Báo cáo khoa học: "Using Conditional Random Fields For Sentence Boundary Detection In Speech" potx

... Proceedings of the 43rd Annual Meeting of the ACL, pages 451–458,Ann Arbor, June 2005.c2005 Association for Computational LinguisticsUsing Conditional Random Fields For Sentence Boundary Detection ... not an-notated according to the guideline used for the train-ing and test data (Strassel, 2003). For BN, we usethe training corpus for the LM for speech recogni-tion. For CTS, we use the Penn ... knowledge sourcesare integrated for detecting sentence boundaries in speech. We have shown that a discriminativelytrained CRF model is a competitive approach for the sentence boundary detection task....

Báo cáo khoa học: "Using Conditional Random Fields to Predict Pitch Accents in Conversational Speech" pptx

... CRF:All in Table5). We get about 0.5% increase in accuracy, 76.1%with a window of size w = 1.Using larger windows resulted in minor increases in the performance of the model, as summarized in Table ... For our textual phonological features, we in- cluded the number of syllables in a word and thenumber of phones (both in citation form as well astranscribed form). Instead of position in a sentence, we ... that in uence pitchaccent placement in natural, conversational speech in a sequence labeling setting. We introduce Con-ditional Random Fields (CRFs) to pitch accent pre-diction task in order...

Báo cáo khoa học: "Using Conditional Random Fields to Extract Contexts and Answers of Questions from Online Forums" docx

... linguistic expression used by a ques-tioner to request information in the form of an an-swer. The sentence containing request focus iscalled question. Context are the sentences contain-ing ... on forum data.Experimental results show that 1) Linear CRFs out-perform SVM and decision tree in both contextand answer detection; 2) Skip-chain CRFs outper-form Linear CRFs for answer ﬁnding, ... participate in dis-cussions. The availability of vast amounts of threaddiscussions in forums has promoted increasing in- terests in knowledge acquisition and summarization for forum threads. Forum...

Báo cáo khoa học: "Scaling Conditional Random Fields Using Error-Correcting Codes" docx

... experimentsimply for illustrative purposes: many columns in this code were unnecessary, yielding only aslight gain in performance over much simplercodes while incurring a very large increase in training ... errors by individual models.Error-correcting CRF training is muchless resource intensive and has a muchfaster training time than a standardlyformulated CRF, while decodingperformance remains quite ... given above for 12classiﬁers. A series of CRFs are trained, eachon a relabelled variant of the training data. Therelabelling for each binary CRF maps the labelsinto binary space using the relevant...

Báo cáo khoa học: "Training Conditional Random Fields with Multivariate Evaluation Measures" potx

... CRF training.1 Introduction Conditional random ﬁelds (CRFs) are a recentlyintroduced formalism (Lafferty et al., 2001) for representing a conditional model p(y|x), whereboth a set of inputs, ... isozaki}@cslab.kecl.ntt.co.jpAbstractThis paper proposes a framework for train-ing Conditional Random Fields (CRFs)to optimize multivariate evaluation mea-sures, including non-linear measures suchas F-score. Our proposed ... training criterion. Therefore, weminimize the following loss function for the MAPcriterion training of CRFs:LMAPλ= LMLλ− log p(λ). (2)There are several possible choices when selectinga...

Báo cáo khoa học: "Using Deep Morphology to Improve Automatic Error Detection in Arabic Handwriting Recognition" pot

... Effect of Training Data Size In order to allow for rapid examination of multi-ple feature combinations, we restricted the size ofthe training set (S) to maintain manageable train-ing times. ... training data. We test this assumption bytaking the best-performing feature sets from Table 5and training new models using twice the trainingdata {S=4000}. The results are shown in Table 6. In ... models were trained on the {S=4000,H=10} training set.train is less than dev and train: 63% versus 81%,respectively on {S=4000}.5.5 Preliminary Combination Analysis In a preliminarily investigation...

Tài liệu Báo cáo khoa học: "Using Automatically Transcribed Dialogs to Learn User Models in a Spoken Dialog System" doc

... distributions in Figure 1, except the user model,are known. The ASR confusion model was esti-mated by transcribing 50 randomly chosen dialogsfrom the training set in Section 4.2 and calculat-ing the ... reasonable,hand-crafted value for θ, and then generated syn-thetic dialogs by following the probabilistic processdepicted in Figure 1. In this way, we were able tocreate synthetic training sets of varying sizes, ... generated dialogd in each training/test set consisted of a sequence ofvalues for all the observed and unobserved variables:d = (S0, U0, A0,˜A0, . . .). For a training/test set D, let...

Tài liệu Báo cáo khoa học: "Trimming CFG Parse Trees for Sentence Compression Using Machine Learning Approaches" pptx

... sometimes elim-inate the original meaning by incorrectlyremoving important parts of sentences, be-cause trimming probabilities only dependon parents’ and daughters’ non-terminals in applied CFG ... tree of the original sentence is marked if it exists in the compressed sentence. In the ﬁgure, the marked terminals arerepresented by circles. Second, each non-terminal in the original parse tree ... Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions, pages 850–857,Sydney, July 2006.c2006 Association for Computational LinguisticsTrimming CFG Parse Trees for Sentence...

Báo cáo khoa học: "Using WordNet to Automatically Deduce Relations between Words in Noun-Noun Compounds" docx

... word in WordNet is linked to a set of senses, with eachsense identifying one particular meaning of thatword. For example, the noun ‘skin’ has senses rep-resenting (i) the cutis or skin of ... sense-pair, using the ﬁrst term ofthe score tuple as the main key for comparison(lines 14 and 15), and using the second term asa tie-breaker (lines 16 to 18). If the PRO score for relation R in the ... human beings, (ii)the rind or peel of vegetables or fruit, (iii) the hideor pelt of an animal, (iv) a skin or bag used as acontainer for liquids, and so on. Each sense con-tains an identifying...

Báo cáo khoa học: "Using Predicate-Argument Structures for Information Extraction" ppt

... parses in TreeBank Release 2. As named entity informationis not available in PropBank/TreeBank we taggedthe training corpus with NE information using anopen-domain NE recognizer, having 96% ... weremainly responsible for the improvements. This isthe rationale for including them in the FS2 set.We also were interested in comparing the results23These results, listed also on the last line ... Deci-sion tree learning provided by C5 (Quinlan, 2002)has built in support for boosting. We used it andobtained improvements for both tasks. The best F-measure obtained for argument constituent...

Xem thêm