id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_5600 | Given the magnitude of these networks, scalably monitoring their activity necessarily relies on automated tools, and notably on NLP tools. | little is known about what characteristics texts communicated through the Darknet have, and how well off-the-shelf NLP tools do on this domain. | contrasting |
train_5601 | For some of the categories, legal and illegal activities are distinguished. | the automatic classification presented in their work focuses on the distinction between different classes of illegal activity, without addressing the distinction between legal and illegal ones, which is the subject of the present paper. | contrasting |
train_5602 | Results when training and testing on forums data are much worse for the neural-based systems, probably due to the much noisier and more varied nature of the data. | the SVM model achieves an accuracy of 85.3% in the full setting. | contrasting |
train_5603 | in Section 5), we conclude that the distribution of parts of speech alone is not as strong a signal as the word forms for distinguishing between the domains. | the SVM model does manage to distinguish between the texts even in this setting. | contrasting |
train_5604 | To reduce human efforts and scale the process, automated CTA transcript parsing is desirable. | this task has unique challenges as (1) it requires the understanding of long-range context information in conversational text; and (2) the amount of labeled data is limited and indirect-i.e., context-aware, noisy, and low-resource. | contrasting |
train_5605 | (2018) explored modeling cognitive knowledge in well-defined tasks with neural models. | for the most general setting that extract cognitive processes from interviews, we still need substantial expertise to interpret the interview transcript. | contrasting |
train_5606 | Recent advances in machine reading comprehension, textual entailment (Devlin et al., 2018) and relation extraction (Zhang et al., 2017) shows the contemporary NLP models have the capability of capturing causal relations in some degree. | it is still an open problem to extract procedural information from text. | contrasting |
train_5607 | As an application-oriented task, it is beneficial to properly introduce human efforts to monitor and optimize our expansion model. | the design of this human-model interaction faces several challenges. | contrasting |
train_5608 | Classic methods prior to the advance of applied machine learning in this domain typically try to produce grammatical English with generative grammar (Chapman and Davida, 1997). | such generation methods fall short in terms of statistical imperceptibility (Meng et al., 2008). | contrasting |
train_5609 | The last term is the entropy of the partitions at the current step which is bounded between zero and k. Hence, the KL divergence is at most k at each step. | if the probability mass is roughly evenly distributed over each of the 2 k bins, then the KL divergence is close to zero. | contrasting |
train_5610 | We observe that the GCNN outperforms the baseline models (CNN-RE/RNN-RE) in both datasets. | in the CDR dataset, the performance of GCNN is 1.6 percentage points lower than the best performing system of (Gu et al., 2017). | contrasting |
train_5611 | In their experiments, the importance of few-shot learning is not taken into account since the criminal charges that appear fewer than 80 times are filtered out. | in reality, a court is able to judge even under rare conditions. | contrasting |
train_5612 | Among those, the adversarial target input (z = z) shows the greatest decrease of 1.87 BLEU points, and removing language models have the least impact on the BLEU score. | language models are still important in reducing the size of the candidate set, regularizing word embeddings and generating fluent sentences. | contrasting |
train_5613 | As the model samples from ground truth word and the sentence-level oracle word at each step, the two sequences should have the same number of words. | we can not assure this with the naive beam search decoding algorithm. | contrasting |
train_5614 | While most neural machine translation (NMT) systems are still trained using maximum likelihood estimation, recent work has demonstrated that optimizing systems to directly improve evaluation metrics such as BLEU can substantially improve final translation accuracy. | training with BLEU has some limitations: it doesn't assign partial credit, it has a limited range of output values, and it can penalize semantically correct hypotheses if they differ lexically from the reference. | contrasting |
train_5615 | This is not surprising given that BLEU is the standard metric for system comparison at test time. | bLEU is not without problems when used as a training criterion. | contrasting |
train_5616 | From thexe examples we can see that when BLEU scores are very different, the semantics of the sentence can still be preserved. | we observe that often in these cases, the SIM scores of the sentences tend to be similar. | contrasting |
train_5617 | These documents form part of a large corpora of knowledge (e.g., research papers and encyclopedias) mostly accessible through text-based search engines. | to better exploit this knowledge (i.e., not just recovering text fragments or URLs) in automated processes, it is necessary to process the text and extract the pieces of information in a semantic format useful for machine analysis. | contrasting |
train_5618 | In the eHealth-KD challenge context, the selection of which algorithms and hyper-parameters to use for each subtask can be framed as a classic AutoML problem. | there are additional high-level decisions, such as whether to solve subtasks sequentially or combined, that cannot be easily represented in traditional AutoML frameworks. | contrasting |
train_5619 | As a case study, we apply our proposal to the eHealth-KD challenge, and achieve state-of-the-art results in one of the evaluation scenarios proposed. | this approach can be extended to other machine learning scenarios. | contrasting |
train_5620 | Furthermore, the challenge defines three evaluation scenarios in which different subtasks are considered. | for the purpose of this research, we focus on Scenario 1, which considers all three subtasks and is thus the most complete and challenging. | contrasting |
train_5621 | A total of 6 researchers submitted their approaches in the eHealth-KD challenge, of which 5 evaluate in Scenario 1. | one of these participants did not submit a description paper, hence, it will not be considered in this research. | contrasting |
train_5622 | Our proposal is based on AutoML, using the metaheuristic grammatical evolution (O'Neill and Ryan, 2001) for defining and exploring the space of solution for dealing with the Scenario 1 of the eHealth-KD challenge. | the optimization process is different to the traditional grammatical evolution formulation because our grammar involves both discrete and continuous parameters. | contrasting |
train_5623 | The performance of a pipeline will depend, in general, of complex interactions between its components that are not completely captured in a simple linear regression model. | there is a large correlation between feature weights and pipeline performance, as demonstrated by the coefficient of determination (R 2 = 0.763) of the regression. | contrasting |
train_5624 | Concretely, we first Then our attention mechanism models (trigger, context word) pair's relevance with a Multi-Layer Perceptron (MLP), and uses a softmax function normalizing relevance scores to attention weights: Given the attention weights, the lexical-specific context representation is summarized as c 0 = And the final lexical-specific representation of instance x is the concatenation of its token representation h 0 and the lexical-specific context representation c 0 , i.e., The lexical-specific representation can effectively disambiguate trigger words by capturing (trigger, context word) associations. | this representation is lexical-specific, thus hard to generalize well to sparse/unseen words. | contrasting |
train_5625 | Then the PCNNs model (Zeng et al., 2015) designed the multi-instance learning paradigm for RE. | the PCNNs model suffers the issue of the selection of sentences. | contrasting |
train_5626 | Graph Convolution Networks (Kipf and Welling, 2016;Schlichtkrull et al., 2017) and Graph attention networks (Veličković et al., 2017) also learn neighborhood based representations of nodes. | they do not learn a query-dependent composition of the neighborhood which is sub-optimal as also seen our in experiments and noted previously (Dettmers et al., 2017). | contrasting |
train_5627 | 2016proposed a neighborhood mixture model which is closely related. | their proposed model learns a fixed mixture over neighbors as opposed to learning an adaptive mixture based on the query, and requires storing an embedding parameter for every entity-relation pair which can be prohibitively large, potentially O(V e × V r ) whereas our model only requires O(V e + V r ). | contrasting |
train_5628 | contribution as a synthesizer and that he is an instrumentalist for Keyboards to infer that he is a musician. | for queries like nationality, the model attends to neighbors like place of birth (see query for Burt Young) and places lived. | contrasting |
train_5629 | On the one hand, while syntactic information (i.e., the dependency tree) can directly connect "will" to "go", it will also promote some noisy words (i.e., "back") at the same time due to the direct links (see the dependency tree in Figure 1). | while deep learning models with the sequential structure can help to downgrade the noisy words (i.e., "back") based on the semantic importance and the close distance with "go", these models will struggle to capture "will" for the factuality of "go" due to their long distance. | contrasting |
train_5630 | , h n ), it is possible to use the hidden vector corresponding to the anchor word h k as the features to perform factuality prediction (as done in (Rudinger et al., 2018)). | despite the rich context information over the whole sentence, the features in h k are not directly designed to focus on the import context words for factuality prediction. | contrasting |
train_5631 | Since this distant data is created using rulebased classifiers, given a large amount of training data, the baseline model can achieve high performance as it learns to infer these rules. | our aim is to improve the performance of the event ordering model on moderately sized datasets, where the knowledge induction from timex embeddings play a larger role. | contrasting |
train_5632 | This demonstrates that the temporal model does not use the knowledge from time expressions when making temporal relation predictions. | in the ELMo setting, we observed a larger drop in performance by masking out the time expressions compared to GloVe embeddings. | contrasting |
train_5633 | Differentiable Neural Computer (DNC) extends the NTM to address the issue by introducing a temporal link matrix, replacing the least used memory when the memory is full. | this method is a rule-based one that cannot maximize the performance on a given task. | contrasting |
train_5634 | TriviaQA), storing the evidence sentences that precede span words may be useful as they may provide useful context. | using only discrete policy gradient method, we cannot preserve such context instances. | contrasting |
train_5635 | ing the best accuracy, which may be due to its ability to capture global relative importance of each memory entry. | the gap between EMR-Transformer and EMR-biGRU diminishes as the size of memory increases, since then the size of the memory becomes large enough to contain all the frames necessary to answer the question. | contrasting |
train_5636 | With treebank data such as Penn Treebank (Marcus et al., 1993), we have pairs of (x, y) to train the two components respectively, and get high-quality language model with the parser providing grammar knowledge. | due to the expensive cost of accurate parsing annotation, Figure 1: Overall framework of neural variational language model (NVLM). | contrasting |
train_5637 | Subsequent work has homed in on language modeling (LM) pretraining, finding that such mod-els can be productively fine-tuned on intermediate tasks like natural language inference before transferring to downstream tasks (Phang et al., 2018). | we identify two open questions: (1) How effective are tasks beyond language modeling in training reusable sentence encoders (2) Given the recent successes of LMs with intermediate-task training, which tasks can be effectively combined with language modeling and each other. | contrasting |
train_5638 | Multitask pretraining or intermediate task training offers modest further gains. | we see several worrying trends: • The margins between substantially different pretraining tasks can be extremely small in this transfer learning regimen and many pretraining tasks struggle to outperform trivial baselines. | contrasting |
train_5639 | propose to firstly parse a question to a coarse logical form then a fine-grained one based on a neural architecture. | these approaches miss the opportunity to utilize question decomposition information for complex question semantic parsing. | contrasting |
train_5640 | The task involves assessing whether a given premise entails a given hypothesis. | to other entailment datasets mentioned previously, the hypotheses in SciTail are created from science questions while the corresponding answer candidates and premises come from relevant web sentences retrieved from a large corpus. | contrasting |
train_5641 | Dependency parsing allows us to design our extraction method such that each S1 and S2 is interpretable as a full sentence in isolation, and the appropriate conceptual relation holds between the pair. | occasionally we get ungrammatical sentences or the wrong pair of sentences for a relation. | contrasting |
train_5642 | We do not allow the classifier to access the underlying discourse relation type and we only provide the individual sentence embeddings as input features. | patterson and Kehler (2013) used a variety of discrete features provided by the pDTB dataset for their classifier, including the hand-annotated relation types. | contrasting |
train_5643 | Adopted by several works later on (Miller et al., 1996;Zettlemoyer and Collins, 2009;Suhr et al., 2018), ATIS has only a single domain for flight planning which limits the possible SQL logic it contains. | to ATIS, SParC consists of a large number of complex SQL queries (with most SQL syntax components) inquiring 200 databases in 138 different domains, which contributes to its diversity in query semantics and contextual dependencies. | contrasting |
train_5644 | In comparison, SParC is significantly different as it (1) contains mode complex contextual dependencies, (2) has greater semantic coverage, and (3) adopts a cross-domain task setting, which make it a new and challenging cross-domain contextdependent text-to-SQL dataset. | sParC has overcome the domain limitation of ATIs by covering 200 different databases and has a significantly larger natural language vocabulary. | contrasting |
train_5645 | The domain by its nature requires the user to express multiple constraints in separate utterances and the user is intrinsically motivated to interact with the system until the booking is successful. | the interaction goals formed by Spider questions are for open-domain and general-purpose database querying, which tend to be more specific and can often be stated in a smaller number of turns. | contrasting |
train_5646 | It can be seen that the DCTbased embedding creation technique needs more components to achieve reasonable performance, as compared to PCA, because PCA learns the basis vectors in a data-driven way while DCT assumes cosine functions as bases. | since it does not need to learn the bases, and therefore makes less errors than PCA, DCT is more performant than PCA when we utilize more components. | contrasting |
train_5647 | The human performance numbers in Table 1 shows that overall our annotators stick it to the Muppets on GLUE. | on MRPC, QQP, and QNLI, Bigbird and BERT outperform our annotators. | contrasting |
train_5648 | In the first example, the model originally (incorrectly) assumes that "Toronto" refers to the city, tagging it as a GPE. | after resolving the semantic role -determining that "Toronto" is the thing getting "smoked" (ARG1) -the entity-typing decision is revised in favor of ORG (i.e. | contrasting |
train_5649 | In the second example, the model initially tags "today" as a common noun, date, and temporal modifier (ARGM-TMP). | this phrase is ambiguous, and it later reinterprets "china today" as a proper noun (i.e. | contrasting |
train_5650 | As with BERT, we observe that the weights are only weakly concentrated for the relations and SPR tasks. | unlike BERT, we see only a weak concentration in the weights on the coreference task, which agrees with the finding of Tenney et al. | contrasting |
train_5651 | The BiLSTM layer in the baseline and our model is capable of capturing high-order information to some extent. | without prior knowledge of high-order parts, it may require more training data to learn this capability than a high-order decoder. | contrasting |
train_5652 | Mean field variational inference slows down training and parsing by 35% and 20% respectively compared with the baseline. | loopy belief propagation slows down training and parsing by 65% and 67% respectively compared with the baseline. | contrasting |
train_5653 | (2016b) proposed a dataset similar to us, i.e., based on the TV show Friends. | their corpus only includes the textual modality and is thus not multimodal in nature. | contrasting |
train_5654 | Due to the speaker overlap across splits, the model can leverage speaker regularities for sarcastic tendencies. | we do not observe the same trend for the best multimodal variant (text + video) where the score barely improves. | contrasting |
train_5655 | (2017), many sentences either do not express an argument or cannot be understood out of context. | our dataset explicitly provides the sequence of claims in an argument path that leads to any particular claim, which can enable an argument generation system to generate relevant claims, with a particular stance and at the right level of specificity. | contrasting |
train_5656 | Some of these studies have shown that simple linear classifiers with uni-gram and n-gram features are effective for this task (Somasundaran and Wiebe, 2010;Hasan and Ng, 2013;Mohammad et al., 2016). | in our setting, since we try to predict the stance between all pairs of claims on an argument path, rather than simply claims that are directed towards the thesis or the parent claim, we find that the models with a hierarchical representation of the argument path, i.e. | contrasting |
train_5657 | These models handle sentiment composition implicitly and predict sentiment polarities only based on embeddings of current nodes. | we model sentiment explicitly. | contrasting |
train_5658 | The model includes a bidirectional LSTM encoder and a unidirectional LSTM decoder with attention, and only the encoder is used for downstream task-specific models. | pre-training is limited by the availability of parallel corpora. | contrasting |
train_5659 | We are surprised to find that BERT's peak performance of 77% on the Argument Reasoning Comprehension Task reaches just three points below the average untrained human baseline. | we show that this result is entirely accounted for by exploitation of spurious statistical cues in the dataset. | contrasting |
train_5660 | To answer our question in the introduction: BERT has learned nothing about argument comprehension. | our investigations confirmed that BERT is indeed a very strong learner. | contrasting |
train_5661 | These reasons are informative about which stance is taken, because they give more details on how a stance is developed. | simply comparing the reasons may not be sufficient to predict stance (dis)agreement, as sometimes people can take the same stance but give different reasons (e.g., the points outlaws having guns and freedom of speech mentioned in Table 1). | contrasting |
train_5662 | This indicates that the refined representation is more learnable and is able to extract the interdependence between aspect and the corresponding target in the context. | the performance of sentiment classification is improved certainly in comparison with the remarkable models (Delayed-memory and SenticLSTM). | contrasting |
train_5663 | Although the neural network-based approach (Potash et al., 2017;Eger et al., 2017) achieves high performances for ASP, it does not explicitly take into account useful linguistic clues. | prior works demonstrate that linguistic features, particularly discourse connectives, are strong clues to predict the structure for ASP (Lawrence and Reed, 2015;Stab and Gurevych, 2017). | contrasting |
train_5664 | The given input is a paragraph consisting of T tokens w 1: the other reason is that consequently, in conclusion, from the above views, although first, as you can see that in this essay , the reasons for why i agree that it is a debatable subject that unfortunately although some argue that furthermore , it 's undeniable that the in short, another thing that put big cities in front of small towns is in conclusion, despite the contribution of it to the society, however, some say that Some people may argue that (AM1) children will be more material, neglect their study for earning money or be exploited by the employers(AC1). | (AM2) if children get good care and instructions from their parents, they can take advantages of the work to learn valuable things and avoid going in a wrong way(AC2). | contrasting |
train_5665 | Instead of making a prediction based on the comparison result of a single alignment process, a stacked model with multiple alignment layers maintains its intermediate states and gradually refines its predictions. | suffering from inefficient propagation of lower-level features and vanishing gradients, these deeper architectures are harder to train. | contrasting |
train_5666 | The simpler implementation of the fusion layer leads to evidently worse performance, indicating that the fu- sion layer cannot be further simplified. | the alignment layer and the prediction layer can be simplified on some of the datasets. | contrasting |
train_5667 | While translational models learn embeddings using simple operations and limited parameters, they produce low quality embeddings. | cNN based models learn more expressive embeddings due to their parameter efficiency and consideration of complex relations. | contrasting |
train_5668 | In contrast, CNN based models learn more expressive embeddings due to their parameter efficiency and consideration of complex relations. | both translational and CNN based models process each triple independently and hence fail to encapsulate the semantically rich and latent relations that are inherently present in the vicinity of a given entity in a KG. | contrasting |
train_5669 | Our model achieves these objectives by assigning different weight mass (attention) to nodes in a neighborhood and by propagating attention via layers in an iterative fashion. | as the model depth increases, the contribution of distant entities decreases exponentially. | contrasting |
train_5670 | One of the main capabilities of these models is that they implicitly exploit entailment relations such as person born in country entails person be from country (Riedel et al., 2013). | entailment relations are not learned explicitly. | contrasting |
train_5671 | This would suggest we can define a new link prediction score based on entailment relations: S ent q,e 1 ,e 2 = max r∈R: r→q S r,e 1 ,e 2 . | since we do not have access to the entailment relations and can only rely on the predictions, Equation 3 is likely to be very noisy. | contrasting |
train_5672 | This is because the other baselines have also access to the whole set of triples ( §4.4). | for evaluating the link prediction model, we compute entailment scores by only considering the predictions in the training set. | contrasting |
train_5673 | Deudon (2018) build a sentence-reformulating deep generative model whose objective is to measure the semantic similarity between a sentence pair. | their work cannot be applied to a multi-class classification problem, and the generative objective is only used in pre-training, not considering the joint optimization of the generative and the discriminative objective. | contrasting |
train_5674 | As described by Glavaš andŠnajder (2014a), there have been efforts that have focused on detecting temporal and spatial subevent containment individually. | it is clear that subevent detection requires both simultaneously. | contrasting |
train_5675 | While ActivityNet starts o↵ harder for BERT (25.5%), it also proves di cult for humans (60%). | wikiHow starts easier for BERT (41.1%) and humans find the domain almost trivial (93.5%). | contrasting |
train_5676 | Recently, there has been a surge in neural encoderdecoder techniques which are trained with input utterances and corresponding annotated output programs (Dong and Lapata, 2016;Jia and Liang, 2016). | the performance of these strongly supervised methods is restricted by the size and the diversity of training data i.e. | contrasting |
train_5677 | Herzig and Berant (2017) propose semantic parsing models using supervised learning in a multi-domain setup and is the closest to our work. | none of the existing works inspect the problem of multi-domain semantic parsing in a weak supervision setting. | contrasting |
train_5678 | These are the two ways for the log-loss approach to make predictions with high accuracy: always giving very high score for the entailment hypothesis and low score for the contradiction hypothesis, but giving either very high or very low score for the neutral hypothesis. | the margin-loss gives more intuitive scores for these two examples. | contrasting |
train_5679 | 1 Conceptually, GLEN is similar to the explicit retrofitting model of , who focus on the symmetric semantic similarity relation. | gLEN has to account for the asymmetric nature of the LE relation. | contrasting |
train_5680 | This reliance on attention may lead one to expect decreased performance on commonsense reasoning tasks (Roemmele et al., 2011;Zellers et al., 2018) compared to RNN (LSTM) models (Hochreiter and Schmidhuber, 1997) that do model word order directly, and explicitly track states across the sentence. | the work of (Peters et al., 2018a) suggests that bidirectional language models such as BERT implicitly capture some notion of coreference resolution. | contrasting |
train_5681 | That is, they are unable to resolve pronouns in light of abstract/implicit referrals that require background knowledge -see (Saba, 2018) for more detail. | this is beyond the task of WSC. | contrasting |
train_5682 | Neural networks have proven highly effective in natural language processing (NLP) tasks, outperforming other machine learning methods and even matching human performance (Hassan et al., 2018;Nangia and Bowman, 2018). | supervised models require many per-task annotated training examples for a good performance. | contrasting |
train_5683 | For the seq2seq-T (S2S-T) model (Qin et al., 2018), the comment is generated mainly based on the clue "ancient costume" in the title. | because "ancient costume" is not frequently seen in the comments (in the training set). | contrasting |
train_5684 | • Our CFNet is significantly better than two baselines (PGNet, NQG), our MSNet, and our CorefNet. | the difference between our CFNet and our FlowNet is not significant. | contrasting |
train_5685 | Neural models for automatic question generation using the standard sequence to sequence paradigm have been shown to perform reasonably well for languages such as English, which have a large number of training instances. | large training sets are not available for most languages. | contrasting |
train_5686 | In preliminary experiments, we explored a learned policy for operator selection. | we observed that the learned policy quickly collapses to a nearly deterministic choice of Rep φ 3 . | contrasting |
train_5687 | (2017) proposed to use an auxiliary model, trained to extract structured records from text, for evaluation. | the extraction model presented in that work is limited to the closed-domain setting of basketball game tables and summaries. | contrasting |
train_5688 | DROP is far from all datasets because it requires quantitative reasoning that is missing from other datasets. | it is relatively close to HOTPOTQA and WIKIHOP, which target multi- Figure 1: A 2D-visualization of the similarity between different datasets using the force-directed placement algorithm. | contrasting |
train_5689 | (2018) generate explanations and predictions for the natural language inference problem (Camburu et al., 2018). | the authors report that interpretability comes at the cost of loss in performance on the popular Stanford Natural Language Inference (Bowman et al., 2015) dataset. | contrasting |
train_5690 | (2018) use human explanations to train a neural network model on the SNLI dataset (Bowman et al., 2015). | they obtain explanations at the cost of accuracy. | contrasting |
train_5691 | We only present annotators with query instances for which both models output the same answer. | we do not restrict these answers to be the ground truth. | contrasting |
train_5692 | Given a pair of words (i, j) and associated word vectors , we compute the similarity between two words by their vector similarity. | we encode this similarity in a weighted adjacency matrix A: nodes are only connected to their knearest neighbors (Section 6.2 examines the sensitivity to k); all other edges become zero. | contrasting |
train_5693 | For ZH, HI, and KO, the improvement comes from selecting better mappings during the adversarial step. | modularity does not improve on all languages (e.g., VI) that are reported to fail by Hoshen and Wolf (2018). | contrasting |
train_5694 | Our results suggest that simple properties of a language-the word inventory and (to a lesser extent) the raw character sequence length-are statistically significant indicators of modeling difficulty within our large set of languages. | we fail to reproduce our earlier results from Cotterell et al. | contrasting |
train_5695 | As Figure 2 shows, the number of merges that maximizes log-likelihood of our dev set differs from language to language. | 8 as we will see in Figure 3, tuning this parameter does not substantially influence our results. | contrasting |
train_5696 | Because it is multiplicative, Model 1 appropriately predicts that in each language , intents with large will not only have larger values but these values will vary more widely. | model 1 is homoscedastic: the variance 2 of log( / ) is assumed to be independent of the independent variable , which predicts that the distribution of should spread out linearly as the information content increases: e.g., ( That assumption is questionable, since for a longer sentence, we would expect log / to come closer to its mean as the random effects of individual translational choices average out. | contrasting |
train_5697 | Speakers prefer shorter dependencies in both production and processing, and average dependency lengths tend to be much shorter than would be expected from randomly-generated parses (Futrell et al., 2015;Liu et al., 2017). | there is substantial variability between languages, and it has been proposed, for example, that head-final languages and case-marking languages tend to have longer dependencies on average. | contrasting |
train_5698 | We did implement this model and test it on data, but not only was fitting it much harder and slower, it also did not yield particularly encouraging results, leading us to omit it from the main text. | c Goodness of fit of our difficulty estimation models Figure 6 shows the log-probability of held-out data under the regression model, by fixing the estimated difficulties (and sometimes also the estimated variance 2 ) to their values obtained from training data, and then finding either MAP estimates or posterior means (by running HMc using STAN) of 29 version (a) is then deficient since it then incorrectly allocates some probability mass to + < 0 and thus < 0 is possible. | contrasting |
train_5699 | This provides clear evidence of M-BERT's multilingual representation ability, mapping structures onto new vocabularies based on a shared representation induced solely from monolingual language model training data. | cross-script transfer is less accurate for other pairs, such as English and Japanese, indicating that M-BERT's multilingual representation is not able to generalize equally well in all cases. | 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.