id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_3600 | WMD helps to bridge the vocabulary gap between the citing context and the cited article. | though not often, the use of distributed representation can also create mistakes. | contrasting |
train_3601 | In another example shown in Table 8, ungrounded mention SNP, though detected, contributes little to supporting the ground truth pair. | when it's grounded to the entry Scottish National Party, the system leverages world knowledge and relates it to the mention of Scotland in the citing context. | contrasting |
train_3602 | More recently, attempts have been made to automatically extract pronunciation dictionaries directly from audio data (Stahlberg et al., 2016). | the requirement of a native speaker, web resources, or audio data specific to the language still blocks development, and the number of g2p resources remains very low. | contrasting |
train_3603 | This expands 56 g2p models (the languages covered by both IPA Help and Wiktionary models) to models for 211 languages. | as shown in Table 8, results are very poor, with a very high WER of 92% using the unioned models and a PER of more than 50%. | contrasting |
train_3604 | Since Xue (2003), most methods formalize the Chinese word segmentation (CWS) as a sequence labeling problem with character position tags, which can be handled with su-pervised learning methods such as Maximum Entropy (Berger et al., 1996;Low et al., 2005) and Conditional Random Fields (Lafferty et al., 2001;Peng et al., 2004;Zhao et al., 2006a). | those methods heavily depend on the choice of handcrafted features. | contrasting |
train_3605 | In order to get exact inference, most sequence-labeling systems address this problem with a Viterbi search, which takes the advantage of their hypothesis that the tag interactions only exist within adjacent characters (Markov assumption). | since our model is intended to capture complete history of segmentation decisions, such dynamic programming algorithms can not be adopted in this situation. | contrasting |
train_3606 | The results of the discrete model and the neural model without fine-tuning are highly similar, showing the usefulness of beam-search. | with fine-tuning, the results are different. | contrasting |
train_3607 | The model with beam size 1 gives better accuracies compared to the other models with the same beam size. | as the beam size increases, the performance increases very little. | contrasting |
train_3608 | The perspectives of our model can be considered a type of feature. | they are implemented by parametric differentiable functions. | contrasting |
train_3609 | We transform the question and the answer toq m andã n analogously using distinct matrices and bias vectors. | to the semantic perspective, we keep the question and answer candidates separate in the wordby-word perspective. | contrasting |
train_3610 | Moreover, linearization of the dependency graph, because it relies on an eigendecomposition, is not differentiable. | we handle the linearization in data preprocessing so that the model sees only reordered word-vector inputs. | contrasting |
train_3611 | This suggests that, to learn true reasoning abilities, MCTest is too simple a dataset-and it is almost certainly too small for this goal. | it may be that human language processing can be factored into separate processes of comprehension and reasoning. | contrasting |
train_3612 | These inferences are feasible in a limited domain, but become difficult the more open-domain reasoning they require. | even a simple lexical overlap classifier could correctly predict the entailment. | contrasting |
train_3613 | The domain of word denotations is then or- dered by the subset operator, corresponding to ordering by hypernymy over the words. | 2 hypernymy is not the only useful partial ordering over denotations. | contrasting |
train_3614 | (2015) for their conversational dialog engine Knowbot, and therefore constitutes a more fair comparison against their results. | we also make use of the full SCITEXT corpus (Clark et al., 2014). | contrasting |
train_3615 | The main challenge in learning a curriculum is that it requires the identification of easy and hard concepts in the given training dataset. | in real-world applications, such a ranking of training samples is difficult to obtain. | contrasting |
train_3616 | These datasets can also help us understand the shaping hypothesis as we can use them to verify if easier questions are indeed getting picked by our incremental learning algorithm before harder questions. | deep learning models (Le-Cun et al., 2015) have recently shown good performance in many standard NLP and vision tasks, including QA. | contrasting |
train_3617 | SPL is different from active learning (Settles, 1995) in the sense that the training examples (and labels) are already provided and the solution only orders the examples to achieve a better solution. | active learning tries to interactively query the user (or another information source) to achieve a better model with few queries. | contrasting |
train_3618 | This process of memory pattern restructuring is difficult to achieve, so it could be the reason for the relatively poor per-formance of naive curriculum learning and SPL strategies. | as we can see from the previous results, the explore and exploit strategy of mixing in some harder examples avoids the problem of having to abruptly restructure memory patterns. | contrasting |
train_3619 | We also tried variants of our models where we used curriculum learning on grade 3 questions, followed by grade 4 and grade 5 questions. | this did not lead to significant improvements. | contrasting |
train_3620 | For instance, the question in Table 1 only contains five words, while the best answer uses 60 words for elaboration. | the best answers from a pool can also be noisy and include extraneous information irrelevant to the question. | contrasting |
train_3621 | In (Feng et al., 2015), GESD outperforms cosine similarity in their models. | the proposed models with GESD as similarity scores do not provide any improvement on the accuracy. | contrasting |
train_3622 | Note that the best result in (Severyn and Moschitti, 2015) was obtained by combining CNN-based features with additional human-defined features. | our attentive LSTM model achieves higher performance without using any human-defined features. | contrasting |
train_3623 | Pasupat and Liang (2015) use tables from Wikipedia to construct a set of QA pairs. | their annotation setup does not impose structural constraints from tables, and does not collect finegrained alignment to table elements. | contrasting |
train_3624 | In both cases only the training partition of each dataset was used to formulate and handcraft tables. | for unbounded topics, additional facts were added to each table, using science education text books and websites. | contrasting |
train_3625 | (2015) propose a neural attention model for abstractive sentence compression which is trained on pairs of headlines and first sentences in an article. | our model summarizes documents rather than individual sentences, producing multi-sentential discourse. | contrasting |
train_3626 | In the standard neural sequence-to-sequence modeling paradigm (Bahdanau et al., 2015), an attention mechanism is used as an intermediate step to decide which input region to focus on in order to generate the next output. | our sentence extractor applies attention to directly extract salient sentences after reading them. | contrasting |
train_3627 | This suggests that an alternative to training the network would be to employ a ranking-based objective or a learning to rank algorithm. | we leave this to future work. | contrasting |
train_3628 | The guidelines state in fact that 'If the negated affix is attached to an adverb that is a complement of a verb, the negation scopes over the entire clause' (Morante et al., 2011, p. 21) and we annotated suffixal negation in this way. | 3 out of 4 examples of suffixal negation in adverbs in the training data (e.g. | contrasting |
train_3629 | Perhaps the most common fixedlength vector representation for texts is the bag-ofwords or bag-of-n-grams (Harris, 1970). | they suffer severely from data sparsity and high dimensionality, and have very little sense about the semantics of words or the distances between the words. | contrasting |
train_3630 | Wherein, the contexts are fixedlength (length is 2k) and sampled from a sliding window over the current sentence. | the word matrix W is shared across sentences. | contrasting |
train_3631 | (2014) built a conversation system using learning to rank and semantic matching techniques. | collecting enough Q-R pairs to build chatbots is often intractable for many domains. | contrasting |
train_3632 | The Interactive Alignment Model proposes that conversational partners prime each other, causing alignment via the primed reuse of structures ranging from individual lexical items to syntactic abstractions (Pickering and Garrod, 2004). | accommodation Theory emphasizes the relatively more communicative and strategic nature of alignment (Giles et al., 1991). | contrasting |
train_3633 | If alignment is driven primarily by priming, it should be relatively consistent across different aspects of a discourse. | from a strategic or communicative perspective, alignment -in which preceding words and concepts are reused -must be balanced against a need to move the conversation forward by introducing new words and concepts. | contrasting |
train_3634 | Distributional methods such as Linguistic Style Matching (LSM) (Niederhoffer and Pennebaker, 2002;Ireland et al., 2011) or the Zelig Quotient (Jones et al., 2014) calculate the similarity between the conversation participants over their frequencies of word or word category use in all utterances within the conversation. | conditional metrics, such as Local Linguistic Alignment (LLA) (Fusaroli et al., 2012;Wang et al., 2014) and the metric used by Danescu-Nicolescu-Mizil et al. | contrasting |
train_3635 | Intuitively, a within-turn topic boundary suggests that the speaker of the current turn is initiating the topic shift. | a betweenturn boundary suggests that the following speaker who first gives substantial contribution is more likely to be the initiator of the next topic. | contrasting |
train_3636 | Several such models have been proposed (Reiter and Dale, 2000;Henschel et al., 2000;Callaway and Lester, 2002;Krahmer and Theune, 2002;Gupta and Bandopadhyay, 2009;Greenbacker and McCoy, 2009). | all of these are fully deterministic, always choosing the same referential form in the same context. | contrasting |
train_3637 | Naive Bayes model vs. RNNs Although the RNNs were able to model individual variation in the choice of referential form to some extent, they did not perform significantly better than the Naive Bayes models, which might have to do with the relatively small dataset. | we think the size of the corpus matches the relatively low complexity of the problem we address. | contrasting |
train_3638 | Intuitively, the average American reader may not know exactly how many people are in Texas, but is familiar enough with the quantity to effectively reason using Texas' population as a unit. | it is less likely that the same reader is familiar with even the concept of Angola's population. | contrasting |
train_3639 | The model is able to capture rephrasings of fact descriptions and reordering of the facts. | it often confuses prepositions and, very rarely, can produce nonsensical utterances. | contrasting |
train_3640 | Qualitatively the neural networks outperform the baseline model in cases where they are able to levage additional knowledge about the entities (see first, third and fifth example in Table 4). | for rare relationships the baseline model appears to perform better, because it is able to produce a reasonable question if only a single example with the same relationship exists in the training set (see eighth example in Table 4). | contrasting |
train_3641 | A robust model might, for example, employ one pre- dictor to copy "The Foundation" from the input, and a another one to find the answer "Issac Asimov" by searching through a database. | training multiple predictors is in itself a challenging task, as no annotation exists regarding the predictor used to generate each output token. | contrasting |
train_3642 | This is informative as it gives an intuition of how many samples can be used without further human post-editing. | it does not provide an illustration on the degree of closeness to achieving the correct code. | contrasting |
train_3643 | Some tasks, such as the generation of queries (Zelle and Mooney, 1996), have overcome this problem by executing the query and checking if the result is the same as the annotation. | we shall leave the study of these methologies for future work, as adapting these methods for our tasks is not triv-ial. | contrasting |
train_3644 | Yet, it is a promising result that the model was able to capture this difference. | in many cases, effects that radically differ from seen ones tend to be generated incorrectly. | contrasting |
train_3645 | We adopt the WAC ("words-as-classifiers") model (Kennington and Schlangen, 2015), which was originally used for reference resolution in situated dialogue. | wAC is essentially a task-independent approach to predicting semantic appropriateness of words in visual contexts and can be flexibly combined with task-dependent decoding procedures. | contrasting |
train_3646 | We observe a clear learning effect in the non-interactive system, meaning that users faced unexpected interpretation problems due to inaccurate expressions, but adapted to the situation to some extent. | both installment systems have stable performance over time, which indicates that system behaviour is immediately understandable and predictable for human users. | contrasting |
train_3647 | every annotated mention could be mapped to at least one entity, and the set of entities included the gold entity. | changes in canonical Wikipedia URLs, accented characters and unicode usually result in mention losses over time, as not all URLs can be mapped to the KB (Hasibi et al., 2016, Sec. | contrasting |
train_3648 | It is of course possible to extend the global single-link model to the multi-focus case, by modifying the model factors and resulting messages. | the simplicity of the star-shaped model, its empirical effectiveness, and ease of learning parameters make it an attractive approach for easily incorporating attention into existing resolution models. | contrasting |
train_3649 | The common practice for ranking coreference resolvers is to use the average of three different metrics. | one cannot expect to obtain a reliable score by averaging three unreliable metrics. | contrasting |
train_3650 | Therefore, m is a coreferent mention that has at least one coreference link in the response entities. | none of its detected coreference links in the response entities are correct. | contrasting |
train_3651 | As can be seen, (1) recall changes for all the metrics except for MUC; (2) both CEAF e recall and precision significantly decrease; and (3) BLANC recall notably decreases so that F 1 drops significantly in comparison to Base. | adding completely incorrect entities to the response entities should not affect the recall and it should decrease the precision. | contrasting |
train_3652 | The following observations can be drawn from these experiments: (1) MUC and LEA are the only measures which give a zero score to the response that contains no correct coreference relations; (2) in our experiments, CEAF e shows an unreasonable drop when the correct link ratio changes from 0% to 20%; and (3), in Figure 2, the BLANC F 1 values are less than or equal to those of B 3 and LEA. | in Figure 3 that contains both coreferent and non-coreferent links, BLANC F 1 is at least 20% higher than that of other metrics. | contrasting |
train_3653 | The following points are worth noting from the results of Figure 7: (1) MUC and CEAF m are the least discriminative metrics when the system output includes extra mentions; (2) except for CEAF e , other metrics rank 3-10 as the worst output;(3) CEAF e recognizes both 2-0 and 3-0 as the worst outputs. | in these outputs the extra mentions are linked together and therefore no incorrect information is added to the correctly resolved entities; and (4) LEA is the only metric that recognizes error 2-0 is less harmful than 1-2 or 1-10. | contrasting |
train_3654 | However, in these outputs the extra mentions are linked together and therefore no incorrect information is added to the correctly resolved entities; and (4) LEA is the only metric that recognizes error 2-0 is less harmful than 1-2 or 1-10. | lEA does not discriminate the different outputs in which only one extra mention is added to an entity. | contrasting |
train_3655 | We believe entity-level information is particularly useful for preventing bad merges between large clusters (see Figure 4 for an example). | it is worth noting that in practice the much more complicated cluster-ranking model brings only fairly modest gains in performance. | contrasting |
train_3656 | 2016extend their mention-ranking model by incorporating entity-level information produced by a recurrent neural network running over the candidate antecedent-cluster. | this is an augmentation to a mention-ranking model, and not fundamentally a clustering model as our cluster ranker is. | contrasting |
train_3657 | For instance, when multiple puzzle-composers write crossword puzzle clues for the same word, they will try to write creative, unique clues to make the puzzle interesting and challenging; clues for "star" could be "Paparazzi's target" or "Sky light." | people writing a descriptive caption for a photograph can adopt a less creative style. | contrasting |
train_3658 | And indeed, this paper will show that, depending on the choice of clustering method, semantics-based similarity measures such as summed word embeddings and deep neural networks can have an advantage over more traditional similarity metrics, such as n-gram counts, n-gram tf-idf vectors, and dependency tree kernels, when applied to creative texts. | unlike in most text similarity tasks, in clustering the choice of similarity metric interacts with both the choice of clustering method and the properties of the text. | contrasting |
train_3659 | TTR is an obvious estimate of the width of the vocabulary of a corpus. | all other things being equal, a corpus of many very short texts triggered by the same stimulus would have more repeated words, proportional to the total number of tokens in the corpus, than would a corpus of a smaller number of longer texts. | contrasting |
train_3660 | Traditionally documents are represented as a bag-of-words (BOW) vectors. | this simple representation suffers from being high-dimensional and highly sparse, and loses semantic relatedness across the vector dimensions. | contrasting |
train_3661 | Taking the logarithm of both sides, we obtain where m ik = L i j=1 δ(z ij = k) counts the number of words assigned with the k-th topic in 6 Variational Inference Algorithm Given the hyperparameters α, γ, µ, the learning objective is to find the embeddings V , the topics T , and the word-topic and document-topic distributions Here the hyperparameters α, γ, µ are kept constant, and we make them implicit in the distribution notations. | the coupling between A, V and T , Z, φ makes it inefficient to optimize them simultaneously. | contrasting |
train_3662 | The word-to-word bilingual constraints in PCLSA are not as effective. | our MTCA model incorporates the bilingual translations using auxiliary distributions which incorporate word distributions from the other language on the topic level and can capture common semantics of multilingual reader comments. | contrasting |
train_3663 | If the dashed link does not exist, both WSBM and SCC can identify two blocks. | if the dashed link does exist, SCC will return only one big block that contains all nodes, while WSBM still keeps the nodes in two reasonable blocks. | contrasting |
train_3664 | Thus B consists of only explicit positive links. | to avoid overfitting, we sample some implicit links and add them to B as explicit negative links. | contrasting |
train_3665 | The only difference between LCH-RTM and LBH-RTM is the block detection algorithm. | their link prediction performance is poles apart-LCH-RTM even fails to outperform RTM. | contrasting |
train_3666 | For the AraSenTi-Trans lexicon, if the tweet contains a negation particle and a positive word, we do not increment the positive word counter. | for tweets containing negative words and negation particles we found that not incrementing the negative word counter degraded the accuracy, so we opted to increment the negative word counter even if a negation particle is found in the tweet. | contrasting |
train_3667 | The performance of the lexicons on external datasets also suggests their ability to be used in classifying new datasets. | there is much room for improvement given the simple method used in evaluation. | contrasting |
train_3668 | In his approach, a Bayesian segmentation algorithm is applied, which is followed by a segment clustering algorithm. | the author tested his approach by using only documents with a few transitions among authors. | contrasting |
train_3669 | Automated Text Scoring (ATS) provides a cost-effective and consistent alternative to human marking. | in order to achieve good performance, the predictive features of the system need to be manually engineered by human experts. | contrasting |
train_3670 | We report the Spearman's rank correlation coefficient ρ, Pearson's product-moment correlation coefficient r, and the root mean square error (RMSE) between the predicted scores and the gold standard on our test set, which are considered more appropriate metrics for evaluating essay scoring systems (Yannakoudakis and Cummins, 2015). | we also report Cohen's κ with quadratic weights, which was the evaluation metric used in the Kaggle competition. | contrasting |
train_3671 | As described above, the SVM model has rich linguistic knowledge and consists of hand-picked features which have achieved excellent performance in similar tasks (Yannakoudakis et al., 2011). | in terms of RMSE, it is among the lowest performing models (8.85), together with 'BLSTM' and 'Twolayer BLSTM'. | contrasting |
train_3672 | (2016) recently used eyetracking recordings to induce PoS models. | a weakness of eye-tracking data is that while eye movement surely does reflect the temporal aspect of cognitive processing, it is only a proxy of the latter and does not directly represent which processes take place in the brain. | contrasting |
train_3673 | Our best explanation for why the incorporation of preceding fMRI measurements is beneficial to our model, is that the grammatical function of a token may be predictable from a reader's cognitive state while reading preceding tokens. | note that the measurements at the far ends of the time window only factor into the token vector to a small degree as a consequence of the Gaussian weighting. | contrasting |
train_3674 | (2013) addressed the ontology mismatch problem (i.e., two ontologies using different vocabularies) by first parsing a sentence into a domainindependent underspecified logical form, and then using an ontology matching model to transform this underspecified logical form to the target ontology. | their method is still hard to deal with the 1-N and the N-1 mismatch problems between natural language and target ontologies. | contrasting |
train_3675 | Traditionally, if a sentence contains a word which is not covered by the lexicon, it will cannot be correctly parsed. | with the help of sentence rewriting, we may rewrite the OOV words into the words which are covered by our lexicons. | contrasting |
train_3676 | This is especially the case for unsupervised resolvers. | a deep network can handle difficult cases like this via learning representations that make lexically different contexts look similar. | contrasting |
train_3677 | Finally, given that our network is trained in a supervised manner, we can extensively employ lex-ical features and use them in combination with other types of features that have been shown to be useful for AZP resolution. | rather than employing words directly as features, we employ word embeddings trained in an unsupervised manner. | contrasting |
train_3678 | They showed that supplementing a Bayesian linear ridge regression model (BLRR) with data from one other source domain is beneficial when there is limited target domain data. | it was shown that simply using the source domain data as extra training data outperformed the EA domain adaptation approach in three out of four cases. | contrasting |
train_3679 | These results indicate that because the preference-ranking model performs so well, only a few target-task training instances are needed for the linear-regression step of All-MTL-cTAP. | all-MTL-TaP uses all of the training instances in its final linear regression step, and performs significantly worse on a number of prompts. | contrasting |
train_3680 | Thus, answering factoid questions will be straightforward once they are converted into the corresponding structured form. | due to complexity of language, converting natural language questions to structure forms remains an open challenge. | contrasting |
train_3681 | There is an alternative factorization p(s, r|q) = p(s|q) • p(r|s, q). | it is rather challenging to estimate p(s|q) directly due to the vast amount of entities (> 10 6 ) in a KB. | contrasting |
train_3682 | As the two problems defined by equation (16) take the standard form of classification, theoretically, cross entropy can used as the training objective. | computing the partition function is often intractable, especially for p θs (s|r, q), since there can be millions of entities in the KB. | contrasting |
train_3683 | Alternative focus with CRF RNN-CRF based models have achieved the state-of-the-art performance on various sequence labeling tasks (Huang et al., 2015;. | the labeling task we consider here is relatively unsophisticated in the sense that there are only two categories of labels -part of subject string (SUB) or not (O). | contrasting |
train_3684 | Insepcting the table horizontally, when BiGRU is employed as the subject network, the accuracy is consistently higher regardless of relation network structures. | the margin is quite narrow, especially compared to the effect of varying the relation network structure the same way. | contrasting |
train_3685 | Accordingly, various semantic classifications of AO-selecting verbs have been developed, e.g., (Kiparsky and Kiparsky, 1970;Karttunen, 1971;Karttunen, 2012), some of them explicitly in the context of NLP (Nairn et al., 2006;Saurí, 2008). | these classifications are constructed manually and often quite limited in coverage. | contrasting |
train_3686 | These classes contain many AO-selecting verbs that were not covered by Levin's classification. | verbNet does not provide information about the modal meaning of AO-selecting verbs and does not reflect fine-grained distinctions between various kinds of modality. | contrasting |
train_3687 | While approaches to exploit the syntactic behavior of verbs for the automatic acquisition of semantic verb classes from corpora have been developed in the past, they were used to recover only small verb classifications: Schulte im Walde (2006)'s work considered a semantically balanced set of 168 German verbs, Merlo and Stevenson (2001) used 60 English verbs from three particular semantic classes. | to previous work, we consider a large set of more than 600 German AO-selecting verbs and focus on their modal meaning (i.e., expressing factuality or uncertainty). | contrasting |
train_3688 | Parsing accuracy on the coordination-enhanced corpus is higher than on the original trees. | the numbers are not strictly comparable, as the test sets contain trees with somewhat different number of constituents. | contrasting |
train_3689 | Using a large-scale crowdsourcing experiment, we demonstrate that human annotators are generally accurate in assessing the traits of others. | they make systematically different types of errors compared to a prediction model trained using the bag-of-words assumption. | contrasting |
train_3690 | Recent results on social media data report a performance of over 90% for gender classification and a correlation of r ∼ 0.85 for age prediction (Sap et al., 2014). | authors can introduce their biases in text (Recasens et al., 2013). | contrasting |
train_3691 | The second is identifying topical aspects and corresponding opinions, e.g., developing topic models to predict fine-grained aspect ratings (Titov and McDonald, 2008;Wang et al., 2011). | all those works emphasize population-level analysis, which applies a global model on all users and therefore fails to recognize the heterogeneity in which different users express their diverse opinions. | contrasting |
train_3692 | (Li et al., 2010) developed an online learning algorithm to continue training personalized classifiers based on a given global model. | all of these aforementioned solutions perform model adaptation from a fixed global model, such that the learning of personalized models is independent from each other. | contrasting |
train_3693 | For example, the phrase "a waste of money" generally represents negative opinions across all users; and it is very unlikely that anybody would use it in a positive sense. | members of some segments of a social structure tend to feel certain emotions more often or more intensely than members of other segments (Hochschild, 1975). | contrasting |
train_3694 | Our motivation to build a concept graph from a technical corpus is to improve performance at the task of reading list generation. | an applied evaluation makes it harder to judge the quality of the concept graph itself. | contrasting |
train_3695 | Another natural evaluation of an automatically generated concept graph would be to compare it to a human-generated gold standard, where an expert has created concept nodes at the optimal level of generality and linked these by her understanding of the conceptual dependencies among concepts in the domain. | there are several difficulties with this approach: (1) It is quite labor-intensive to manually generate a concept graph; (2) we expect only moderate agreement between graphs produced by different experts, who have different ideas of what concepts are important and distinct and which concepts are important to understanding others; and (3) the concept graphs we learn from a collection of documents will differ significantly from those we imagine, without these differences necessarily being better or worse. | contrasting |
train_3696 | In this work, we assume that a topic model provides a reasonable proxy for the concepts a person might identify in a technical corpus. | topic modeling approaches are better at finding general areas of research than at identifying fine-grained concepts like those shown in Figure 1. | contrasting |
train_3697 | Stolcke (2002) find that the approximation actually decreases perplexity, which we also see in the experiments ( §6). | approximation only happens when the model backs off, which is less likely to happen in fluent sentences used for perplexity scoring. | contrasting |
train_3698 | 3 The log probability of tuning data w is which expands according to the definition of p LL n i The gradient with respect to λ i has a compact form where CH is cross entropy. | computing the cross entropy directly would entail a sum over the vocabulary for every word in the tuning data. | contrasting |
train_3699 | In the same direction, Rothe and Schütze (2015) trained embeddings by mixing words, lexemes and synsets, and introducing a set of features based on calculations on the resulting representations. | none of these techniques takes full advantage of the semantic information contained in embeddings. | contrasting |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.