id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_97600 | (ii) For the first time, we apply maximum expected BLEU training on a data set as large as four million sentence pairs. | due to the renormalization component, this results in a total of 6.08M features that are updated with GT using the same data. | neutral |
train_97601 | In higher iterations, the update steps for good features keep growing and we observe an exponential increase of the objective function. | the triplet features did not finish in time, so we applied the feature sets (a), (b) and (d), 45M features in total. | neutral |
train_97602 | Some rare words do not receive a vector representation after running Word2Vec, and we simply remove phrases containing those words, resulting in a total of 0.65m phrases for Arabic, 0.18m phrases for Urdu, and 1.2m phrases for English. | the algorithms presented in the previous section require rapid retrieval of neighboring phrases in continuous space. | neutral |
train_97603 | 2014, we use a baseline MT system to translate the Arabic or Urdu phrases and add their translations to the English phrase set. | this extra computation takes only a negligible amount of time, since the number of labeled phrases on the source side is significantly smaller than the number of phrases on the target side. | neutral |
train_97604 | ), but are more general, with semantic categories relevant to common nouns and verbs as well. | here we build on prior work with an inventory of semantic classes (for nouns and verbs) known as supersenses. | neutral |
train_97605 | In addition to an overwhelming task of trying to capture all words and expressions that can convey a sentiment there are many other problems to solve: resolving the scope of negation to determine the shift of polarity (Lapponi et al., 2012), determining if an opinion is present in interrogative or conditional sentences (Narayanan et al., 2009), dealing with irony (Tsur, 2010), etc. | the recall for the positive class is substantially low than for the negative class. | neutral |
train_97606 | Our next technique (Frequency Method) modifies the phrase table by assigning short bit codes to frequent words, and long bit codes to infrequent words. | "Recalls" indicates how many participants returned to type their memorized English sequences, and "Correct Recalls" tells how many sequences were accurately remembered. | neutral |
train_97607 | A random, computergenerated 60-bit string is much more secure. | instead of a 2048-word dictionary, we use a 32,7868-word dictionary. | neutral |
train_97608 | Feature types are shared across categories, e.g., categories CLOTHING (k1), BIRDS (k2), and FOOD (k3) are all associated with feature type color (g2). | we compared the performance of BCF against BayesCat, a Bayesian model of category acquisition (Frermann and Lapata, 2014) and Strudel, a pattern-based model which extracts concept features from text (Baroni et al., 2010). | neutral |
train_97609 | Both n-gram factors will therefore be included as fixed effects and as by-subject random slopes in the baselines of the remaining evaluations in this study. | it is common for psycholinguistic models to include a measure of n-gram predictability for each fixated word conditioned on its context, but unless probabilities for words between fixations are also included, the probabilities used in this calculation are Table 1, the standard bigram factor (top line) predicts that the reading time of the region that ends with word 6 depends on word 5, but the probability of word 5 given its context is never included in the model, so an improbable transition between words 4 and 5 would not be caught. | neutral |
train_97610 | Qualitative evaluation shows that the model makes reasonable predictions of the level of formality of social network ties in well-known movies. | to compute the predictive log-likelihood of the address terms, we hold out a randomly-selected 10% of films. | neutral |
train_97611 | In contrast, a status-based network theory would penalize non-transitive triads such as β >>< . | this is a natural next step from prior work that computes the frequency of triads in explicitly-labeled signed social networks (Leskovec et al., 2010b). | neutral |
train_97612 | As a result, the morphological analysis happens within a different model compared to the model in which the resulting morphemes are consequently used. | for German, we use the Gur350 and ZG222 datasets (Zesch and Gurevych, 2006). | neutral |
train_97613 | The proposed method greatly improves the flexibility of translation rules at the cost of only a 30% increase in decoding time, and we demonstrate a 1.2-1.9 BLEU improvement over a strong tree-to-tree baseline. | in this paper we have proposed flexible nonterminals for dependency tree-to-tree translation. | neutral |
train_97614 | In this section, we propose a second phrase selection method based on the results from the syntactic analysis of source language data. | bold face indicates the highest coverage for each number of additional words. | neutral |
train_97615 | This method allows for improvement of coverage with fewer additional words than sentence selection, achieving higher efficiency by reducing the amount of data unnecessarily annotated. | the accuracy is conversely degraded if we use only phrase pairs with confidence level 3. | neutral |
train_97616 | Our empirical research questions are as follows: • can we control the production of honorifics in neural machine translation via side constraints? | our approach is not specific to this architecture. | neutral |
train_97617 | The actual benefit from this reduction is highly implementation-and architecture-dependent. | recall that c i = e i−1 i−n+1 , f a i +m a i −m . | neutral |
train_97618 | reviews for a movie, or arguments for a controversial social issue), and then outputs a one-sentence abstractive summary that describes the opinion consensus of the input. | we further constructR to contain all the pairwise differences (r p − r q ).L is a vector of the same size asR with each element as 1. | neutral |
train_97619 | Since we have three systems we performed this pairwise study thrice. | while clusters that balance the entities are preferable, it is also acceptable to have clusters where one of the entities is sparsely represented (or not represented at all). | neutral |
train_97620 | Albeit the encouraging performance of our proposed approach on summarizing student responses, when applied to the DUC 2004 dataset (Hong et al., 2014) and evaluated using ROUGE we observe only comparable or marginal improvement over the ILP baseline. | notable systems include maximal marginal relevance (Carbonell and Goldstein, 1998), submodular functions (Lin and Bilmes, 2010), jointly extract and compress sentences (Zajic et al., 2007), optimize content selection and surface realization (Woodsend and Lapata, 2012), minimize reconstruction error (He et al., 2012), and dual decomposition (Almeida and Martins, 2013). | neutral |
train_97621 | Thesaurus synonyms are not that helpful for generating inference rules (or else we will generate rules like produce → percolate ). | for example, the rule be the president of → be not the president of , will be rejected. | neutral |
train_97622 | But not all relations are distributive in this sense. | this work has availed itself of increasingly sophisticated features of the semantics of the units to be related (Braud and Denis, 2015); but as the PDtB does not provide full discourse structures for texts, it is not relevant to our concerns here. | neutral |
train_97623 | For our experiments we used a corpus collected from chats involving an online version of the game The Settlers of Catan described in Afantenos et al., 2015). | for example, we could complicate (2) slightly: ( In 3, the SDRS graph would be: this SDRS entails that a is explained by [c, d] and that b is explained by [c, d]. | neutral |
train_97624 | As decoding proceeds, the influence of the initial input on decoding (i.e., the source sentence representation) diminishes as additional previously-predicted words are encoded in the vector representations. | intuitively, it seems desirable to take into account not only the dependency of responses on messages, but also the inverse, the likelihood that a message will be provided to a given response. | neutral |
train_97625 | For instance, by using a basic unigrambased definition of discussion points, we do not account for the context or semantic sense in which these points occur. | often the debates are quite tight: for 30% of the debates, the difference between the winning and losing sides' deltas is less than 10%. | neutral |
train_97626 | This figure is above the average inter-annotator agreement of 0.67, which has been referred to as the ceiling performance in most work up to now. | contrary to the SimLex-999 experiments, starting from the Paragram vectors did not lead to superior performance, which shows that injecting the application-specific ontology is at least as important as the quality of the initial word vectors. | neutral |
train_97627 | cheaper and pricey) is critical for the performance of dialogue systems. | in our opinion, the average inter-annotator agreement is not the only meaningful measure of ceiling performance. | neutral |
train_97628 | A new feature of this version is that it assigns relation types to its word pairs. | a dialogue system can be led seriously astray by false synonyms. | neutral |
train_97629 | The entropy is shown at the bottom of the figure. | moreover, linguistic studies have shown that action verbs such as cut and slice often denote some change of state as a result of the action (Hovav and Levin, 2010;Hovav and Levin, 2008). | neutral |
train_97630 | Also note that the whole prepositional phrase "from the drawer" is identified as the source rather than "the drawer" alone. | the traditional SRL is not targeted to represent verb semantics that are grounded to the physical world so that artificial agents can truly understand the ongoing activities and (learn to) perform the specified actions. | neutral |
train_97631 | The focus of our experiments is on metaphorical expressions in verb-subject, verb-direct object and adjectival modifier-noun constructions. | the thesholds appear to be relatively stable, with a standard deviation of 0.03 for MIXLAtE; 0.02 for WORDCOS (linguistic); and 0.05 for PHRASECOS1 (visual). | neutral |
train_97632 | We then use this English caption to retrieve images using the En-Image CorrNet. | en-Image CorrNet: This is the CorrNet model trained using only Z 1 as defined earlier in this section. | neutral |
train_97633 | The MSCOCO dataset 2 contains images and their English captions. | (A plate with meat and green veggies mixed with sauce.) | neutral |
train_97634 | encodes the input v j into a hidden representation h and then g V j (.) | there is no parallel data available between the non-pivot views. | neutral |
train_97635 | Table 6: Images that were assigned an incorrect sense in the PRED setting. | no comparable resource is available for verbs (see Section 2.1). | neutral |
train_97636 | Using WordNet we construct a superset G containing all possible parent relations for the relations in S by replacing their arguments o 1 , o 2 by all their possible hypernyms. | we use seven spatial relations and allow natural language relations that represent a larger array of higher level semantics. | neutral |
train_97637 | Second, it is manipulated using both push and pop operations. | the generation algorithm also requires slightly modified constraints. | neutral |
train_97638 | Instead, we rely on the LSTM supertagger to implicitly model the dependencies-a task that becomes more challenging with longer dependencies. | (2013) with a shiftreduce parser and a chart-based model. | neutral |
train_97639 | These models are 0.3 and 1.5 F1 more accurate than the C&C baseline respectively, which is well within the margin of improvement obtained by our model. | parsing models either use these scores directly (Auli and Lopez, 2011b), or as a form of beam search (Clark and Curran, 2007), typically in conjunction with models of the dependencies or derivation. | neutral |
train_97640 | For each position i, we select the most probable supertag from the output distribution. | we achieve strong accuracies compared to (Wang et al., 2015) using feed-forward neural network model trained on local context, showing that this task does not require bi-LSTMs. | neutral |
train_97641 | During testing we simply apply the word segmenter to the sentences. | in addition, adapting the word segmentation with NER partiallylabeled data gives a further gain for both CTB6 and PKU, with an F-measure of 86.96% and 87.64% respectively. | neutral |
train_97642 | ), sentences are represented as strings of characters without similar natural delimiters. | from the adaptation the system learns to put a word boundary between '亲' and '四' and then the correct slot value '四季酒店' (four Seasons Hotel) is extracted. | neutral |
train_97643 | After that the performance seems to quickly saturate. | the joint-learning process generally assumes the availability of manual word segmentations for the training data, which limits the use of this approach. | neutral |
train_97644 | During training, we use L-BFGS to compute the maximumlikelihood estimates of φ. | • Event trigger: the word or phrase that clearly expresses its occurrence. | neutral |
train_97645 | In order to capture the inter-dependencies between triggers and argument roles, we introduce memory vectors/matrices to store the prediction information during the course of labeling the sentences. | after that, higher level features has been investigated to improve the performance (Ji and Grishman, 2008;Gupta and Ji, 2009;Patwardhan and Riloff, 2009;Liao and Grishman, 2010;Liao and Grishman, 2011;Hong et al., 2011;McClosky et al., 2011;Huang and Riloff, 2012;Li et al., 2013). | neutral |
train_97646 | (2015) by means of visualization experiments and cell activation statistics in the context of character-level language modeling. | it is desirable to have a more principled mechanism allowing us to inspect recurrent architectures from a linguistic perspective. | neutral |
train_97647 | This work was supported by a Google Faculty Research award to the third author. | these two vectors Figure 1: A fragment of our model with latent variable zt, which only illustrates discourse information flow from sentence (t − 1) to t. the information from sentence (t − 1) affects the distribution of zt and then the words prediction within sentence t. are then linearly combined in the generation function for word y t,n , where c t−1 is set to the last hidden state of the previous sentence. | neutral |
train_97648 | As such, it can be viewed as a form of multi-task learning (Caruana, 1997), where we learn a shared representation that works well for discourse relation prediction and for language modeling. | but because these graphical models represent uncertainty for every element in the model, adding too many layers of latent variables makes them difficult to train. | neutral |
train_97649 | As a result, the model can act as both a discourse relation classifier and a language model. | unlike this prior work, in our approach, the latent variables carry a linguistic interpretation, and are at least partially observed. | neutral |
train_97650 | Of particular relevance are applications of neural architectures to PDTB implicit discourse relation classification Zhang et al., 2015;Braud and Denis, 2015). | the discrete latent variables in our model are easy to sum and maximize over. | neutral |
train_97651 | As described in Section 2, the only previous attempt at automatically deriving the meaning of phonesthemes is due to Abramova et al. | our second similarity test is a check on the overall semantic cohesiveness of the candidate phonesthemic clusters. | neutral |
train_97652 | An overview of the results can be seen in Figure 1. | in the first part of our study (Section 4) we present a stricter validation method for candidate phonesthemes that also includes considerations related to morphological diversity, which were ignored in previous work. | neutral |
train_97653 | Hence, we would expect a negative correlation between cluster size and semantic similarity. | our phonestheme validation procedure is stricter compared to previous work since we use sets of words that share a random two-consonant prefix as baseline and, importantly, take into account morphological relatedness. | neutral |
train_97654 | We also include features indicating the relation of the typed dependency of the chunks. | since certain adjacent words tend to be discarded or kept together, we reinforce this property by adding a bigram POs feature of w h to encode its context. | neutral |
train_97655 | 496,567 tokens on the target side) and about 2,691 pairs of parallel sentences for testing (approx. | i suggest you visit first the cathedral of " Le UNK UNK " because it is the most UNK building in the area . | neutral |
train_97656 | As a first step towards developing full-fleged learning systems that leverage all signals available within a communicative setup, in our extended model we incorporate information regarding the objects that caregivers are holding. | it could be that MSG has simply learned to treat, say, the lamb visual vector as an arbitrary signature, functioning as a semantically opaque ID for the relevant object, without exploiting the visual resemblance between lamb and sheep. | neutral |
train_97657 | To use maximum likelihood or minimum cross-entropy, it assumes that the model distribution is peaked. | we notice that, in ondevice model, user queries tend to be short, with on average 2.4 words in a query, shown in Table. | neutral |
train_97658 | Different from RCRFs and conventional RNNs that in essence apply the multinomial logistic regression on the output layer, RSVMs optimize the sequence-level max-margin training criterion used by structured support vector machines (Tsochantaridis et al., 2005) on the output layer of RNNs. | model training can be sped up by skipping the weight updating for non-support vector training samples. | neutral |
train_97659 | 6 Both the development and test sets were annotated by a human judge and an author of this paper. | it is worth noting that the DDoS attack on RBS, Ulster Bank and Natwest was actually on 2015 July 31. | neutral |
train_97660 | All experiments were run with the default settings except for a distortion-limit of 12 in the JP-EN experiment, as suggested by (Goto et al., 2013). | to put these results in the broader context of machine translation research, our approach (even without special handling of unknown words) achieved gains of up to 5.6 BLEU points over strong phrase-based and hierarchical phrase-based Moses baselines, with the help of an ensemble technique. | neutral |
train_97661 | To overcome this issue, we propose an agreement model for neural machine translation and show its effectiveness on large-scale Japaneseto-English and Chinese-to-English translation tasks. | their approach was concerned with feedforward networks, which can not make full use of rich contextual information. | neutral |
train_97662 | There has been little work on deception detection in written language and most of it has focused on either discriminating between sincere and insincere arguments (Mihalcea and Strapparava, 2009) or opinion spam (Ott et al., 2011;Jindal and Liu, 2008). | while their results are promising, our focus is on written text only. | neutral |
train_97663 | The werewolves are motivated to hide their roles, as in every round there is a majority of non-werewolves. | hesitations (um, er, uh), hedges (sort of, kind of, almost), and polite forms are markers of powerless language (Sparks and Areni, 2008). | neutral |
train_97664 | For simplicity, denote the entity set as E = {e 1 , • • • , e |E| }, and the template set as T . | many slot filling algorithms requires the full information of the event schemas and the labeled corpus. | neutral |
train_97665 | By incorporating templates and slots as latent topics, probabilistic graphical models learns those templates and slots that best explains the text. | the edge weight between two points is their template level similarity / slot level similarity. | neutral |
train_97666 | The most extensively developed resource for English is the MRC Psycholinguistic Database (Section 2). | the most extensively developed resource for English is the MRC Psycholinguistic Database (Section 2). | neutral |
train_97667 | • Simple Ridge: Test word is assigned the property value as predicted by a Ridge regressor trained with the 15 aforementioned lexical features. | it is far from complete, most likely due to the inherent cost of manually entering such properties. | neutral |
train_97668 | At the same time, we want to give special thanks to the anonymous reviewers for their insightful comments as well as suggestions. | all those works for document representation paid little attention to the variability of intra-topic documents. | neutral |
train_97669 | For example, this approach was used by and was originally used for PLTM. | figure 7: oPLTM vs. mlhPLTM: perplexity comparison (left); performance comparison on the CLIR task (right). | neutral |
train_97670 | One alternative approach to longer documents that has received attention in the past has been to directly model local-i.e., Markov-dependencies among tokens. | the VB approach (Blei et al., 2003) offers more efficient computation but as in the case of Gibbs sampling requires iterating over the whole collection multiple times (e.g. | neutral |
train_97671 | If a h ≥ 1.5 and a t ≥ 1.5, then r is labeled as M-M. 1.4%, 8.9%, 14.6% and 75.1% of the test triples belong to a relation type classified as 1-1, 1-M, M-1 and M-M, respectively. | our new KB completion model STransE chooses W r,1 , W r,2 and r so that W r,1 h + r ≈ W r,2 t. That is, a TransE-style relationship holds in some relation-dependent subspace, and crucially, this subspace may involve very different projections of the head h and tail t. So W r,1 and W r,2 can highlight, suppress, or even change the sign of, relation-specific attributes of h and t. For example, for the "purchases" relationship, certain attributes of individuals h (e.g., age, gender, marital status) are presumably strongly correlated with very different attributes of objects t (e.g., sports car, washing machine and the like). | neutral |
train_97672 | Neural Network models like Convolutional Neural Networks and Recurrent Neural Networks (LSTM, GRU) have recently been been successfully used to tackle various sequence labeling problems in NLP. | the embeddings are trained on a large unlabeled biomedical dataset, compiled from three sources, the English Wikipedia, an unlabeled EHR corpus, and PubMed Open Access articles. | neutral |
train_97673 | In this sentence, the true labels are Adverse Drug Event(ADE) for "bronchiolitis obliterans" and Drugname for "ABVD chemo". | (1994) showed that learning long term dependencies in recurrent neural networks through gradient decent is difficult. | neutral |
train_97674 | More interestingly, we find that performance can be improved if the system scores and human ratings are aggregated over several topic cardinalities before computing the correlation. | 2014proposed an automated approach to the word intrusion task. | neutral |
train_97675 | Transition actions are treated as an atomic output component in each feature instance. | for scoring the action SHIFT-Lw-Lp, S 0 w is instantiated into S 0 w-SHIFT-Lw-Lp, where Lw is the word to shift and Lp is its POS. | neutral |
train_97676 | To make quality vectors, we regard that the probability of the target word y j involves the quality information about whether the target word y j in target sentence is properly translated from source sentence. | feature selection is to select the best features by using selection algorithms, such as Gaussian processes (Shah et al., 2015) and heuristic (González-Rubio et al., 2013), among already extracted features. | neutral |
train_97677 | 7 From the extended prediction method of (4), the probability of the target word y j is computed by using information of relevant source words in source sentence x and all target words y y j surrounding the target word y j in target sentence. | by decoding t j , we are able to get quality vector q y j for the target word y j ∈ R Ky at position j of target sentence. | neutral |
train_97678 | Word Embeddings for Verbs and Adjectives. | 4 For all context types other than BOW we use the word2vec package of (Levy and Goldberg, 2014), 5 which augments the standard word2vec toolkit with code that allows arbitrary context definition. | neutral |
train_97679 | Moreover, the SP-based model is much more compact than the alternatives, making its training an order of magnitude faster. | sG-sP and sG-Coor, which take 11 minutes and 23 minutes respectively to train, are substantially faster than the other w2v-sG models. | neutral |
train_97680 | In order to tackle this problem, Scheirer et al. | we investigated the problem via reducing the open space risk, and proposed a solution based on cen- ter-based similarity space learning. | neutral |
train_97681 | We also tried gated recurrent units (GRUs) (Cho et al., 2014) and the basic RNN, but the results were generally lower than LSTM. | naive Bayes: (Lendvai and Geertzen, 2007). | neutral |
train_97682 | We used the NN shown in Figure 1. | we treated 252 neurons in the final hidden layer as dedicated neurons in weight initialization. | neutral |
train_97683 | According to the structure shown in Figure 1, W 1 is a matrix of weights that is updated during training, thus the distributed representations contained in W 1 are learned simultaneously with the training of BLSTM-RNN on any supervised learning tasks. | recently, many state-of-the-art systems of tagging related tasks are implemented with bidirectional long short-term memory (BLSTM) recurrent neural network (rNN), for example, slot filling (Mesnil et al., 2013), part-of-speech tagging (Huang et al., 2015), and dependency parsing (Dyer et al., 2015) etc. | neutral |
train_97684 | The matrix factorization family only uses the statistics of co-occurrence counts, disregarding of the position of word in sentence and word order. | it should be noted that the speed of our approach is acceptable compared with previous neural network language model based methods, including (Bengio et al., 2003;Mikolov et al., 2010;Mnih and Hinton, 2007), as our model uses a much simpler output layer which only has two nodes, avoiding the time consuming computation of the big softmax output layer in language model. | neutral |
train_97685 | Therefore in bi-directional RNNs, not only a history vector of word w t is regarded but also a future vector. | with our neural models, we achieved new state-of-the-art results on the SemEval 2010 task 8 benchmark data. | neutral |
train_97686 | We re-implemented the approach by Pichotta and Mooney (2014) with the exception that we use v(e subj , e dobj , e iobj ) instead of v(e subj , e obj , e prep ) to represent events. | for future work, this lack of coverage could be compensated for by backing off to the P&M model. | neutral |
train_97687 | We can then write unary Thus we can regard the bilinear form as a function computing a weighted inner product over some real embedding v y U representing state y and some real embedding representing input factor t. The rank of W gives us the intrinsic dimensionality of the embedding. | within this context, we are interested in the task of predicting semantic tuples for images. | neutral |
train_97688 | Interestingly, in 36% of games, the team arrives at a better answer than any of the individual guesses (c best > 0). | for example, the conversation in figure 4 has two confident phrases, underlined in red. | neutral |
train_97689 | One possible clue as to why vision is better at predicting a concept's properties is given by the fact that it obtains better results on concepts such as PANTS or STOOL, where the only difference to very similar concepts like TROUSERS or CHAIR are visual (a STOOL has no backrest as opposed to a CHAIR). | table 2 shows some examples for each of the 10 property types as defined and annotated in MCRAE. | neutral |
train_97690 | The published dataset contains a total of 2526 features, with a mean of 13.7 features per concept. | close-to-perfect performance in this task is impossible, since almost 30% of the features only occur with one concept, and hence can't be reconstructed for that particular concept. | neutral |
train_97691 | Cross-lingual Wikification is the task of grounding mentions written in non-English documents to entries in the English Wikipedia. | finally, Section 7 concludes the paper. | neutral |
train_97692 | We choose the CCA-based model because we can obtain multilingual word and title embeddings for all languages in Wikipedia without any additional data beyond Wikipedia. | they only focus on English Wikification. | neutral |
train_97693 | Previous work in the area have proposed a number of methods for identifying and extracting task knowledge from search query sessions (Mehrotra and Yilmaz, 2015b;Wang et al., 2013;Lucchese et al., 2011;Verma and Yilmaz, 2014;Mehrotra and Yilmaz, 2015a). | in comparison, however, the baseline methods failed to identify diagnostic clusters. | neutral |
train_97694 | The Z r , Z t (f ) and Z n (f ) terms denote the partition functions of each model, and φ r , φ n and φ t are functions mapping nonterminals and production rules to feature vectors. | we iteratively developed these templates to cover the data set, and the lexicon generated by these templates can correctly parse 96% of the examples in FOODCHAINS. | neutral |
train_97695 | The suffix-prefix-based approach results to Bundes-finanz-ministerium and the prefixsuffix method to Bund-esfinanz-ministerium. | our method does not require any linguistic knowledge and is initialized using a large monolingual corpus. | neutral |
train_97696 | We have examined schemes of priority ordering for integrating information from different candidate sets, e.g. | most existing systems rely on dictionaries or are trained in a supervised fashion. | neutral |
train_97697 | The latter automaton uses integers as state names: it is 0 xn −→ n . | it removes γ i:j , e x i and e x j+1 and in fact, it is precisely a weighting of the edits in our original FST F , without further considering the context in which an edit is applied. | neutral |
train_97698 | In an error analysis, it turned out that FF2010 (i.e., SMOR) cannot process 2% of the gold samples. | section 4 shows two efficient and flexible data structures used for our statistical compound splitter, which is described in section 5. | neutral |
train_97699 | To conduct our experiments, we follow the single mode setting of Hermann et al. | to visualize the difference in embeddings learned with BRAVE-S and BRAVE-D, we selected sentiment words and identified crosslanguage nearest neighbors in table 5. | 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.