Báo cáo khoa học: "Using Parse Features for Preposition Selection and Error Detection" ppt

... 353–358,Uppsala, Sweden, 11-16 July 2010.c2010 Association for Computational LinguisticsUsing Parse Features for Preposition Selection and Error DetectionJoel TetreaultEducational Testing ServicePrincetonNJ, ... 65.2lexical +parse 68.1all features ( +Parse) 68.5Table 2: Accuracy on preposition selection task for various feature combinationsThe Preposition Head and Complement Mixed features are created by ... T&C08 and +Parse models on the task of preposition selection onwell-formed texts written by native speakers. For every preposition in the test set, we compare thesystem’s top preposition for that...

Báo cáo khoa học: "Novel Semantic Features for Verb Sense Disambiguation Dmitriy Dligach" ppt

... (1) the lexical features only (2) the lexical and the DDN features (3) the lexical and the object features. The object features consist of the head word of the NP object and the head word's ... % Error Rate, % All Features – DDN 82.38 17.62 All Features 82.88 17.12 Table 5. Performance of the full-fledged VSD system The DDN features improved performance by 0.5% (3% drop in error ... Contribution of Semantic Features The DDN features outperform the other two types of semantic features used separately and in conjunction. The difference in performance is sta-tistically...

Báo cáo khoa học: "A Graphical Interface for MT Evaluation and Error Analysis" doc

... systems,documents and segments. This kind of results isvaluable for rapid evaluation and ranking of trans-lations and systems. However, it is unfriendly for MT developers that need to manually analyze and compare ... review the type of information used byeach family of measures according to their classiﬁ-cation, and how this information can be used for MT error analysis purposes.Lexical information. There ... prepositions between areference and a candidate). In addition, depen-dency and constituency parse trees allow for captur-ing other aspects of the translations. For instance,DP-HCWM is a speciﬁc...

Báo cáo khoa học: "Hierarchical Sequential Learning for Extracting Opinions and their Attributes" ppt

... i)where gO and g′Oare feature vectors defined for Opinion extraction, gP and g′Pare feature vectorsdefined for Polarity extraction, and gS and g′Sarefeature vectors defined for Strength ... com-mon features and polarity extraction features inSection 3), and another that extracts opinion ex-pressions only with the intensity attribute (usingcommon features and intensity extraction features in ... Transition Features Transition features are employed to help withboundary extraction as follows:Polarity Transition Features Polarity transition features are features that areused only for g′O(α,...

Tài liệu Báo cáo khoa học: "Web-Scale Features for Full-Scale Parsing" doc

... and this work, and the relative error reduction. Results are for dependency parsing on the dev set for iters:5,training-k:1.tal errors break down by gold head. For example,the 12.1% total error ... NN IN time, for some NN IN money, for annual NNS IN reports, onTable 8: The highest-weight features (thresholded at a count of400) of the before-word schema for a noun head and preposition argument ... then integrate our features into full-scale dependency and con-stituent parsers. We show relative error re-ductions of 7.0% over the second-order depen-dency parser of McDonald and Pereira (2006),9.2%...

Tài liệu Báo cáo khoa học: "Using Structural Information for Identifying Similar Chinese Characters" pdf

... characters for charac-ters that appeared in Figure 2 and Figure 3. For instance, we can tell that “搞” and “篙” are similar because their Cangjie codes share “卜口月”, which in fact represent “高”. Unfortunately, ... complete codes for the components in our database. For in-stance, in our database, the codes for “巠”, “脛”, “徑”, “頸”, and “勁” are, respectively, “一女女一”, “月一女女一”, “竹人一女女一”, “一女女一一月山金”, and “一女女一大尸”. ... have a machine readable lexicon that provides information about pronuncia-tions of characters and when we ignore special pat-terns for tone sandhi in Chinese (Chen, 2000). In contrast, it...

Tài liệu Báo cáo khoa học: "Kinds of Features for Chinese Opinionated Information Retrieval" pdf

... thefound set of features for text classification (index-ing) for an OIR query of the first level (finds opin-ionated information) and for an OIR query of thesecond level (finds opinionated information ... world(the mainland China, Taiwan, Hong Kong, Singa-pore and numerous Chinese communities all overthe world) is getting more and more influential overthe world economy and politics.We therefore believe ... a classification-based approach for sentiment extraction and sentiment polarity detec-tion (for example, Pang et al. (2002), Turney (2002),Kim and Hovy (2004) and others), however, the re-search...

Tài liệu Báo cáo khoa học: "Using Confidence Bands for Parallel Texts Alignment" pptx

... confidenceinterval [ ax + b – error( x); ax + b + error( x)],where ax + b is the linear regression line equa-tion and error( x) is the error admitted at the xco-ordinate. The upper and lower limits of ... (pt) and Spanish (es).2 Webpage address: curia.eu.int3 The same languages as those in footnote 1 plusFinnish (fi) and Swedish (sv).4 No Written Questions and Debates texts for Finnish and ... classes. For example, for Figure 1, wehave 3338 points and the distances betweenexpected and real positions range from 0 to35997. Thus, the number of classes is1 + log23338 ≅ 12.7 → 13 and the...

Báo cáo khoa học: "Linguistically Motivated Features for Enhanced Back-of-the-Book Indexing" ppt

... that uses a Na¨ıve Bayes learn-ing algorithm and several features, including infor-mation theoretic features such as tf.idf and positional features reﬂecting the position of the words with ... IntroductionBooks represent one of the oldest forms of writ-ten communication and have been used since thou-sands of years ago as a means to store and trans-mit information. Despite this fact, given ... original ordering. For training and evaluation purposes, we used arandom split of the collection into 90% training and 10% test. This yields a training corpus of 259 docu-ments and a test data...

Báo cáo khoa học: "Cognitively Motivated Features for Readability Assessment" pot

... presents the average error results for the baseline system and our three regression models. We can see that the model trained on the shallow and parse- related features out-performs the model trained ... primarily use it for evaluation pur-poses (not for training the readability models). 5 Linguistic Features and Readability We now describe the set of features we investi-gated for assessing ... representation and deciding how each entity is related to others. On the other hand, when a text refers to few entities, less work is required both for semantic encoding and for in-tegrating...

Xem thêm