539 research outputs found
Sort by
Treex: Modular NLP Framework
Get PDFTreex is a multi-purpose open-source natural language processing (NLP) framework implemented in Perl programming language. It allows for fast and efficient development of NLP applications by exploiting a wide range of software modules already integrated in Treex, such as tools for sentence segmentation, tokenization, morphological analysis, POS tagging, shallow and deep syntax parsing, named entity recognition, anaphora resolution, tree-to-tree translation, natural language generation, word-level alignment of parallel corpora, and other tasks. One of the most complex applications of Treex is the English-Czech machine translation system TectoMT. Several modules are available also for other languages (German, Russian, Arabic). Where possible, modules are implemented in a language-independent way, so they can be reused in many applications
Strojový překlad přes tektogramatickou rovinu v systému TectoMT
Get PDFEnglish-to-Czech machine translation implemented in TectoMT system consists of three phases: analysis, transfer, and synthesis. Transfer is performed on the tectogrammatical layer which is modified for MT purposes. Each phase is divided into so-called blocks which solve particular linguistically interpretable tasks (e.g. tagging with statistic tagger or clitic shifting according to hand-written rules). TectoMT system is designed in a modular way - blocks can be substituted with alternative implementations. The talk presents basic steps of the whole translation and focuses on improvements implemented in the last year, especially:
(a) tectogrammatical LM and Hidden Markov Tree Models,
(b) new translation dictionaries trained on parallel corpus CzEng using Maximum Entropy
English-Hindi Translation – Obtaining Mediocre Results with Bad Data and Fancy Models
No full textWe describe our attempt to improve on previous English to Hindi machine translation results, using two open-source phrase-based MT systems: Moses and Joshua. We use several approaches to morphological tagging: from automatic word classes, through stem-suffix segmentation, to a
POS tagger. We also experiment with factored language models. We evaluate various combinations of training data sets and other existing English-Hindi resources. To our knowledge, the BLEU score we obtained is currently the best published result for the IIIT-TIDES dataset
Pokyny k překladu určené překladatelům, revizorům a korektorům textů z Wall Street Journal pro projekt PCEDT
No full textPCEDT is planned to be a corpus of syntactically annotated parallel texts (in English and Czech) intended chiefly for machine translation experiments. The texts for PCEDT were taken from Penn Treebank, which means there are mostly economical articles from the Wall Street Journal. 2312 documents were used in PCEDT (approximately 49,000 sentences) that are manually annotated with constituent trees in Penn Treebank. For the Czech part of PCEDT, the English texts have to be translated into Czech. This book contents the guidelines for translators
Trainable Tokenizer
No full textTrainable Tokenizer is able to tokenize and segment most languages based on supplied configuration and sample data. The tokenizer is not aimed e.g. for Chinese with no explicit delimitation of words
Jak se dělá strojový překlad
No full textA quick introduction to problems of machine translation for the participants of Pralinka corespondence series
AJAX-CAT: Demo of a web editor with machine translation
No full textAJAX-CAT demonstrates an integration of machine translation system into a simple web text editor. For a given input sentence, AJAX-CAT presents a table of translation options of source phrases and also provides a suggestion box with output continuations when the user starts typing the translation
Manually Flagged Errors in WMT09 Test Set
No full textA collection of outputs of 4 machine translation systems. Errors in the output are manually flagged and classified
CzEng 0.9
No full textCzEng 0.9 is the third release of a sentence-parallel Czech-English corpus compiled at the Institute of Formal and Applied Linguistics (ÚFAL) freely available for non-commercial and research purposes.
CzEng 0.9 contains 8.0 million parallel sentences (93 million English and 82 million Czech tokens) from seven different types of sources automatically annotated at surface and deep (a- and t-) layers of syntactic representation
UMC003: Czech-English-Russian Tri-parallel Test Set for MT
No full textUMC003 is a cleaned tokenized development and test set to accompany the training data in UMC 0.1 aimed at Czech-English-Russian machine translation. For more information about the test set, see the README file in the UMC003 package