G. Kokkinakis

Computer-Based Authorship Attribution Without Lexical Measures

The most important approaches to computer-assistedauthorship attribution are exclusively based onlexical measures that either represent the vocabularyrichness of the author or simply comprise frequenciesof occurrence of common words. In this paper wepresent a fully-automated approach to theidentification of the authorship of unrestricted textthat excludes any lexical measure. Instead we adapt aset of style markers to the analysis of the textperformed by an already existing natural languageprocessing tool using three stylometric levels, i.e.,token-level, phrase-level, and analysis-levelmeasures. The latter represent the way in which thetext has been analyzed. The presented experiments ona Modern Greek newspaper corpus show that the proposedset of style markers is able to distinguish reliablythe authors of a randomly-chosen group and performsbetter than a lexically-based approach. However, thecombination of these two approaches provides the mostaccurate solution (i.e., 87% accuracy). Moreover, wedescribe experiments on various sizes of the trainingdata as well as tests dealing with the significance ofthe proposed set of style markers.

Towards an adaptive natural language interface to command languages

Operating system command languages assist the user in executing commands for a significant number of common everyday tasks. On the other hand, the introduction of textual command languages for robots has provided the opportunity to perform some important functions that leadthrough programming cannot readily accomplish. However, such command languages assume the user to be expert enough to carry out a specific task in these application domains. On the contrary, a natural language interface to such command languages, apart from being able to be integrated into a future speech interface, can facilitate and broaden the use of these command languages to a larger audience. In this paper, advanced techniques are presented for an adaptive natural language interface that can (a) be portable to a large range of command languages, (b) handle even complex commands thanks to an embedded linguistic parser, and (c) be expandable and customizable by providing the casual user with the opportunity to specify some types of new words as well as the system developer with the ability to introduce new tasks in these application domains. Finally, to demonstrate the above techniques in practice, an example of their application to a Greek natural language interface to the MS-DOS operating system is given.

Segmental Duration Modeling for Greek Speech Synthesis

In this paper we cope with the task of modeling phoneme duration for Greek speech synthesis. In particular we apply well established machine learning approaches to the WCL-1 prosodic database for predicting segmental durations from shallow morphosyntactic and prosodic features. We employ decision trees, instance based learning and linear regression. Trained on a 5500 word database, both CART and linear regression models proved to be the most effective in terms for the task with a root mean square error off 0. 0252 and 0.0251 respectively.

Improving the robustness of noisy MFCC features using minimal recurrent neural networks

We describe a novel technique for improving speech recognition performance in real environments. We investigate the special case of speech recognition in the car environment for SNRs ranging from -10 to 20 dB. Our approach makes use of a feature set that is composed of uncorrelated variables in order to create a group of neural networks each one dedicated to a sole variable of the feature vector. This technique results in neural networks of much smaller total number of weights than reported cases and consequently in faster training and execution performance. Furthermore, contextual information regarding a features history is incorporated into the network by making use of recurrent neural networks. We evaluate the performance in comparison with the standard MLPs and TDNNs in order to prove that they compare favourably to them in terms of recognition improvement over a wide range of SNRs.

Development and evaluation of a prosodic database for Greek speech synthesis and research

Abstract In this article the definition, construction and statistical evaluation of a prosodic database of Greek speech are presented. The main motivation for the development of such a database was its use as a research tool for Text-to-Speech synthesis and the study of prosody in general. Beginning with the task of text selection we came to a final set containing sentences with almost 95% of all Greek syllables, extracted from a widely used Greek dictionary. Then, a professional radio actress was instructed to utter these sentences in reading style and at reading rate; this was recorded at 44 kHz/16 bits, in the anechoic chamber of a professional studio. The intonational phenomena were transcribed on the corresponding speech signals by a trained phonetician using the ToBI annotation model adapted to Greek prosodic patterns. The speech data were segmented to the phoneme level employing a phoneme recognizer based on the HTK platform. All files were aligned so that possible relations among text, intonational and durational labelling could be identified. For database management, the EMU speech database system was utilized. Extensive measurements of numerous annotated events presented in histograms and tables provide detailed information on the database. Finally, we evaluate prediction models of prosodic phrase breaks and pitch accents derived from our database. Performance of these models was also compared to models derived under the same experimental conditions with a limited domain corpus of Greek speech.

Improving environmental robustness of speech recognition using neural networks

This paper presents a method for improving speech recognition in noisy environment by using neural networks. Two multilayer perceptrons (MLPs) are used. The first MLP minimises the difference between noisy and clean speech and the second one measures the degree of noise in the speech signal and adjusts the time interval between subsequent frames of the processed speech signal accordingly. If we use the technique presented in this paper as a pre-processing stage of a speech recognition system we can extend the application of the system to different environments without re-training it. We need only to train the preprocessing stage with a small portion of noisy data which is created by conducting part of the original clean speech database used for training the speech recognizer through the desired environment. There is no need for creating a new database in the desired working environment. Our method was tested on a vowel spotting system, and is trained with two well known databases: TIMIT and NTIMIT. The evaluation of the system through a vowel spotting process, shows a significant improvement of the recognition rate of the system.

Extraction and recognition of handwritten alphanumeric characters from application forms

This paper presents a reading system capable of extracting the handwritten text and recognizing the alphanumeric characters from application forms. The system has been designed and implemented in the framework of the LE project ACCESS. The application forms are scanned and the handwritten parts are automatically separated. The character recognition is based on discrete hidden Markov models. In our system the estimation of the HMM parameters has been simplified by using a left-to-right HMM with step one. The system recognizes 60 alphanumeric characters (26 English upper-case letters, 24 Greek upper-case letters and 10 digits). The experiments carried out achieved a recognition rate of 93% in character level and 88% in word level. The latter improved to 97% by lexical confirmation. A novelty of this system is the feature extraction algorithm applied to the characters and the resulting very fast recognition.

Enhancing Text Retrieval by Using Advanced Stylistic Techniques

Text retrieval techniques have long focused on the topic of texts rather than the pragmatic role they play per se. In this article, we address two other aspects in text processing that could enhance text retrieval: (a) the detection of functional style in retrieved texts, and (b) the detection of writers attitude towards a given topic in retrieved texts. The former is justified by the fact that current text databases have become highly heterogeneous in terms of document inclusion, while the latter is dictated by the need for advanced and intelligent retrieval tools. Towards this aim, two generalised methodologies are presented in order to achieve the implementation of the findings in both aspects in text processing respectively. Particularly, the first one is fully developed and thus is analysed and evaluated in detail, while for the second one the theoretical framework is given for its subsequent computational implementation. Both approaches are as language independent as possible, empirically driven, and can be used, apart from information retrieval purposes, in various natural language processing applications. These include grammar and style checking, natural language generation, summarisation, style verification in real-world texts, recognition of style shift between adjacent portions of text, and author identification.

Using Functional Style Features to Enhance Information Extraction from Greek Texts

Current Information Extraction (IE) systems extract, in most cases, fixed information from documents [1,2]. This information pertains only to four distinct tasks: named entity recognition, coreference identification, template elements filling, and scenario-based template elements filling. Thus, providing these systems with the capability of locating stylistic features in a text and thus detecting its genre, it would be possible to meet specific user interests. For instance, users are often looking for texts on a certain topic with particular, quite narrow generic properties, such as authoritatively written documents, opinion pieces, scientific articles, and so on.

Vowel-non vowel decision using neural networks and rules

This paper describes a speaker independent vowel/non-vowel classifier based on neural networks and several rules. RASTA-PLP analysis of the speech signal resulting to mel-cepstral coefficients and a formant tracking method are used in order to provide the feature vectors for the MLP. To train and test the system we used a part of the TIMIT database. The results indicate that the performance of this classifier for speaker independent vowel classification is approximately 98.5% so it can be favorably used for speaker recognition or speech labeling purposes.