these findings into account. The results suggest that judicious over-specification can facilitate search in a precisely defined class of problematic conditions (but less so in other cases) even if the hearer has previous knowledge about the domain. The implications of these findings are discussed and an algorithm for the generation of referring expressions is proposed that reflects them as closely as possible.

Statistical Surface Realisation of Portuguese Referring Expressions

Natural Language Generation systems usually require substantial knowledge about the structure of the target language in order to perform the final task in the generation process --- the mapping from semantic representation to text known as surface realisation. Designing knowledge bases of this kind, typically represented as sets of grammar rules, may however become a costly, labour-intensive enterprise. In this work we take a statistical approach to surface realisation in which no linguistic knowledge is hard-coded, but rather trained automatically from large corpora. Results of a small experiment in the generation of referring expressions show significant levels of similarity between our (computer-generated) text and those produced by humans, besides the usual benefits commonly associated with statistical NLP such as low development costs, domain- and language-independency.

Overspecified Reference in Hierarchical Domains: Measuring the Benefits for Readers

It is often desirable that referring expressions be chosen in such a way that their referents are easy to identify. In this paper, we investigate to what extent identification becomes easier by the addition of logically redundant properties. We focus on hierarchically structured domains, whose content is not fully known to the reader when the referring expression is uttered.

A Machine Learning Approach to Portuguese Pronoun Resolution

Anaphora resolution is an essential component of most NLP applications, from text understanding to Machine Translation. In this work we discuss a supervised machine learning approach to the problem, focusing on instances of anaphora ubiquitously found in a corpus of Brazilian Portuguese texts, namely, third-person pronominal references. Although still limited to a subset of the more general co-reference resolution problem, our present results are comparable to existing work in the field in both English and Portuguese languages, representing the highest accuracy rates that we are aware of in (Brazilian) Portuguese pronoun resolution.

Generating Relational Descriptions Involving Mutual Disambiguation

This paper discusses the generation of relational referring expressions in which target and landmark descriptions are allowed to help disambiguate each other. Using a corpus of referring expressions in a simple visual domain - in which these descriptions are likely to occur - we propose a classification approach to decide when to generate them. The classifier is then embedded in a REG algorithm whose results outperform a number of naive baseline systems, suggesting that mutual disambiguation is fairly common in language use, and that this may not be entirely accounted for by existing REG algorithms.

Generating Spatial Referring Expressions in Interactive 3D Worlds

This paper discusses the computational problem of generating referring expressions (REG) in 3D virtual worlds. We propose a REG algorithm that attempts to make adequate choices of spatial relations for the purpose of disambiguation (as opposed to, e.g., determining the localisation of a previously identified object). The decisions made by the algorithm are based on existing computational models of spatial reference, and further refined by the use of domain knowledge obtained from a corpus of instructions in virtual environments. The proposed approach is shown to outperform a number of baseline systems, and provides evidence on how a standard REG approach may be applied to the generation of these descriptions in 3D virtual worlds.

USP-EACH frequency-based greedy attribute selection for referring expressions generation

Both greedy and domain-oriented REG algorithms have significant strengths but tend to perform poorly according to humanlikeness criteria as measured by, e. g., Dice scores. In this work we describe an attempt to combine both perspectives into a single attribute selection strategy to be used as part of the Dale & Reiter Incremental algorithm in the REG Challenge 2008, and the results in both Furniture and People domains.

Referring Expression Generation: Taking Speakers’ Preferences into Account

We describe a classification-based approach to referring expression generation (REG) making use of standard context-related features, and an extension that adds speaker-related features. Results show that taking speakers’ preferences into account outperforms the standard REG model in four test corpora of definite descriptions.