José Gabriel Pereira Lopes

Using LocalMaxs Algorithm for the Extraction of Contiguous and Non-contiguous Multiword Lexical Units

The availability of contiguous and non-contiguous multiword lexical units (MWUs) in Natural Language Processing (NLP) lexica enhances parsing precision, helps attachment decisions, improves indexing in information retrieval (IR) systems, reinforces information extraction (IE) and text mining, among other applications. Unfortunately, their acquisition has long been a significant problem in NLP, IR and IE. In this paper we propose two new association measures, the Symmetric Conditional Probability (SCP) and the Mutual Expectation (ME) for the extraction of contiguous and non-contiguous MWUs. Both measures are used by a new algorithm, the LocalMaxs, that requires neither empirically obtained thresholds nor complex linguistic filters. We assess the results obtained by both measures by comparing them with reference association measures (Specific Mutual Information, o 2 , Dice and Log-Likelihood coefficients) over a multilingual parallel corpus. An additional experiment has been carried out over a part-of-speech tagged Portuguese corpus for extracting contiguous compound verbs.

Unified logic programming approach to the abduction of plans and intentions in information-seeking dialogues☆

We propose a framework that supports the recognition of plans and intentions behind speech acts through abductive inferences over discourse sentences. These inferences allow each agent to have an active and intelligent participation in dialogues, namely, in cooperative information-seeking dialogues. In our framework, the possible actions, events, states, and world knowledge are represented by extended logic programs (LP with explicit negation), and the abductive inference porcess is modeled by the framework proposed by Pereira et al. [13], which is based on the Well Founded Semantics augmented with explicit negation (WFSX) and contradiction removal semantics (CRSX). It will be shown how this framework supports abductive planning with Event Calculus [5], and some examples will be shown [10, 14] in the domain of information-seeking dialogues. Finally, some open problems will be pointed out.

Syntactic-Based Methods for Measuring Word Similarity

This paper explores different strategies for extracting similarity relations between words from partially parsed text corpora. The strategies we have analysed do not require supervised training nor semantic information available from general lexical resources. They differ in the amount and the quality of the syntactic contexts against which words are compared. The paper presents in details the notion of syntactic context and how syntactic information could be used to extract semantic regularities of word sequences. Finally, experimental tests with Portuguese corpus demonstrate that similarity measures based on fine-grained and elaborate syntactic contexts perform better than those based on poorly defined contexts.

Selection Restrictions Acquisition from Corpora

This paper describes an automatic clustering strategy for acquiring selection restrictions. We use a knowledge-poor method merely based on word cooccurrence within basic syntactic constructions; hence, neither semantic tagged corpora nor man-made lexical resources are needed for generalising semantic restrictions. Our strategy relies on two basic linguistic assumptions. First, we assume that two syntactically related words impose semantic selectional restrictions to each other (cospecification). Second, it is also claimed that two syntactic contexts impose the same selection restrictions if they cooccur with the same words (contextual hypothesis). In order to test our learning method, preliminary experiments have been performed on a Portuguese corpus.

Measuring the impact of cognates in parallel text alignment

This paper evaluates the impact of considering cognates for aligning parallel texts using our aligner. By considering that two words may be cognates if their similarity is higher than a certain threshold and if they are automatically selected as aligners, we tested different degrees of cognativeness, by varying threshold values. A new methodology to measure the quality of resulting alignments is presented. Two language pairs are addressed: Portuguese-Spanish, and Portuguese-English

Multilingual Document Clustering, Topic Extraction and Data Transformations

This paper describes a statistics-based approach for clustering documents and for extracting cluster topics. Relevant Expressions (REs) are extracted from corpora and used as clustering base features. These features are transformed and then by using an approach based on Principal Components Analysis, a small set of document classification features is obtained. The best number of clusters is found by Model-Based Clustering Analysis. Data transformations to approximate to normal distribution are done and results are discussed. The most important REs are extracted from each cluster and taken as cluster topics.

Extracting Equivalents from Aligned Parallel Texts: Comparison of Measures of Similarity

Extraction of term equivalents is one of the most important tasks for building bilingual dictionaries. Several measures have been proposed to extract translation equivalents from aligned parallel texts. In this paper, we will compare 28 measures of similarity based on the co-occurrence of words in aligned parallel text segments. Parallel texts are aligned using a simple method that extends previous work by Pascale Fung & Kathleen McKeown and Melamed but which, in contrast, does not use statistically unsupported heuristics to filter reliable points.

Modelling Dynamic Aspects of Intentions

In this paper, we define a formal model of intentions that accounts for both static and dynamic aspects of intentions. By static aspects, we mean its relation with desires and beliefs and the properties that this relation has (namely, consistency and avoidance of side-effects). By dynamic aspects, we mean establishing criteria, other than those derived from commitment, to drop intentions (namely, based on the cause-effect relation between desires and intentions) and mechanisms that triggers the reasoning process without imposing a significant additional burden on the agent. Our approach is to define those conditions that make the agent start reasoning about intentions as constraints over its beliefs.

A Distributed Approach for a Robust and Evolving NLP System

We present in this paper some aspects concerning the design and implementation of an architecture that is the basis for the development of a natural language processing system that, besides the obvious goal of building some computational representation (at a desired level) of the input, has two main objectives: to be robust and to evolve. To be robust in the sense that the non recognition of some input should not block the system but, instead, should lead the system to an automatic recovery process. To evolve, so that when some incompleteness/incorrectness is detected (or suspected) during a recovery process, the component responsible for the mistake should be updated accordingly, so that in future analogous situations the system can perform better. In order to achieve this goal we propose the definition of a distributed architecture.

A two-headed architecture for intelligent multimedia man-machine interaction

Abstract In this paper a two-headed architecture designed for handling intelligent multimedia man-machine interaction is described. In this work we are mainly interested in interactions that may consist only of natural language sentences and graphical actions and events (multimodal interactions). The architecture is extended from Lopes [Lop91] and is composed of two decisione centers (heads): an interactional and a reflectional one. The interactional decision center is responsible for the interaction with the user and consists of a handler that activates the adequate low-level executive modules (input interpreters, planners and executive generators of planned actions) that perform the desired actions and use and update the context of interaction. The reflectional module is responsible for detecting faults and deadlocks in a interaction and for propagating solutions that will lead to their repair. It also consists of a handler and of executive modules that use the interactional context as object as well as other knowledge sources. It is shown how this architecture can be applied to multimodal interactions. An implemented new environment for the creation of dynamic lexical databases is described as an application.