the average of the joint probability and span prob-ability features, and re-calculating the conditionalprobabilities from the averaged joint probabilities.6 Related WorkIn addition to the previously mentioned phrasealignment techniques, there has also been a signif-icant[r]
significantly advanced the development of machine translation (MT). However, lack of sufficient bilingual linguistic resources for many languages and domains is still one of the major obstacles for further advancement of automated translation. At the same time, comparable corpora, i.e., non-p[r]
and discovered, through the alignment process, in the other one. Our procedure presents three main differences over other approaches: we do not force term translations to fit within specific patterns, we consider the whole sentences, thus enabling us to remove some ambiguities, and<[r]
with severe fading [1, 3, 8].In the literature, there are several blind criteria for the es-timation of a specific user with knowledge only of its spread-ing code. Some authors proposed the use of MMSE cri-teria to exploit the signal subspace defined by the desireduser’s code [2, 8]. In [9], a blind d[r]
adheres closely to the algorithm of Collins (2002).Although exact decoding improves alignment per-formance over the approximate search approach, thegain is marginal and not significant. This seems toindicate that the simulated annealing search strategyis fairly effective at avoiding loc[r]
method described in Section 4.1, while “grow-diag-final” shows the accuracy when phrases areextracted using the standard phrase extraction al-gorithm described in (Koehn et al., 2003).It is obvious that, for building the global phrasereordering model, our phrase extraction[r]
Hermjakob 2009; Liu et al. 2010). For example, Fossum et al. (2008) used a discriminatively trained model to identify and delete incorrect links from original word alignments to improve string-to-tree transformation rule extraction, which incor-porates four types of features such as le[r]
regions are detected by analyzing the robust edges or homogeneous color/grayscale components that belong to characters. For example, Cai et al. [1] detect text edges in video sequences using a color edge detector and then apply a low threshold to filter out definite non-edge points. Real text[r]
less patterns are extracted from corpus. Some lin-guistic rules (Argamon,1999) are introduced intoour model. It is observed that candidate patternshould contain content words. Some patterns areonly organized by pure function words, such as themost frequent patterns “the to”, “of the”. Thesepatterns[r]
(Liu et al., 2007; Zhang et al., 2008a) are two interesting modeling methods with promising results reported. Forest-based modeling aims to improve translation accuracy through digging the potential better parses from n-bests (i.e. forest) while tree sequence-based modeling aims to model non-syntact[r]
aligned (and likewise constraint 4 for f).5 ApplicationsThe need to find an optimal phrase alignment for aweighted sentence pair arises in at least two appli-cations. First, a generative phrase alignment modelcan be trained with Viterbi EM by finding optimalphrase al[r]
posite meanings.Ibrahim et al. (2003) combine the two approaches:aligned monolingual corpora and parsing. Theyevaluated their system with human judges who wereasked whether the paraphrases were “roughly inter-changeable given the genre”, scored an average of41% on a set of 130 paraphrases, wi[r]
also explore NLP techniques that we propose for the analysis of Vietnamese texts, focusing on the advanced candidate phrases recogni-tion phase as well as part-of-speech (POS) tagging. Finally, we review the results of sev-eral experiments that have examined the im-pacts of strategies chosen for Vie[r]
level alignments are nonetheless a common sourceof error in machine translation systems.A natural solution to several of these issues isunite the word-level and phrase-level models intoone learning procedure. Ideally, such a procedurewould remedy the deficiencies of word-level align-men[r]
Hindawi Publishing CorporationEURASIP Journal on Advances in Signal ProcessingVolume 2009, Article ID 629030, 9 pagesdoi:10.1155/2009/629030Research ArticleAutomated Intelligibility Assessment of PathologicalSpeech Using Phonological FeaturesCatherine Middag,1Jean-Pierre Martens,1Gwen Van Nuffelen,2[r]
from noisy alignments at the sentence and corpuslevel. The significant improvements above the base-line carry through when this method is combinedwith other phrasal and word level methods. Furtherexperimentation is required to fully appreciate therobustness of this technique, especially[r]
Farsi to English Model-4 word alignments areused, but we try different combination methodsand analysis the final alignment set and the result-ing phase translation table. Table 1 presents somestatistics. Each row corresponds to a particularcombination. The first two are intersection (I)[r]
languages the reordering could be incorporated intocorrectly ordered target phrases as discussed previ-ously.For pairs of languages with radically differentword order (e.g. English-Japanese), there needs tobe a global reordering of words similar to the casein the SFST-based translation system. Also,[r]
et al., 2009) produces improvements over the meth-ods above, while reducing the computational costby using weighted alignment matrices to representthe alignment distribution over each parallel sen-tence. However, their results were limited by thefact that they had no method for extract[r]