Catálogo de publicaciones - libros

In this paper, we propose a phrase reordering model for statistical machine translation. The model is derived from the bracketing ITG, and integrates the local and global reordering model. We present a method to extract phrase pairs from a word-aligned bilingual corpus in which the alignments satisfy the ITG constraint, and we also extract the reordering information for the phrase pairs, which are used to build the re-ordering model. Through experiments, we show that this model obtains significant improvements over the baseline on a Chinese-English translation.

Automatic pronoun resolution may improve the performance of natural language systems, such as translators, generators and summarizers. Difficulties may arise when there is more than one potential candidate for a referent. There has been little research on pronoun resolution for Portuguese, if compared to other languages, such as English. This paper describes a variant of Hobbs’ syntactic algorithm for pronoun resolution in Portuguese. The system was evaluated comparing the results with the ones obtained with another syntactic algorithm for pronoun resolution handling, the Lappin and Leass’ algorithm. The same Portuguese corpora were used and significant improvement was verified with Hobbs’ algorithm.

It is well known that lexical knowledge sources such as WordNet, HowNet are very important to natural language processing applications. In those lexical resources, attributes play very important roles for defining and distinguishing different concepts. In this paper, we propose a novel method to automatically discover the attribute hosts of HowNet’s attribute set. Given an attribute, we model the solving of its host as a problem of selectional constraint resolution. The World Wide Web is exploited as a large corpus to acquire the training data for such a model. From the training data, the attribute hosts are discovered by using a statistical measure and a semantic hierarchy. We evaluate our algorithm by comparing the result with the original hand-coded attribute specification in HowNet. Some experimental results about the performance of the method are provided.

The exponential growth of the Internet has allowed the development of a market of on-line job search sites. This paper aims at presenting the E-Gen system (Automatic Job Offer Processing system for Human Resources). E-Gen will implement two complex tasks: an analysis and categorisation of job postings, which are unstructured text documents (e-mails of job listings possibly with an attached document), an analysis and a relevance ranking of the candidate answers (cover letter and ). This paper aims to present a strategy to resolve the first task: after a process of filtering and lemmatisation, we use vectorial representation before generating a classification with Support Vector Machines. This first classification is afterwards transmitted to a ”corrective” post-process which improves the quality of the solution.

In Natural Language Processing there are different problems to solve: lexical ambiguity, summarization, information extraction, speech processing, etc. In particular, lexical ambiguity is a difficult task that nowadays is still open to new approaches. In fact, there is still a lack of systems that solve efficiently this kind of problem. At present, we find two different approaches: knowledge systems and machine learning systems. Recent studies demonstrate that machine learning systems obtain better results than knowledge systems but there is a problem: the lack of annotated contexts and corpus to train the systems. In this work, we try to avoid this situation by combining a new machine learning system with a knowledge based system.

The use of biologically inspired, feature extraction methods has improved the performance of artificial systems that try to emulate some aspect of human communication. Recent techniques, such as independent component analysis and sparse representations, have made it possible to undertake speech signal analysis using features similar to the ones found experimentally at the primary auditory cortex level. In this work, a new type of speech signal representation, based on the spectro-temporal receptive fields, is presented, and a problem of phoneme classification is tackled for the first time using this representation. The results obtained are compared, and found to greatly improve both an early auditory representation and the classical front-end based on Mel frequency cepstral coefficients.

This paper shows results obtained in the Automatic Speech Recognition (ASR) task for a corpus of digits speech files with a determinate noise level immerse. In the experiments, we used several speech files that contained Gaussian noise. We used HTK (Hidden Markov Model Toolkit) software of Cambridge University in the experiments. The noise level added to the speech signals was varying from fifteen to forty dB increased by a step of 5 units. We used an adaptive filtering to reduce the level noise (it was based in the Least Measure Square –LMS- algorithm) and two different wavelets (Haar and Daubechies). With LMS we obtained an error rate lower than if it was not present and it was better than wavelets employed for this experiment of Automatic Speech Recognition. For decreasing the error rate we trained with 50% of contaminated and originals signals to the ASR system. The results showed in this paper are focused to try analyses the ASR performance in a noisy environment and to demonstrate that if we are controlling the noise level and if we know the application where it is going to work, then we can obtain a better response in the ASR tasks. Is very interesting to count with these results because speech signal that we can find in a real experiment (extracted from an environment work, i.e.), could be treated with these technique and we can decrease the error rate obtained. Finally, we report a recognition rate of 99%, 97.5% 96%, 90.5%, 81% and 78.5% obtained from 15, 20, 25, 30, 35 and 40 noise levels, respectively when the corpus mentioned before was employed and LMS algorithm was used. Haar wavelet level 1 reached up the most important results as an alternative to LMS algorithm, but only when the noise level was 40 dB and using original corpus.

An algorithm for decoding commands spoken in an intelligent environment through iterative vocabulary reduction is presented. Current research in the field of speech recognition focuses primarily on the optimization of algorithms for single pass decoding using large vocabularies. While this is ideal for processing conversational speech, alternative methods should be explored for different domains of speech, specifically commands issued verbally in an intelligent environment. Such commands have both an explicitly defined structure and a vocabulary limited to valid task descriptions. We propose that a multiple pass context-driven decoding scheme utilizing dictionary pruning yields improved accuracy; this occurs when one deals with command structure and a reduced vocabulary. Each iteration incorporates the hypothesis of the previous into its decoding scheme by removing unlikely words from the current language model. We have applied this decoding method to a comprehensive set of spoken commands through the use of Sphinx-4, an Automatic Speech Recognition (ASR) engine using the Hidden Markov Model (HMM). When decoding via HMM, multiple previous states are used to determine the current state, thus utilizing context to aid in intelligent recognition. Our results show that within a fixed domain, multiple pass decoding yields recognition accuracy. Further research must be conducted to optimize practical context driven decoding and to apply the method to larger domains, primarily those of intelligent environments.

Emotion recognition aims to make computer understand ambiguous information of human emotion. Recently, research of emotion recognition is actively progressing in various fields such as natural language processing, speech signal processing, image data processing or brain wave analysis. We propose a method to recognize emotion in dialogue text by using originally created Emotion Word Dictionary. The words in the dictionary are weighted according to the occurrence rates in the existing emotion expression dictionary. We also propose a method to judge the object of emotion and emotion expressivity in dialogue sentences. The experiment using 1,190 sentences proved about 80% accuracy.

The Human Computer Interaction (HCI) Technology has emerged in the different fields in applications in computer vision and recognition systems such as virtual environment, video games, e-business and multimedia management. In this paper we propose a framework of designing the Mental State Transition (MST) of a human being or virtual character. The expressions of human emotion can be easily remarked by facial expressions, gestures, sound and other visual characteristics. But the potential MST modeling in affective data are always hidden actually. We analysis the framework of MST and employ DBNs to construct the MST networks and finally the experiment has been implemented to derive the ground truth of the data and verify the effectiveness.