In order to provide open access to data of public interest, it is often necessary to perform several data curation processes. In some cases, such as biological databases, curation involves quality control to ensure reliable experimental support for biological sequence data. In others, such as medical records or judicial files, publication must not interfere with the right to privacy of the persons involved. There are also interventions in the published data with the aim of generating metadata that enable a better experience of querying and navigation. In all cases, the curation process constitutes a bottleneck that slows down general access to the data, so it is of great interest to have automatic or semi-automatic curation processes. In this paper, we present a solution aimed at the automatic curation of our National Jurisprudence Database, with special focus on the process of the anonymization of personal information. The anonymization process aims to hide the names of the participants involved in a lawsuit without losing the meaning of the narrative of facts. In order to achieve this goal, we need, not only to recognize person names but also resolve co-references in order to assign the same label to all mentions of the same person. Our corpus has significant differences in the spelling of person names, so it was clear from the beginning that pre-existing tools would not be able to reach a good performance. The challenge was to find a good way of injecting specialized knowledge about person names syntax while taking profit of previous capabilities of pre-trained tools. We fine-tuned an NER analyzer and we built a clusterization algorithm to solve co-references between named entities. We present our first results, which, for both tasks, are promising: We obtained a 90.21% of F1-micro in the NER task—from a 39.99% score before retraining the same analyzer in our corpus—and a 95.95% ARI score in clustering for co-reference resolution.

Discriminating antonyms and synonyms is an important NLP task that has the difficulty that both, antonyms and synonyms, contains similar distributional information. Consequently, pairs of antonyms and synonyms may have similar word vectors. We present an approach to unravel antonymy and synonymy from word vectors based on a siamese network inspired approach. The model consists of a two-phase training of the same base network: a pre-training phase according to a siamese model supervised by synonyms and a training phase on antonyms through a siamese-like model that supports the antitransitivity present in antonymy. The approach makes use of the claim that the antonyms in common of a word tend to be synonyms. We show that our approach outperforms distributional and patternbased approaches, relaying on a simple feed forward network as base network of the training phases.

No abstract

In this paper we describe a rule-based formalism for the analysis and labelling of texts segments. The rules are contextual rewriting rules with a restricted form of negation. They allow to underspecify text segments not considered relevant to a given task and to base decisions upon context. A parser for these rules is presented and consistence and completeness issues are discussed. Some results of an implementation of this parser with a set of rules oriented to the segmentation of texts in propositions are shown.

We describe the TEMANTEX annotation scheme for temporal expressions and other lexical indicators of temporality and we analyze a first annotation experience. TEMANTEX is mainly a revision of the markup language TIMEX3, but with some additions and a different treatment for relative expressions. Our alternative proposal is justified for two reasons. First, our system aims to cover other temporality-related lexical elements by defining annotations for what we call temporal indicators, which do not have an equivalent in the TimeML system. Second, regarding temporal expressions, our scheme has relevant differences that improve the annotation process and the interpretation potential. A first task of corpus annotation on a set of 2.300 words, comprising 33 temporal expressions and 35 temporal indicators, showed encouraging results.