Dealing with Small, Noisy and Imbalanced Data
TSD '08 Proceedings of the 11th international conference on Text, Speech and Dialogue
| Hi-index | 0.00 |
There are over 400 million speakers of Balto-Slavonic languages world-wide (synonymously used: Balto-Slavic). As of 2007, almost a third of the 23 official European Union languages are Balto-Slavonic, i.e. Bulgarian, Czech, Latvian, Lithuanian, Polish, Slovak and Slovene. The two most recent rounds of the EU Enlargement fundamentally raised the interest in these languages: translators and interpreters for new language pairs need to be found, the interest in Machine (Aided) Translation systems has risen and tools that help language specialists and information-seeking individuals are now highly sought after. For some of the countries speaking Balto-Slavonic languages, there is a rich linguistic heritage, and computational linguistics research and development is rather advanced. For others, however, there has been little development. This is due to the often small number of speakers of that language (Latvian, for instance, is spoken natively by around 1.5 Million people) combined with a lack of access to basic resources needed for Natural Language Processing such as machine-readable corpora and dictionaries, morphological analysers, part-of-speech taggers, parsers, etc. This leads to a linguistic brain drain as some of the best computational linguistics students go abroad or --- even when staying in their mother country --- work on developing new systems for English, French or German because resources for these languages are readily available. Even when linguistic resources and tools are available to the scientific community, methods that have been successfully applied to Germanic and Romance languages cannot simply be ported to languages from the Balto-Slavonic group. The most well-known linguistic phenomena making Balto-Slavonic text analysis harder are the highly inflectional character and the related phenomenon of free word order in these languages. The invited speaker at the workshop, Adam Przepiorkowski from the Polish Academy of Sciences, explains these and a number of further specific linguistic phenomena typical for this language group. Interestingly, he points out that the differences may not always make text analysis tasks harder, but that they can also make some tasks easier. When proposing this workshop to the Association for Computational Linguistics, we knew that Language Technology for some of the languages is not very advanced and that, to date, not much work has been carried out in the area of Information Extraction. The objective of this workshop was thus to promote the work on Balto-Slavonic languages by helping scientists to describe and share their resources and to describe their efforts, hoping that the experiences of a few will be useful for many others. We did not expect to receive papers presenting highly novel approaches to Information Extraction, but rather an adaptation of known methods to new languages and solutions for specific challenges regarding the Balto-Slavonic group. We found, however, that the current work on applying known approaches to a different language type did produce some very interesting work. We received in total 20 submissions, of which only eight were specifically about Information Extraction, i.e. named entity recognition or the extraction of definitions. One of these papers targeted a more high-level Information Extraction task: an application combining more than one entity to fill a scenario template. The other 12 papers fall under the category Enabling Technology, describing mostly morphological tools and resources, taggers, corpora, WordNet developments and topic segmentation of texts. Each of the submissions was reviewed by at least three peers. Finally, we have accepted 9 papers for oral presentations (acceptance rate of 45%) and selected a further three for poster presentations. We want to use this opportunity to thank the Programme Committee for their thorough reviews and mostly extensive and useful comments, as well as for keeping to the deadlines.