Báo cáo khoa học: "Contrastive Estimation: Training Log-Linear Models on Unlabeled Data∗" potx

... Arbor, June 2005. c 2005 Association for Computational Linguistics Contrastive Estimation: Training Log-Linear Models on Unlabeled Data ∗ Noah A. Smith and Jason Eisner Department of Computer ... neighborhood). This table contained no big surprises; we note only that most models were the best on their own criterion, and among unsupervised models, LENGTH performed best on t...

Ngày tải lên: 23/03/2014, 19:20

... 1991 (91a) and the second half of 1998 (98b). For instance, the training- test configuration (S i=91a 98b , U i=91a 98b , T j=98b ) represents the training- test configuration where the test set ... occurrence of the name in the contempo- rary unlabeled data contributes to its correct clas- sification in the test set. 5.2 Is more older unlabeled data better? The second question we addresse...

Ngày tải lên: 20/02/2014, 09:20

... cell. Saturation of 14 CO 2 evolution rate by the mutant strain as a function of concentration depended on sub- strate. The rate of 14 CO 2 evolution was saturated when the concentration of glucose ... uniform con- centrations of all pathway intermediates at K M , the approximate concentration traditionally assumed to be necessary to permit pathways to function optimally. Along with other c...

Ngày tải lên: 19/02/2014, 18:20

... relation is called a joint or conditional (probabilistic) relation and constitutes a statistical model. Such models can be efficiently restricted, manipulated or combined using rational operations as ... a smaller conditional relation (Gerde- mann and van Noord, 1999), 5 (5) by conditionaliza- tion of a joint relation as discussed below. A central technique is to define a joint relation as a...

Ngày tải lên: 20/02/2014, 21:20

... NP, personal pronoun, demonstrative pronoun, possessive pronoun, proper name) 4. ante agree agreement in person, gender, number 5. ante semanticclass semantic class of antecedent (human, concrete ... this, we used a conventional C4.5 decision tree classi- fier (J48 implementation, default settings) on labeled training data sets of the same size used for the re- spective Co -Training exper...

Ngày tải lên: 20/02/2014, 21:20

... givenness, to indicate that it constitutes contrastive information. 2 Predicting contrastive accent In this section I discuss two approaches to predicting contrastive accent, which were put ... placement of con- trastive accent is essential for the assign- ment of correct accentuation patterns in spoken language generation. I discuss two approaches to the generation of contrastive .....

Ngày tải lên: 22/02/2014, 03:20

... 1 http://homepages.inf.ed.ac.uk/lzhang10/maxent_tool kit.html Feature Name Description Phrase Type Phrase type of the constitu- ent Position+Predicate Voice Concatenation of constitu- ent position relative to verb and verb voice Predicate ... in terms of improvement. We use the second op- tion for all experiments here. In each iteration, one can label entire the unlabeled data or...

Ngày tải lên: 07/03/2014, 22:20

... Co -Training for Monolingual Hyponymy-Relation Acquisition Jong-Hoon Oh, Kiyotaka Uchimoto, and Kentaro Torisawa Language Infrastructure Group, MASTAR Project, National Institute of Information ... 2 Translate reliable parts of classification results Training Training Training Training … … Translate reliable parts of classification results Figure 1: Concept of bilingual co -training...

Ngày tải lên: 08/03/2014, 00:20

... calculate the sum- mations in Equation 1 unless an approximation is applied. Several feature functions will be defined on top of the posterior distribution. One of them is based on HMM word alignment ... active features in our transla- tion engine include translation models in two di- rections, lexicon weights in two directions, lan- guage model, lexicalized distortion models, sen- tenc...

Ngày tải lên: 08/03/2014, 01:20

... subcategorisation and selectional preferences, collocations, and verb alternations), 496 and for NLP applications in sparse data situations. In this paper, we provide an evaluation based on a language ... class model under various conditions, we used different fractions of the training data in different training regimes. Because of time and memory constraints, we only used training tu...

Ngày tải lên: 08/03/2014, 01:20