likely terminological units and show what kind of disambiguation LEXTER has to perform. As we already pointed out, LEXTER has been achieved in an industrial context; from the beginning of the project, we had decided to focus upon a strongly restrictive criterium : applying and testing the system ove[r]
specialising the grammar using an au-tomatic corpus-based method, and thencompiling the resulting specialised gram-mars into CFG form. Translation betweenlanguage centered and domain centeredsemantic representations is carried out byALTERF, another Open Source toolkit,which comb[r]
Introduction 5Within a task-based methodological framework, the authors present four tasks based on dierent types of corpora and designed to illustrate contrastive ob-jectives. e tasks are for novice translators doing non-specialised B-A translation (translation from a foreign langua[r]
are content words, then to extract loanwords correctly, we have to identify the original form using stemming. In this paper, we propose methods for extracting loanwords from Cyrillic Mongolian and producing a Japanese–Mongolian bilingual dictionary. We also propose a stemming method to identi[r]
The heuristic conversion of the EDR corpus intoan HPSG treebank consists of the following steps. Asentence ‘((kare/NP-he wo/PP-ACC) (koro/VP-killshi/VP-ENDING ta/VP-DECL))’ ([I] killed him yes-terday) is used to provide examples in some steps.Phrase type annotation Phrase type labels suchas N[r]
which pattern it is most similar—a task easier forhumans than WSD is. This principle seems par-ticularly promising for verbs as words expressingevents, which resist the traditional word sense dis-ambiguation the most.A novel approach to semantic taggingWe present the semantic pattern recognition asa[r]
Noun CompoundFigure 1: System Architecturewhether it is reasonable to expect that all NCsshould be interpretable with a fixed set of semanticrelations.Based on the pioneering work on Levi (1979)and Finin (1980), there have been efforts in com-putational linguistics to arrive at largely task-sp[r]
tatively reliably by non-domain experts when us-ing a two-way distinction. They also perform firstexperiments on automatic classification.Medlock & Briscoe (2007) develop a weaklysupervised system for hedge classification in avery narrow subdomain in the life sciences. Theystart with a small se[r]
hypothesis to become apparent. It should alsonot be too large to become inefficient in termsof cost, recording fatigue, or hours of daylight.As a general rule, several small quadrats will givemore information than few large quadrats ofthe same total area, but will be more costly toidentify and measu[r]
and Weld, 2007; Bellare and McCallum, 2007; Palet al., 2007). Our work was inspired by Mintz et al.(2009) who used Freebase as a knowledge base bymaking the DS assumption and trained relation ex-tractors on Wikipedia. Previous works (Hoffmannet al., 2010; Yao et al., 2010) have pointed out thatthe D[r]
Workshop on Computational Terminology. Montreal, Canada.Lin, D. 1998b. Automatic Retrieval and Clustering of SimilarWords. In Proceedings of COLING-ACL98. Montreal, Canada.Lin, D. (1994). Principar - an Efficient, Broad-Coverage,Principle-Based Parser. In Proceedings of COLING-94. Kyoto,Japan[r]
grammar recognition performance.3 Language modellingTo generate the different trigram language modelswe used the SRI language modelling toolkit (Stol-cke, 2002) with Good-Turing discounting.The first model was generated directly from theMP3 corpus we got from the GF grammar. Thissimple SLM (na[r]
and grammar, is essential for language learners.However, intonation has been less-emphasizedboth in classroom and computer-assisted languageinstruction (Chun, 1998). Outside of tone lan-guages, it can be difficult to characterize the fac-tors that lead to non-native prosody in learnerspeech, and it i[r]
selves. Considering this, it is highly possible thatprecision will improve as the size of the feedbackcorpus increases.5 ConclusionsThis paper has proposed a feedback-augmentedmethod for distinguishing mass and count nounsto complement the conventional rules for detect-ing grammatical errors. The ex[r]
tive ratio that favours one DEO per context. Ourmethod is not guaranteed to increase classificationcertainty between iterations, but we will show thatit does increase certainty very quickly in practice.The key observation that allows us to resolvethe tension between trusting the initialization andenf[r]
the SC of a proper name is computed fairly accu-rately using a named entity (NE) recognizer, manyresolvers simply assign to a common noun the first(i.e., most frequent) WordNet sense as its SC (e.g.,Soon et al. (2001), Markert and Nissim (2005)). Itis not easy to measure the accuracy of this heuristi[r]
to linguistic enquiry, and vice versa—has always seemed anomalous. However, thisanomaly may have roots deeper than being simply a matter of turf wars. Recently,Burrows (2003) has characterized the split between scientific method and literarycriticism as a comparison between Descartes’s retirin[r]
quences of POSs as grammatical errors in the writ-ing of learners. As just discussed above, existingtechniques often fail in sequences of POSs that havea grammatical error. For instance, an existing POStagger likely tags the sentence I frightened. as I/PRPfrightened/VBD ./. as we have just seen, and[r]
points throughout the year, relative to what they need to learn by the end of the year. ese assessments can provide an early prediction of how well students might perform on the year-end state-mandated test; as such, they are referred to as prospective. Another type of local assessment is retrospec[r]
Chapter 6Control Technology of Solidification and Cooling in theProcess of Continuous Casting of SteelQing Liu, Xiaofeng Zhang, Bin Wang andBao Wang Additional information is available at the end of the chapterhttp://dx.doi.org/10.5772/514571. IntroductionSolidification and cooling control, which is[r]