In the present paper we report on a recent effort that resulted in the establishment of a unique multimodal database, referred to as the PROMETHEUS database. This database was created in support of research and development activities, performed within the European Commission FP7 PROMETHEUS project, aiming at the creation of a framework for monitoring and interpretation of human behaviours in unrestricted indoors and outdoors environments. In the present paper we discuss the design and the implementation of the audio part of the database and offer statistical information about the audio content. Specifically, it contains single-person and multi-person scenarios, but also covers scenarios with interactions between groups of people. The database design was conceived with extended support of research and development activities devoted to detection of typical and atypical events, emergency and crisis situations, which assist for achieving situational awareness and more reliable interpretation of the context in which humans behave. The PROMETHEUS database allows for embracing a wide range of real-world applications, including smart-home and human-robot interaction interfaces, indoors/outdoors public areas surveillance, airport terminals or city park supervision, etc. A major portion of the PROMETHEUS database will be made publically available by the end of year 2010.

The present paper outlines the Vergina speech database, which was developed in support of research and development of corpus-based unit selection and statistical parametric speech synthesis systems for Modern Greek language. In the following, we describe the design, development and implementation of the recording campaign, as well as the annotation of the database. Specifically, a text corpus of approximately 5 million words, collected from newspaper articles, periodicals, and paragraphs of literature, was processed in order to select the utterances-sentences needed for producing the speech database and to achieve a reasonable phonetic coverage. The broad coverage and contents of the selected utterances-sentences of the database ― text corpus collected from different domains and writing styles ― makes this database appropriate for various application domains. The database, recorded in audio studio, consists of approximately 3,000 phonetically balanced Modern Greek utterances corresponding to approximately four hours of speech. Annotation of the Vergina speech database was performed using task-specific tools, which are based on a hidden Markov model (HMM) segmentation method, and then manual inspection and corrections were performed.

The present paper reports on a recent effort that resulted in the establishment of a unique multimodal affect database, referred to as the PlayMancer database. This database was created in support of the research and development activities, taking place within the PlayMancer project, which aim at the development of a serious game environment in support of treatment of patients with behavioural and addictive disorders, such as eating disorders and gambling addictions. Specifically, for the purpose of data collection, we designed and implemented a pilot trial with healthy test subjects. Speech, video and bio-signals (pulse-rate, SpO2) were captured synchronously, during the interaction of healthy people with a number of video games. The collected data were annotated by the test subjects (self-annotation), targeting proper interpretation of the underlying affective states. The broad-shouldered design of the PlayMancer database allows its use for the needs of research on multimodal affect-emotion recognition and multimodal human-computer interaction in serious games environment.

This paper describes Eksairesis, a system for learning economic domain knowledge automatically from Modern Greek text. The knowledge is in the form of economic terms and the semantic relations that govern them. The entire process in based on the use of minimal language-dependent tools, no external linguistic resources, and merely free, unstructured text. The methodology is thereby easily portable to other domains and other languages. The text is pre-processed with basic morphological annotation, and semantic (named and other) entities are identified using supervised learning techniques. Statistical filtering, i.e. corpora comparison is used to extract domain terms and supervised learning is again employed to detect the semantic relations between pairs of terms. Advanced classification schemata, ensemble learning, and one-sided sampling, are experimented with in order to deal with the noise in the data, which is unavoidable due to the low pre-processing level and the lack of sophisticated resources. An average 68.5% f-score over all the classes is achieved when learning semantic relations. Bearing in mind the use of minimal resources and the highly automated nature of the process, classification performance is very promising, compared to results reported in previous work.

The present paper deals with the design and the annotation of a Greek real-world emotional speech corpus. The speech data consist of recordings collected during the interaction of naïve users with a smart-home dialogue system. Annotation of the speech data with respect to the uttered command and emotional state was performed. Initial experimentations towards recognizing negative emotional states were performed and the experimental results indicate the range of difficulties when dealing with real-world data.

A speech and noise corpus dealing with the extreme conditions of the motorcycle environment is developed within the MoveOn project. Speech utterances in British English are recorded and processed approaching the issue of command and control and template driven dialog systems on the motorcycle. The major part of the corpus comprises noisy speech and environmental noise recorded on a motorcycle, but several clean speech recordings in a silent environment are also available. The corpus development focuses on distortion free recordings and accurate descriptions of both recorded speech and noise. Not only speech segments are annotated but also annotation of environmental noise is performed. The corpus is a small-sized speech corpus with about 12 hours of clean and noisy speech utterances and about 30 hours of segments with environmental noise without speech. This paper addresses the motivation and development of the speech corpus and finally presents some statistics and results of the database creation.

This paper describes a corpus consisting of audio data for automatic space monitoring based solely on the perceived acoustic information. The particular database is created as part of a project aiming at the detection of abnormal events, which lead to life-threatening situations or property damage. The audio corpus is composed of vocal reactions and environmental sounds that are usually encountered in atypical situations. The audio data is composed of three parts: Phase I - professional sound effects collections, Phase II recordings obtained from action and drama movies and Phase III - vocal reactions related to real-world emergency events as retrieved from television, radio broadcast news, documentaries etc. The annotation methodology is given in details along with preliminary classification results and statistical analysis of the dataset regarding Phase I. The main objective of such a dataset is to provide training data for automatic recognition machines that detect hazardous situations and to provide security enhancement in public environments, which otherwise require human supervision.

For the present work, we deal with the significant problem of high imbalance in data in binary or multi-class classification problems. We study two different linguistic applications. The former determines whether a syntactic construction (environment) co-occurs with a verb in a natural text corpus consists a subcategorization frame of the verb or not. The latter is called Name Entity Recognition (NER) and it concerns determining whether a noun belongs to a specific Name Entity class. Regarding the subcategorization domain, each environment is encoded as a vector of heterogeneous attributes, where a very high imbalance between positive and negative examples is observed (an imbalance ratio of approximately 1:80). In the NER application, the imbalance between a name entity class and the negative class is even greater (1:120). In order to confront the plethora of negative instances, we suggest a search tactic during training phase that employs Tomek links for reducing unnecessary negative examples from the training set. Regarding the classification mechanism, we argue that Bayesian networks are well suited and we propose a novel network structure which efficiently handles heterogeneous attributes without discretization and is more classification-oriented. Comparing the experimental results with those of other known machine learning algorithms, our methodology performs significantly better in detecting examples of the rare class.

For the present work, we introduce and evaluate a novel Bayesian syntactic shallow parser that is able to perform robust detection of pairs of subject-object and subject-direct object-indirect object for a given verb, in a natural language sentence. The shallow parser infers on the correct subject-object pairs based on knowledge provided by Bayesian network learning from annotated text corpora. The DELOS corpus, a collection of economic domain texts that has been automatically annotated using various morphological and syntactic tools was used as training material. Our shallow parser makes use of limited linguistic input. More specifically, we consider only part of speech tagging, the voice and the mood of the verb as well as the head word of a noun phrase. For the task of detecting the head word of a phrase we used a sentence boundary detector. Identifying the head word of a noun phrase, i.e. the word that holds the morphological information (case, number) of the whole phrase, also proves to be very helpful for our task as its morphological tag is all the information that is needed regarding the phrase. The evaluation of the proposed method was performed against three other machine learning techniques, namely naive Bayes, k-Nearest Neighbor and Support Vector Machines, methods that have been previously applied to natural language processing tasks with satisfactory results. The experimental outcomes portray a satisfactory performance of our proposed shallow parser, which reaches almost 92 per cent in terms of precision.

Previous studies on automatic extraction of lexical similarities have considered as semantic unit of text the word. However, the theory of contextual lexical semantics implies that larger segments of text, namely non-compositional multiwords, are more appropriate for this role. We experimentally tested the applicability of this notion applying automatic collocation extraction to identify and merge such multiwords prior to the similarity estimation process. Employing an automatic WordNet-based comparative evaluation scheme along with a manual evaluation procedure, we ascertain improvement of the extracted similarity relations.