id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_99100 | Furthermore, the matrices V and U are obtained by a simple iteration, from Lee (2000), where the initial matrices and are updated. | it is impossible to find the optimal initial value. | neutral |
train_99101 | We used data sets that had less than 5,000 data elements. | (2005) show the equivalence between spectral clustering and NMF. | neutral |
train_99102 | LSI constructs orthogonal space from document space. | in addition, the accuracy of the obtained clustering result was higher than those of NMF, CLUTO and Mcut. | neutral |
train_99103 | Regarding the object function, initial accuracy must be improved. | as shown in Table 2, there are some data sets for which the clustering accuracy of NMF-rfn is worse than that of Mcut. | neutral |
train_99104 | In asymmetrical constructions, the first complement of the mother-category is co-indexed with the first complement of V1. | (2) nanwu-e mek-ta 'divide and eat', kkulhi-e mek-ta 'boil and eat', mandul-e mek-ta 'make and eat', cap-a mek-ta 'catch and eat', cip-e mek-ta 'pick up and eat', ssip-e mek-ta 'chew and eat', kwu-e mek-ta 'broil and eat', ppal-a mek-ta 'suck and eat',… The purpose of this study is to provide an overall picture of KSVCs within the framework of the unification-based grammar 2 , in particular, Head-driven Phrase Structure Grammar (HPSG) 3 . | neutral |
train_99105 | I treat the construction like (4c) as KSVCs on the ground of this difference. | 17 constraints have been built, which support the type hierarchies for Korean serial verb constructions. | neutral |
train_99106 | The name might not be important for an analysis, the reference system of the origin however hints on identities communities maintain. | reference systems and references, as opposed to textual or editorial annotations, constitute a conceptual framework for corpus annotation which are important for cross-cultural and cross-linguistic comparisons. | neutral |
train_99107 | The references to time on the tombstone, for example, can be used to analyze historical developments in the form and content of tombstones. | for some researchers looking at tombstones, neither the exact date (which we will call reference), nor the exact wording (which we will call meaning) of a date might be interesting, but instead what kind of calendar is used. | neutral |
train_99108 | The finding that the SEP is not distinguished from the SEB regarding TTR or D, indicates that prime ministers refer to a similar diversity of political content and in a similarly balanced way in the two types of addresses. | the results indicate that the SEP has a significantly larger N than the SEB, and the two types of addresses are not distinguished by D. 9 In Figure 3, the solid line represents the mean value of ttR of the SEP, the dashed line represents the mean value of ttR of the SEB and dotted lines represent 95% confidence intervals of ttR of both types of addresses, incremented by 500 tokens constructed by Monte Carlo simulations. | neutral |
train_99109 | He succeeded politically partly because he was able to attract people by his way of speaking (Otake, 2006). | table 1 shows the number of addresses, the number of tokens and the number of types of the SEP and the SEB after 1953. | neutral |
train_99110 | An analysis was conducted of the learners' differential choice of modal markers for each block of 10 questions that required each modal marker as the correct answer from the viewpoints of L1 transfer and the development of interlanguage. | this implies that the global cue is the core/common meaning between two usages as the monosemic approach stated. | neutral |
train_99111 | We divided the data into three equal-sized folds in the same manner as the previous work (Pang et al., 2002). | the values are determined from the results obtained from training data by using each classifier constructed from it. | neutral |
train_99112 | Boosting is one of the most famous techniques in ensemble learning methods (Freund and Schapier, 1996). | we obtained knowledge that the misclassifications of each classifier are often different, that is exclusive misclassifications. | neutral |
train_99113 | Currently, Wikipedia is available in 252 languages. | an NE candidate is first input to the Korean Wikipedia, and the title of the matched article's Chinese version is treated as the NE's translation in Chinese. | neutral |
train_99114 | Hence, it substantially improves the performance of KCIR. | our method's performance on KCIR is over five times better than that of a dictionary-based translation system. | neutral |
train_99115 | This phenomenon makes Korean-Chinese information retrieval more difficult because reporters usually use one or two common translations to write articles. | these examples show that there may be many Chinese translations for a Korean name. | neutral |
train_99116 | In contrast to the traditional analyses, all of these studies base their arguments on spoken data beyond sentence level, making their analyses more convincing. | the use of demonstratives as filled pauses (FPs) has been sensed by many linguists. | neutral |
train_99117 | Furthermore, both Li and Chang have argued that a grammaticalizational effect concerning semantic shift is demonstrated in the use of Taiwanese proximal deixis. | in addition, a careful scrutiny of the transformation in its syntactic distribution or semantic content also supports the idea that hitlo is undergoing a grammaticalizational effect. | neutral |
train_99118 | When used to refer to an object in an extralinguistic situation, demonstratives often accompany gestures or eye contact. | previous studies relating to Taiwanese demonstratives frequently restrict their analyses to syntactic patterns or semantic diversities. | neutral |
train_99119 | Also, still some other common functions which demonstratives can serve such as an anaphoric usage are not investigated in this paper. | as for the age effect, there was no clear evidence shown in the overall comprehension scores. | neutral |
train_99120 | To interpret NPs on non-subject position one extends the domain of type <1 > quantifiers: they are treated as reducers of arity of their arguments, that is as functions which when applied to n-ary relations give (n-1)ary relations. | this sentence can mean, at least theoretically, either (2a), the co-referential reading of reflexive, or (2b), the bound reflexive: (2a) Only Leo washed Leo. | neutral |
train_99121 | Biotech L.L.C., Myelodysplastic Syndromes (MDS), Azacitidine, Lifezar 50 mg & 100 mg Tablet). | the researchers' analysis proceeded from the assumption that advertising English should be represented as a continuum of text functions fluctuating between "informing" and "manipulating" in accordance with the idea that advertising is an example of covert communication. | neutral |
train_99122 | Because they refer to the company (GenAsia) or product (Losartan), (9) and (10) make use of NPs (e.g. | the absence of subjects and auxiliaries may make these ads more relevant to the reader because less processing effort is expended when reading them (Wilson and Sperber, 2004). | neutral |
train_99123 | With respect to distinguishing between glossing and translation, meaning is assigned to the text through glossing only indirectly through the unavoidable fact that the ID-gloss, which is primarily intended to identify a sign, actually uses an English word that bears a relationship to the meaning of the sign. | generally speaking, transcriptions are usually created as reference points for, or stages in, linguistic analysis, such as in the creation of scripts for writing systems, for phonological analysis, or for grammatical analysis. | neutral |
train_99124 | Annotators log in to a special researchers' reference view which includes much more information than in the public view (including the ID-gloss), as well as many more additional signs (e.g., variant signs and newly identified signs). | the ID-gloss is still not intended as a translation. | neutral |
train_99125 | Consider the logical contrast of contradiction involving PA in (25a) and the compatibility between the negative pair of opposite gradable adjectives in (25b): (25) a. | the following conjunctions/discourses with PA type conjunction/DM are not acceptable: haciman swusek-un mot haysse 'But she didn't ace the exam'). | neutral |
train_99126 | In addition, our system also made use of the HMM/NN hybrid phone recognizers provided by the Brno University of Technology (BUT) 2 . | we interpret the acoustic sound classes to represent some general vocal tract configurations in producing various speech sounds. | neutral |
train_99127 | One might claim that this job is too big to be practical, but efforts of even greater size have happened once people realize a need (wikipedia for example), and join together on community projects. | we use wordNet as our English lexicon (Fellbaum, 1991). | neutral |
train_99128 | In the above example, what is presupposed in the traditional approach applies to (20), which share exactly the same property as the hp. | this analysis, building upon the scalar model proposed by Fillmore, et al (1988) and Kay's (1990) analysis of even, extends Shudo's analysis (1998, 2002) of mo and claims that even behaves like scalar mo and to. | neutral |
train_99129 | The Creator has limited time and resources and thus prioritizes the creation of resources that are deemed relevant. | any resource that is a tool that supports the creation of language resources is also defined to be a language resource. | neutral |
train_99130 | We propose that there are six necessary and sufficient conditions for the use of a language resource. | the sustainability of language depends in part on the sustainability of the language resources that contribute to language development, which will lead to the production of new language resources that can in turn enable further development, and so the cycle of sustainability continues. | neutral |
train_99131 | It checks if the verb_phrase object contains an input string that matches a causative suffix, instantiates, when a matching string is found, a causative suffix object in the Suffix class, and registers the suffix object as a component of the verb phrase object, before returning to the system the verb phrase object with the added information. | when it has been found and registered, that means a suffix that may appear before the inflection has also been recorded in the verb phrase instance. | neutral |
train_99132 | For instance, "書かした" (/kakasita/ with the meaning of having caused someone to write) is a combination of /kak/ meaning writing, /as/ indicating causation, and /ta/ representing the past inflection, but it is incorrectly parsed as illustrated below: Basic form The meaning of causation contained by this string is not regognized in this analysis, and it would be difficult to obtain the real meaning of "書かした" (/kakasita/ with the meaning of having caused someone to write) from this analysis. | the system can be embedded in a larger system, and be used when the larger system starts parsing verb phrases. | neutral |
train_99133 | Both entropies appear to address the issue of occurrence concerning CCB JJD and CCB JJD NNC. | klein(2005) experimented with a combination of the said context-constituency model and a dependency model klein and Manning(2004), which was then applied to English, German, and Chinese. | neutral |
train_99134 | In sentence (13), the role of the non-subject actor Lisa is determined by the matrix verb binalak 'planned' although the actor verb sumayaw 'dance' occurs closer to this argument. | although the argument of the linked core 'sumayaw 'dance' does not appear in the clause syntactically, we can conclude that the semantic role of this argument will be absolutive as assigned by its verb. | neutral |
train_99135 | (4) Three-way ANOVA showed a significant main effect in S1 [F(1, 69)=5.29, p<0.05], indicating that words filling up semantic gaps are different in their log normalized ratios from those not filling up a gap. | secondly, as unobtrusiveness, in fact, explains for various levels of factors, including phonology, morphology, semantics, and foreign borrowings, its operational criteria would also involve different levels of information in the hope that the scores would be close to the original model's prediction. | neutral |
train_99136 | Moreover, storing the actual statements used will provide consistency in the generated text should the user choose to revisit the same object. | it is applicable to all categories and subcategories. | neutral |
train_99137 | It is possible that speakers of Mandarin are implicitly sensitive to this property of -zi and that -zi could function as a cue to individuation. | experiment 2 then examined how adult speakers interpret both -zi and ge by contrasting them with bare nouns in two word learning tasks. | neutral |
train_99138 | If they selected neither, they were asked to describe in writing what kind of entity the noun could denote. | syntactic cues to individuation may emerge much later in Mandarin than in English One problem in comparing classifiers to count syntax is that whereas count syntax specifies only individuation, classifiers also encode item-specific conceptual information, such as the shape, animacy, etc. | neutral |
train_99139 | The sense means Korean translation word for the ambiguous English word. | the discussion of the previous sections will be summarized in the concluding section 6. | neutral |
train_99140 | The context information was converted to sense vectors. | we needed a new step for semi-automatic tuning of translation engine modules to cut tuning time. | neutral |
train_99141 | Finally, the classifiers are combined together with the three different weighted voting schemes. | frequently occurring words are rarely NEs. | neutral |
train_99142 | For example, less fluent sentences that are shorter were assigned higher ranks than the longer fluent ones. | the same set was also ranked by three language modelers using the different versions of Language Evaluation. | neutral |
train_99143 | n-gram language modeling is a popular statistical language modeling (SLM) technique used to improve performance of various natural language processing (NLP) applications such as speech recognition (SR), machine translation (MT), and information retrieval (IR). | the initial results showed that the thresholds were not effective in its purpose such that only an insignificant amount of proxy trigrams were being filtered out. | neutral |
train_99144 | The classification of Hanzi semantic symbols Under "telic" and "participating," we add a column for "related semantic symbol" to show and link the related deriving concepts. | in conclusoin, our study of four hoofed anminals support our original thesis that the basic concepts as conventionalized by radicals represents a domain ontology of a cluster of concepts derived and marked by that radical. | neutral |
train_99145 | An adjective tall now means a set of degrees which the individual x has on the scale of tallness. | the only criteria which distinguish adjectives from verbs and nouns could be whether they appear in comparative construction or not. | neutral |
train_99146 | As mentioned above, a simple adjective cannot construct a predicate in Chinese. | degree abstraction is lexical in Japanese-type languages as argued in Beck et al (2004). | neutral |
train_99147 | System performance for SENSEVAL-4 (Task 17) is based on the average results on individual target words from all systems, as well as results from two individual systems (System 1 and 4, both using Support Vector Machines), reported in Pradhan et al. | we must also look into concreteness at the sense level in order to study its impact on the information demand of individual words in wSD, and hence varied information susceptibility (Kwong, 2005). | neutral |
train_99148 | Three human judges were asked to rate the words and senses in the sample on a 7-point scale of concreteness, with 1 for highly abstract, and 7 for highly concrete. | looking at the predominant sense, a similar situation to SENSEVAL-1 data was found, i.e., those with the first sense at either end of the continuum are better disambiguated. | neutral |
train_99149 | In Korean, there are two groups of lexical items which realize coherence relations. | for the test we collected such cue phrases from different sources. | neutral |
train_99150 | At first, we can divide coherence relations into 'cause' and 'addition' according to the 'source' notion. | we collected the items from the tagged corpus 'Sejong corpus'. | neutral |
train_99151 | Its basic idea is to utilize the distinct distributions of terms and non-terms in different corpora to facilitate term extraction. | each ordinance can be regarded as a sub-corpus for a "field", e.g., "Public Finance Ordinance", "Forest and Countryside Ordinance", "Hospital Authority Ordinance", and "Marriage Ordinance". | neutral |
train_99152 | Terms are linguistic representations of domain specific concepts to encode our special knowledge about a subject field. | a term candidate may get a negative score from (1) above, which will, unreasonably, weaken the total sum of scores from the comparison to all other background corpora. | neutral |
train_99153 | {not(a,b) ∧ property(b,VAR1)} The hypernym axiom (Table 5) is designed for nouns and verbs. | we adopt this approach and extend it with some post-processing of the semantic representations with a suite of axioms. | neutral |
train_99154 | These semantic representations are designed to abstract away syntactic variations, allowing sophisticated automated processing of Indonesian texts. | the complex machinery described in the last three subsections essentially plays one role: to abstract away the syntactic variations from paraphrases that essentially convey the same thing. | neutral |
train_99155 | The antonym axioms (Table 4) assert new literals explicitly stating the negation of the opposing concept of adjectives appearing in the semantic representation. | it asserts event(e2,memakan), agent(e2,a), and patient(e2,p). | neutral |
train_99156 | Using Hoppes and Thompson's (1980) transitivity hypothesis and Dixon's (1994) syntactic-semantic primitive cited in Adeva's (2005) study, the results revealed that in actor focus, two affixes (mi-/ni-~m-replesive) were evident and three affixes (i-, -un, -an) were evident in goal the focus. | hence, a study on semantic change of Cebuano words should be conducted. | neutral |
train_99157 | To validate the reason in using a metaphor in writing Cebuano literature such as short stories and feature articles, five (5) informants were asked about their personal reasons in using a metaphor in writing short stories and feature articles. | a study on the semantic change of Cebuano words should be conducted to fill this gap. | neutral |
train_99158 | Note that the verbalized predicates in (10b) and (11b) show the difference in passivizability. | another question here is why (11b) cannot be passivized as shown in (12b) while (13b) can as shown in (13a). | neutral |
train_99159 | This method obtains good accuracy. | in their analysis, reviews almost always include the attitude, nominative and object. | neutral |
train_99160 | We form lexicons from syntactic pieces (Aoki and Yamamoto, 2007). | we regarded a blog as being an emotion-blog when it satisfied both requirements 1 and 2. e This section constructs the emotion classifier model. | neutral |
train_99161 | A syntactic piece is the minimum unit of syntactic structure of an expression. | most blogs fail to fulfil the two requirements. | neutral |
train_99162 | We found that messages that were categorically assigned to the true positive group all contain several technical keywords that also occurred in the anchor document; we initially assumed that the presence of these keywords is the primary reason why these messages were classified as relevant. | this characteristic can also be observed in most messages that contained technical discussions. | neutral |
train_99163 | Though trend scores of words are updated whenever new documents arrive, 6 v (the vector of d 6 ) is never updated even if the new documents contain same words in d 6 . | moreover, ( ) The first term For simplification, we regard the current (t=t n ) difference bet een two curves in Figure 2 as one. | neutral |
train_99164 | This layer expresses a pack of POS tags in each bunsetsu. | the double-headed arrows denote the link between a node and a sub-graph enclosed with a dashed line. | neutral |
train_99165 | In addition, we change the value of n in each step of the bootstrap approach. | in this paper, we proposed a method of sentiment sentence extraction based on a graph-based approach, called Hierarchical Directed Acyclic Graph (HDAG). | neutral |
train_99166 | (2006) have proposed methods for opinion extraction. | also they have proposed a method for building lexicon for sentiment analysis (Kaji and Kitsuregawa 2007). | neutral |
train_99167 | Also we evaluated our method with 5 recognizers. | the vocabulary file defines words with its pronunciations (i.e., phoneme sequences) for each category. | neutral |
train_99168 | To apply the method in related work (Utsuro et al. | 2004) and to extract similar word pairs (Komatani et al. | neutral |
train_99169 | And the output of the process is the word clusters wherein words with similar senses are grouped into a cluster. | to the size of the corpora and lexicon used, the system used significantly smaller sized corpora (100K for source and 300K for target), versus the resources used in the related literature, wherein the smallest corpora resource used was 39M (Sadat et al., 2003) and the lexicon resource was 16,380 (Rapp, 1999). | neutral |
train_99170 | Analogously, pronoun j3 also co-refers with i1 1 , and pronouns j2 and j4 both co-refer with i2: Scientists i1 know that the phenomenon i2 occurs once every three to seven years i3 : they j1 can detect when it j2 is coming, they j3 perceive when it j4 is going away. | in order to select the most useful features for solving the problem at hand we started by taking the entire set of 20 learning features into account. | neutral |
train_99171 | As we are still in the process of defining which precise features are applicable to the task at hand, our investigation is currently based on the training data set only, leaving the test data reserved for future use. | the above confusion matrix is to be interpreted as follows: 2141 instances (being 311 coreferent and 1830 non co-referent) were correctly classified (89.47% success rate); 125 coreferent instances were misclassified as non co-referent (5.22%), and 127 non co-referent instances were misclassified as co-referent (5.31%.) | neutral |
train_99172 | (2001), McCarthy and Lehnert (1995) and Ng & Cardie (2002). | results of a standard decision-tree induction algorithm using the new features are presented in Section 4. | neutral |
train_99173 | In addition to that, as we do not intend to explicitly write any anaphora resolution algorithms or rules (but rather induce them automatically) our work is mainly related to machine learning approaches such as Soon et. | it should be pointed out that when building the present training data, the annotators were selective in the choice of training instances of anaphora to be addressed. | neutral |
train_99174 | ten-fold cross-validation decision-tree induction (Quinlan, 1993). | ly, our work does not include instances of reference to compound antecedents (e.g., "The boy and the girl" referred as "They"), which may partially explain the higher success rates 3 . | neutral |
train_99175 | For a rule-based MT, ambiguitiesresolving rules seem to play important roles, whereas for a statistics-based MT, rather the rules that normalize the source expressions so that the decoding can be performed more easily seem to play more important roles. | most of the efforts to bring out the off-the-shelf products into the market have failed mainly due to the difficulty in Korean syntactic parsing and the failure in the domain adaptation. | neutral |
train_99176 | The short message is characterized by informal, highly interactive and other features of oral Chinese. | analysis of short message corpus is an important foundation for research of automatic short message processing technology. | neutral |
train_99177 | And also some statistical analysis are made on the manually processed results. | it has many differences with the normal oral Chinese. | neutral |
train_99178 | We show that this approach is robust when applied to other texts. | we propose a measure of top-bag-of-word similarity for comparing the language variants contained in Chinese GigaWord corpus. | neutral |
train_99179 | If the score computed by expression (1) is a positive number, the syntactic piece occurs frequently in trouble reports. | in this section, we construct the trouble dictionary. | neutral |
train_99180 | Troubles has a higher score because it occurs more often in trouble reports. | we think that if we can connect Basis (2) and Basis (1) pieces, Basis (2) pieces can be called troubles. | neutral |
train_99181 | In this paper, we provide alternative methods to cope with those problems within our user interaction environment. | the customization process includes construction of translation resources specialized in scientific papers, modification of the engine to reflect the linguistic characteristics of academic papers. | neutral |
train_99182 | That is, BE refers to the objective property while BE/FEEL requires experiencer as its argument and refers to one's psychological state. | the Fourth characteristic of the conceptual structures in (13) is that (13a) and (13c) differ from each other with respect to the functions that take ([X] Last, the subscripts on the function FEELING helps to capture the further difference between property/evaluation adjectives and psychological state adjectives. | neutral |
train_99183 | The state of research on sign language linguistics shall also present challenges. | by bringing the Deaf into active partnership with academe, scientists, engineers, there shall be further validation of this language community. | neutral |
train_99184 | Sign language recognition systems need to examine five elements: gross arm movement, hand location, hand shape, hand orientation, and non-manual signals (NMS) which include facial expressions and body posture. | it is an important thrust based on the fact that "creative-sector occupations in science and technology… have grown since 1980 from 12 percent of the work force to between 30 and 40 percent in most advanced countries today…" and at the same time, "…disturbing that the poorest countries and regions around the world continue to export more than half of their scientific and engineering talent to advanced economies" (Florida, 2006). | neutral |
train_99185 | Then, we used the K-means clustering method with 16 clusters because the size of the tag set that was used to annotate the test set is 16. | several points remain for our future work as follows: A further analysis of the classification results would be useful. | neutral |
train_99186 | The works presented here explored the use of learning algorithms to automatically extract templates from training examples provided by the user. | in P4, the "house-wall" relationship is missing because "wal" is not found in Unisyn to produce the word "wall". | neutral |
train_99187 | In case that there is relative clause, we subtract the number of relatives from the number of main verbs. | table 3: Main causes of the errors. | neutral |
train_99188 | They recognize coordinate structures in a Japanese sentence by constructing a similarity matrix between bunsetsus and searching a path with the highest parallelism in the similarity matrix using a dynamic programming method. | the final parsing result is produced by parsing the whole sentence, with the recognized coordinate structure substitued by that node. | neutral |
train_99189 | It should be noted that the testing was just a single-pass because the lexicon is not yet fully functional. | then the result is subjected to the EVAL function, where the output of the GEN function is checked against the two lexicons (root and affixes) and subjected to some constraints to be able to get the right root of the verb. | neutral |
train_99190 | If you consult a dictionary, you will find that its English translation is "to die," which is the intended meaning. | our proposed system uses the morphological analyzer developed in (Roxas, 1989), which is being extended to cover more possible verb conjugations. | neutral |
train_99191 | In HowNet the range of attribute types of relations is constrained by their attribute-values|屬性 值. | we claim that all word interpretations involve two types of sense: relation sense and content sense. | neutral |
train_99192 | the K means, Expectation-Maximization, spectral clustering, Information Bottleneck, Probabilistic Latent Semantic Analysis, cost-based pairwise clustering (Brew and Schulte im Walde, 2002;Schulte im Walde, 2006;Korhonen et al., 2008;Sun and Korhonen, 2009;Vlachos et al., 2009). | after providing a short introduction to the basic principles of manual verb classification, this paper reviews recent research in automatic classification -particularly focussing on work conducted in English -and discusses then the various current challenges that need to be met for substantial further advances. | neutral |
train_99193 | When we check the top-5 and top-10 sentences, the acceptable rate rises to 80%. | as shown in Table 6, both of the scores are above 60, which means that the overall quality of the generated first sentences is acceptable. | neutral |
train_99194 | By exploiting characteristics in Wikipedia, our approach is able to achieve high accuracy. | 11,600,000 page counts) Celine Dion is not listed in WordNet, while Johnny Cash is listed as an instance of singer#n#1. | neutral |
train_99195 | In Table 3, we can see that the distribution varies depending on attribute types. | for NER, the naïve approach is rule-based, but its big disadvantage is the difficulty of ruledesign (feldman, 2002) or rule-learning (Etzioni et al., 2005), which needs to handle various named-entity expressions. | neutral |
train_99196 | This indicates this kind of information extraction needs more effort. | because those NER technologies mainly focused on news documents, it was so dependent on text information, such as capitalization information and corpus type, that their performance dropped much when working on web data because those cues are rather noisy. | neutral |
train_99197 | These features are also probably more discriminative in French than in English, since it is morphologically richer. | 6 The feature templates we used for designing our French tagging model is a superset of the features used by (Ratnaparkhi, 1996) and (Toutanova and Manning, 2000) for English (these were largely language independent). | neutral |
train_99198 | 12 Our baseline tagger significantly outperforms the retrained version of TreeTagger, with an improvement of over 10% on unknown words. | we performed a series of experiments by training MElt f fr on various subcorpora and sublexicons. | neutral |
train_99199 | It predicts the POS tags based on feature information of words collected in a predefined window size while ME or CRF predicts them based on the information of the whole sentence. | a confusion matrix for an n-way classification task is an n-by-n matrix, C, where the cell C (x: y) contains the number of times (in percentage) an item with correct classification x was classified by the model as y with respect to the total number of errors during the classification task. | neutral |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.