HHousen/ParaSCI · Datasets at Hugging Face

sentence1 stringlengths 16 446	sentence2 stringlengths 14 436
the weights 位 m in the log-linear model were trained using minimum error rate training with the news 2009 development set .	the weights of the different feature functions were optimised by means of minimum error rate training on the 2013 wmt test set .
li et al use sense paraphrases to estimate probabilities of senses and carry out wsd .	li et al design several systems that use latent topics to find a most likely sense based on the sense paraphrases and context .
in this setting , traditional transfer methods will always predict the same label .	unlike these methods , our approach assumes no label annotations in the target domain .
each of them was lemmatised and tagged using the treetagger .	the corpus was automatically pos-tagged with treetagger .
table 2 presents the translation performance in terms of various metrics such as bleu , meteor and translation edit rate .	the mt performance in terms of translation edit rate and bleu is shown in figure 4 .
semantic role labeling ( srl ) is the task of identifying the arguments of lexical predicates in a sentence and labeling them with semantic roles ( cite-p-13-3-3 , cite-p-13-3-11 ) .	semantic role labeling ( srl ) is the task of automatic recognition of individual predicates together with their major roles ( e.g . frame elements ) as they are grammatically realized in input sentences .
searchqa and triviaqa show that our system achieves significant and consistent improvement as compared to all baseline methods .	experimental results on real-world datasets show that our model can capture useful information from noisy data and achieve significant improvements on ds-qa as compared to all baselines .
the third diachronic distributional model we will consider comes from bamler and mandt .	this choice of hyperparameters comes from bamler and mandt .
morphologically , arabic is a non-concatenative language .	arabic is a morphologically rich language that is much more challenging to work , mainly due to its significantly larger vocabulary .
we have presented an approach to building a test collection from an existing collection of research papers .	we present an approach to building a test collection of research papers .
representations of word context are based on potential substitutes of a word .	thus substitute vectors represent individual word contexts , not word types .
we extend the existing word embedding learning algorithm and develop three neural networks to learn sswe .	following , we develop a new convolutional neural network based semantic model for semantic parsing .
cook et al and fazly et al take a different approach , which crucially relies on the concept of canonical form .	cook et al and fazly et al rely crucially on the concept of canonical form .
in this paper , we , the fbk-tr team , describe our system participating in task 3 .	in this paper , we describe our system participating in the task 3 , at semeval 2014 .
sentiment analysis is the process of identifying and extracting subjective information using natural language processing ( nlp ) .	sentiment analysis is the task in natural language processing ( nlp ) that deals with classifying opinions according to the polarity of the sentiment they express .
to build a corpus for robocup , 300 pieces of coach advice were randomly selected from the log files of the 2003 robocup coach competition , which were manually translated into english .	for clang , 300 instructions were randomly selected from the log files of the 2003 robocup coach competition and manually translated into english .
in this paper , we apply kernel methods , which enable an efficient comparison of structures .	in this paper , we have engineered and studied several models for relation learning .
zarrie脽 and kuhn argue that multiword expressions can be reliably detected in parallel corpora by using dependency-parsed , word-aligned sentences .	on the other hand , zarrie脽 and kuhn make use of translational correspondences when identifying multiword expressions .
although word embeddings have been successfully employed in many nlp tasks , the application of word embeddings in re is very recent .	word embeddings have become increasingly popular lately , proving to be valuable as a source of features in a broad range of nlp tasks .
heilman et al extended this approach and worked towards retrieving relevant reading materials for language learners in the reap 3 project .	heilman et al combined a language modeling approach with grammarbased features to improve readability assessment for first and second language texts .
we use the ontonotes datasets from the conll 2011 shared task 6 , only for training the out-of-the-box system .	throughout this work , we use the datasets from the conll 2011 shared task 2 , which is derived from the ontonotes corpus .
we propose novel linear associative units ( lau ) to reduce the gradient propagation length inside the recurrent unit .	we propose a linear associative unit ( lau ) which makes a fusion of both linear and nonlinear transformation inside the recurrent unit .
neural networks have recently gained much attention as a way of inducing word vectors .	more recently , neural networks have become prominent in word representation learning .
the language model component uses the srilm lattice-tool for weight assignment and nbest decoding .	uedin has used the srilm toolkit to train the language model and relies on kenlm for language model scoring during decoding .
multi-task learning using a related auxiliary task can lead to stronger generalization and better regularized models .	multi-task learning can integrate different objectives into one model and has previously been shown to help improve model generalisation .
this is a gui-enabled convenience tool that manages datasets and uses the python-based scikitlearn machine learning toolkit .	kindred is a python package that builds upon the stanford corenlp framework and the scikit-learn machine learning library .
methods make use of the information from only one language side .	the existing methods use only the information in either language side .
tests show that using a situated model significantly improves performances over traditional language modeling methods .	results indicate that integration of situational context dramatically improves performance over traditional methods alone .
we used kenlm with srilm to train a 5-gram language model based on all available target language training data .	we used the srilm toolkit to train a 4-gram language model on the english side of the training corpus .
we propose a robust nonparanormal model to learn the stochastic dependencies among the image , the candidate descriptions , and the popular votes .	we propose a robust nonparanormal approach ( cite-p-23-3-18 ) to model the multimodal stochastic dependencies among images , text , and votes .
ideally the third item can be estimated by the forward-backward algorithm recursively for the firstorder or second-order hmms .	ideally , it can be estimated by using the forward-backward algorithm recursively for the first-order or second-order hmms .
we use the mallet implementation of conditional random fields .	our system is based on the conditional random field .
specifically , we use the liblinear svm package as it is well-suited to text classification tasks with large numbers of features and texts .	in particular , we use the liblinear 3 package which has been shown to be efficient for text classification problems such as this .
semantic role labeling ( srl ) is a kind of shallow semantic parsing task and its goal is to recognize some related phrases and assign a joint structure ( who did what to whom , when , where , why , how ) to each predicate of a sentence ( cite-p-24-3-4 ) .	semantic role labeling ( srl ) is the process of extracting simple event structures , i.e. , “ who ” did “ what ” to “ whom ” , “ when ” and “ where ” .
for implicit discourse relation recognition , previous works attempted to automatically generate training data by removing explicit discourse connectives from sentences .	due to the lack of benchmark data for implicit discourse relation analysis , earlier work used unlabeled data to generate synthetic implicit discourse data .
for english , we use the stanford parser for both pos tagging and cfg parsing .	we use the stanford parser for obtaining all syntactic information .
to get a dictionary of word embeddings , we use the word2vec tool 2 and train it on the chinese gigaword corpus .	for word-level embeddings , we pre-train the word vectors using word2vec on the gigaword corpus mentioned in section 4 , and the text of the training dataset .
using the deep learning framework caffe , we extracted image embeddings from a deep convolutional neural network that was trained on the imagenet classification task .	we extract the 4096-dimensional pre-softmax layer from a for-ward pass through a convolutional neural network , which has been pretrained on the imagenet classification task using caffe .
park and levy proposed a language-modeling approach to whole sentence error correction but their model is not competitive with individually trained models .	park and levy proposed an em-based unsupervised approach to perform whole sentence grammar correction , but the types of errors must be predetermined to learn the parameters for their noisy channel model .
the dimensionality of our word embedding layer was set to size 300 , and we use publicly available pre-trained glove word embeddings that we finetune during training .	since it is operated on the word level , we use pre-trained 300-dimensional glove embeddings and keep them fixed during training .
a pun is a form of wordplay in which a word suggests two or more meanings by exploiting polysemy , homonymy , or phonological similarity to another word , for an intended humorous or rhetorical effect ( cite-p-15-3-1 ) .	a pun is a form of wordplay in which one signifier ( e.g. , a word or phrase ) suggests two or more meanings by exploiting polysemy , or phonological similarity to another signifier , for an intended humorous or rhetorical effect .
we use the skll and scikit-learn toolkits .	we used the svm implementation of scikit learn .
segmentation is the first step in a discourse parser , a system that constructs discourse trees from elementary discourse units .	segmentation is the task of dividing a stream of data ( text or other media ) into coherent units .
the probabilistic language model is constructed on google web 1t 5-gram corpus by using the srilm toolkit .	all feature models are estimated in the in-domain corpus with standard techniques .
to test whether a performance difference is statistically significant , we conduct significance tests following the paired bootstrap approach .	to compare the relative quality of different metrics , we apply bootstrapping re-sampling on the data , and then use paired t-test to determine the statistical significance of the correlation differences .
previous methods have used the first or second order co-occurrence , parts of speech , and local collocations .	previous methods have used first or second order co-occurrences , parts of speech , and grammatical relations .
by an unsupervised one , we may raise the question as to whether the end of supervised nlp comes in sight .	we will raise the question whether the end of supervised parsing is in sight .
we used the 300-dimensional glove word embeddings learned from 840 billion tokens in the web crawl data , as general word embeddings .	we use the pre-trained glove 50-dimensional word embeddings to represent words found in the glove dataset .
in this proposal , we propose a corpus based study to examine doctor-patient conversation of antibiotic treatment negotiation .	in this proposal , we propose a corpus-based study of doctor-patient conversations of antibiotic treatment negotiation in pediatric consultations .
our model is effective in dealing with negation phrases ( a typical case of sentiment expressed by sequence ) .	an interesting case study on negation expression processing shows a promising potential of the architecture dealing with complex sentiment phrases .
sentiment analysis is a recent attempt to deal with evaluative aspects of text .	sentiment analysis is the process of identifying and extracting subjective information using natural language processing ( nlp ) .
tiedemann propose cache-based language and translation models , which are built on recently translated sentences .	tiedemann proposed a cache-model to enforce consistent translation of phrases across the document .
as well as temporal constraints , we are able to extract text-to-text .	we are able to achieve significantly better results than with a text-to-words wtmf model .
this paper has described an unsupervised method for inducing semantic frames from instances of each verb .	this paper presents a method for automatically building verb-specific semantic frames from a large raw corpus .
the results of automatic evaluation and manual assessment of title quality show that the output of our system is consistently ranked higher than that of non-hierarchical baselines .	the results of automatic evaluation and manual assessment confirm the benefits of this design : our system is consistently ranked higher than non-hierarchical baselines .
we parse each document using stanford corenlp in order to acquire both dependency , named entity , and coreference resolution features .	for these data , we preprocess the text including using stanford corenlp to split the review documents into sentences and tokenizing all words .
in addition , we use an english corpus of roughly 227 million words to build a target-side 5-gram language model with srilm in combination with kenlm .	to rerank the candidate texts , we used a 5-gram language model trained on the europarl corpus using kenlm .
more recently , watanabe et al and chiang et al presented a learning algorithm using the mira technique .	recently , watanabe et al and chiang et al have developed tuning methods using the mira algorithm as a nucleus .
in this paper , we propose an approach for identifying curatable articles .	in this paper , we demonstrate how our system is constructed .
the log-lineal combination weights were optimized using mert .	the decoding weights were optimized with minimum error rate training .
and the task includes resolving not just a certain type of noun phrase ( e . g . , pronouns ) .	the approach learns from a small , annotated corpus and the task includes resolving not just pronouns but general noun phrases .
keyphrases are useful for a variety of tasks such as summarization , information retrieval and document clustering .	keyphrases are useful in many tasks such as information retrieval , document summarization or document clustering .
mada-arz is an egyptian arabic extension of the morphological analysis and disambiguation of arabic tool .	mada-arz is an egyptian arabic extension of the morphological analysis and disambiguation of arabic .
from the business language testing service ( bulats ) , the proposed approach is found to outperform gps and dnns with mcd in uncertainty-based rejection .	this method outperforms gps and monte-carlo dropout in uncertainty based rejection for automatic assessment .
it includes a top-level ontology developed following the procedure outlined by russell and norvig and originally covered the tourism domain encoding knowledge about sights , historical persons and buildings .	it includes a top-level developed following the procedure outlined by russell and norvig and originally covered the tourism domain encoding knowledge about sights , historical persons and buildings .
part-of-speech tagging is the assignment of syntactic categories ( tags ) to words that occur in the processed text .	part-of-speech tagging is a key process for various tasks such as ` information extraction , text-to-speech synthesis , word sense disambiguation and machine translation .
for all experiments , we used a 4-gram language model with modified kneser-ney smoothing which was trained with the srilm toolkit .	for the language model , we used sri language modeling toolkit to train a trigram model with modified kneser-ney smoothing on the 31 , 149 english sentences .
the level of the agreement was then assessed using the kappa statistic .	the interannotator agreement was measured using the kappa statistic .
negation is a linguistic phenomenon that can alter the meaning of a textual segment .	negation is a complex phenomenon present in all human languages , allowing for the uniquely human capacities of denial , contradiction , misrepresentation , lying , and irony ( horn and wansing , 2015 ) .
wizard could also tell when the query results did not contain the requested title .	the most successful wizard could also tell when the query results did not contain the requested title .
with an empirical evaluation on the penn discourse treebank ( pdtb ) ( cite-p-11-3-7 ) dataset , which yields an f 1 score of 0 . 485 .	our approach achieves an f 1 score of 0.485 on the implicit relation labeling task for the penn discourse treebank .
in portantly , this type of evaluation can measure how well , fourth darpa speech and natural language .	in portantly , this type of evaluation can measure how well , fourth darpa speech and natural language worka structural theory of text can perform .
the tilde system is a moses phrase-based smt system that was trained on the tilde mt platform .	the smt system is a standard phrase-based system that was trained on the tilde mt platform with moses .
we implemented all models in python using the pytorch deep learning library .	we used pytorch to implement the embedding model and gurobi as our ilp solver .
the berkeley framenet project is an ongoing effort of building a semantic lexicon for english based on the theory of frame semantics .	the berkeley framenet is an ongoing project for building a large lexical resource for english with expert annotations based on frame semantics .
jacy is a hand-crafted japanese hpsg grammar that provides semantic information as well as linguistically motivated analysis of complex constructions .	jacy is a type of hand-crafted japanese grammar based on hpsg that can compute a detailed semantic representation .
the penn discourse treebank corpus is the best-known resource for obtaining english connectives .	the penn discourse treebank is the largest available discourseannotated resource in english .
they used stanford parser to create the parse trees for all sentences .	the stanford parser was used to generate the dependency parse information for each sentence .
gram language models are trained over the target-side of the training data , using srilm with modified kneser-ney discounting .	a 5-gram language model with kneser-ney smoothing is trained using s-rilm on the target language .
vector based word representation has powerful capability that captures the phenomenon that words having the similar meanings should appear together .	the word embeddings can provide word vector representation that captures semantic and syntactic information of words .
the annotation scheme leans on the universal stanford dependencies complemented with the google universal pos tagset and the interset interlingua for morphological tagsets .	the annotation scheme is based on an evolution of stanford dependencies , google universal part-ofspeech tags , and the interset interlingua for morphosyntactic tagsets .
to train our reranking models we used svm-light-tk 7 , which encodes structural kernels in svmlight solver .	to train our models , we adopted svm-light-tk 7 , which enables the use of structural kernels in svm-light , with default parameters .
collobert et al adjust the feature embeddings according to the specific task in a deep neural network architecture .	collobert et al apply generic neural network architectures to several sequence labelling tasks and obtain competitive results despite of the task-specific variations .
klementiev et al treat the task as a multi-task learning problem where each task corresponds to a single word , and task relatedness is derived from co-occurrence statistics in bilingual parallel data .	klementiev et al treated the task as a multi-task learning problem where each task corresponds to a single word , and the task relatedness is derived from cooccurrence statistics in bilingual parallel corpora .
later , miwa and bansal have implemented an end-to-end neural network to construct a context representation for joint entity and relation extraction .	miwa and bansal adopted a bidirectional tree lstm model to jointly extract named entities and relations under a dependency tree structure .
word sense disambiguation ( wsd ) is a difficult natural language processing task which requires that for every content word ( noun , adjective , verb or adverb ) the appropriate meaning is automatically selected from the available sense inventory 1 .	word sense disambiguation ( wsd ) is a widely studied task in natural language processing : given a word and its context , assign the correct sense of the word based on a predefined sense inventory ( cite-p-15-3-4 ) .
for language models , we use the srilm linear interpolation feature .	in the case of the trigram model , we expand the lattice with the aid of the srilm toolkit .
we successfully apply the attention scheme to detect word senses and learn representations according to contexts with the favor of the sememe annotation .	we also analyze several cases in wsd and wrl , which confirms our models are capable of selecting appropriate word senses with the favor of sememe attention .
word sense disambiguation is the task of determining the particular sense of a word from a given set of pre-defined senses .	word sense disambiguation is the task of computationally determining the meaning of a word in its context .
our base model is a transition-based neural parser of chen and manning .	our parsing model is built based on the work of chen and manning .
we use the pre-trained glove 50-dimensional word embeddings to represent words found in the glove dataset .	as input to the aforementioned model , we are going to use dense representations , and more specifically pre-trained word embeddings , such as glove .
we used the meetings from the icsi meeting data , which are recordings of naturally occurring meetings .	we used the icsi meeting corpus , which contains naturally occurring meetings , each about an hour long .
at a large scope , they facilitate tremendously the costly but unavoidable process of semi-automatic lexical acquisition .	nevertheless , large-scope lrs are justified because they facilitate the unavoidable process of large-scale semi-automatic lexical acquisition .
semantic role labeling ( srl ) is the task of labeling predicate-argument structure in sentences with shallow semantic information .	semantic role labeling ( srl ) is the process of extracting simple event structures , i.e. , “ who ” did “ what ” to “ whom ” , “ when ” and “ where ” .
we trained a trigram language model on the chinese side , with the srilm toolkit , using the modified kneser-ney smoothing option .	we used srilm to build a 4-gram language model with interpolated kneser-ney discounting .
our system is based on the phrase-based part of the statistical machine translation system moses .	our translation system is an in-house phrasebased system analogous to moses .
the discursive relations used in this work came from the penn discourse treebank .	they had shown that the penn discourse treebank style discourse relations are useful .
we employed the machine learning tool of scikit-learn 3 , for training the classifier .	for the feature-based system we used logistic regression classifier from the scikit-learn library .
the smt systems were built using the moses toolkit .	we use the popular moses toolkit to build the smt system .
socher et al , 2012 , uses a recursive neural network in relation extraction , and further use lstm .	socher et al train a composition function using a neural network-however their method requires annotated data .

the weights 位 m in the log-linear model were trained using minimum error rate training with the news 2009 development set .

the weights of the different feature functions were optimised by means of minimum error rate training on the 2013 wmt test set .

li et al use sense paraphrases to estimate probabilities of senses and carry out wsd .

li et al design several systems that use latent topics to find a most likely sense based on the sense paraphrases and context .

in this setting , traditional transfer methods will always predict the same label .

unlike these methods , our approach assumes no label annotations in the target domain .

each of them was lemmatised and tagged using the treetagger .

the corpus was automatically pos-tagged with treetagger .

table 2 presents the translation performance in terms of various metrics such as bleu , meteor and translation edit rate .

the mt performance in terms of translation edit rate and bleu is shown in figure 4 .

semantic role labeling ( srl ) is the task of identifying the arguments of lexical predicates in a sentence and labeling them with semantic roles ( cite-p-13-3-3 , cite-p-13-3-11 ) .

semantic role labeling ( srl ) is the task of automatic recognition of individual predicates together with their major roles ( e.g . frame elements ) as they are grammatically realized in input sentences .

searchqa and triviaqa show that our system achieves significant and consistent improvement as compared to all baseline methods .

experimental results on real-world datasets show that our model can capture useful information from noisy data and achieve significant improvements on ds-qa as compared to all baselines .

the third diachronic distributional model we will consider comes from bamler and mandt .

this choice of hyperparameters comes from bamler and mandt .

morphologically , arabic is a non-concatenative language .

arabic is a morphologically rich language that is much more challenging to work , mainly due to its significantly larger vocabulary .

we have presented an approach to building a test collection from an existing collection of research papers .

we present an approach to building a test collection of research papers .

representations of word context are based on potential substitutes of a word .

thus substitute vectors represent individual word contexts , not word types .

we extend the existing word embedding learning algorithm and develop three neural networks to learn sswe .

following , we develop a new convolutional neural network based semantic model for semantic parsing .

cook et al and fazly et al take a different approach , which crucially relies on the concept of canonical form .

cook et al and fazly et al rely crucially on the concept of canonical form .

in this paper , we , the fbk-tr team , describe our system participating in task 3 .

in this paper , we describe our system participating in the task 3 , at semeval 2014 .

sentiment analysis is the process of identifying and extracting subjective information using natural language processing ( nlp ) .

sentiment analysis is the task in natural language processing ( nlp ) that deals with classifying opinions according to the polarity of the sentiment they express .

to build a corpus for robocup , 300 pieces of coach advice were randomly selected from the log files of the 2003 robocup coach competition , which were manually translated into english .

for clang , 300 instructions were randomly selected from the log files of the 2003 robocup coach competition and manually translated into english .

in this paper , we apply kernel methods , which enable an efficient comparison of structures .

in this paper , we have engineered and studied several models for relation learning .

zarrie脽 and kuhn argue that multiword expressions can be reliably detected in parallel corpora by using dependency-parsed , word-aligned sentences .

on the other hand , zarrie脽 and kuhn make use of translational correspondences when identifying multiword expressions .

although word embeddings have been successfully employed in many nlp tasks , the application of word embeddings in re is very recent .

word embeddings have become increasingly popular lately , proving to be valuable as a source of features in a broad range of nlp tasks .

heilman et al extended this approach and worked towards retrieving relevant reading materials for language learners in the reap 3 project .

heilman et al combined a language modeling approach with grammarbased features to improve readability assessment for first and second language texts .

we use the ontonotes datasets from the conll 2011 shared task 6 , only for training the out-of-the-box system .

throughout this work , we use the datasets from the conll 2011 shared task 2 , which is derived from the ontonotes corpus .

we propose novel linear associative units ( lau ) to reduce the gradient propagation length inside the recurrent unit .

we propose a linear associative unit ( lau ) which makes a fusion of both linear and nonlinear transformation inside the recurrent unit .

neural networks have recently gained much attention as a way of inducing word vectors .

more recently , neural networks have become prominent in word representation learning .

the language model component uses the srilm lattice-tool for weight assignment and nbest decoding .

uedin has used the srilm toolkit to train the language model and relies on kenlm for language model scoring during decoding .

multi-task learning using a related auxiliary task can lead to stronger generalization and better regularized models .

multi-task learning can integrate different objectives into one model and has previously been shown to help improve model generalisation .

this is a gui-enabled convenience tool that manages datasets and uses the python-based scikitlearn machine learning toolkit .

kindred is a python package that builds upon the stanford corenlp framework and the scikit-learn machine learning library .

methods make use of the information from only one language side .

the existing methods use only the information in either language side .

tests show that using a situated model significantly improves performances over traditional language modeling methods .

results indicate that integration of situational context dramatically improves performance over traditional methods alone .

we used kenlm with srilm to train a 5-gram language model based on all available target language training data .

we used the srilm toolkit to train a 4-gram language model on the english side of the training corpus .

we propose a robust nonparanormal model to learn the stochastic dependencies among the image , the candidate descriptions , and the popular votes .

we propose a robust nonparanormal approach ( cite-p-23-3-18 ) to model the multimodal stochastic dependencies among images , text , and votes .

ideally the third item can be estimated by the forward-backward algorithm recursively for the firstorder or second-order hmms .

ideally , it can be estimated by using the forward-backward algorithm recursively for the first-order or second-order hmms .

we use the mallet implementation of conditional random fields .

our system is based on the conditional random field .

specifically , we use the liblinear svm package as it is well-suited to text classification tasks with large numbers of features and texts .

in particular , we use the liblinear 3 package which has been shown to be efficient for text classification problems such as this .

semantic role labeling ( srl ) is a kind of shallow semantic parsing task and its goal is to recognize some related phrases and assign a joint structure ( who did what to whom , when , where , why , how ) to each predicate of a sentence ( cite-p-24-3-4 ) .

semantic role labeling ( srl ) is the process of extracting simple event structures , i.e. , “ who ” did “ what ” to “ whom ” , “ when ” and “ where ” .

for implicit discourse relation recognition , previous works attempted to automatically generate training data by removing explicit discourse connectives from sentences .

due to the lack of benchmark data for implicit discourse relation analysis , earlier work used unlabeled data to generate synthetic implicit discourse data .

for english , we use the stanford parser for both pos tagging and cfg parsing .

we use the stanford parser for obtaining all syntactic information .

to get a dictionary of word embeddings , we use the word2vec tool 2 and train it on the chinese gigaword corpus .

for word-level embeddings , we pre-train the word vectors using word2vec on the gigaword corpus mentioned in section 4 , and the text of the training dataset .

using the deep learning framework caffe , we extracted image embeddings from a deep convolutional neural network that was trained on the imagenet classification task .

we extract the 4096-dimensional pre-softmax layer from a for-ward pass through a convolutional neural network , which has been pretrained on the imagenet classification task using caffe .

park and levy proposed a language-modeling approach to whole sentence error correction but their model is not competitive with individually trained models .

park and levy proposed an em-based unsupervised approach to perform whole sentence grammar correction , but the types of errors must be predetermined to learn the parameters for their noisy channel model .

the dimensionality of our word embedding layer was set to size 300 , and we use publicly available pre-trained glove word embeddings that we finetune during training .

since it is operated on the word level , we use pre-trained 300-dimensional glove embeddings and keep them fixed during training .

a pun is a form of wordplay in which a word suggests two or more meanings by exploiting polysemy , homonymy , or phonological similarity to another word , for an intended humorous or rhetorical effect ( cite-p-15-3-1 ) .

a pun is a form of wordplay in which one signifier ( e.g. , a word or phrase ) suggests two or more meanings by exploiting polysemy , or phonological similarity to another signifier , for an intended humorous or rhetorical effect .

we use the skll and scikit-learn toolkits .

we used the svm implementation of scikit learn .

segmentation is the first step in a discourse parser , a system that constructs discourse trees from elementary discourse units .

segmentation is the task of dividing a stream of data ( text or other media ) into coherent units .

the probabilistic language model is constructed on google web 1t 5-gram corpus by using the srilm toolkit .

all feature models are estimated in the in-domain corpus with standard techniques .

to test whether a performance difference is statistically significant , we conduct significance tests following the paired bootstrap approach .

to compare the relative quality of different metrics , we apply bootstrapping re-sampling on the data , and then use paired t-test to determine the statistical significance of the correlation differences .

previous methods have used the first or second order co-occurrence , parts of speech , and local collocations .

previous methods have used first or second order co-occurrences , parts of speech , and grammatical relations .

by an unsupervised one , we may raise the question as to whether the end of supervised nlp comes in sight .

we will raise the question whether the end of supervised parsing is in sight .

we used the 300-dimensional glove word embeddings learned from 840 billion tokens in the web crawl data , as general word embeddings .

we use the pre-trained glove 50-dimensional word embeddings to represent words found in the glove dataset .

in this proposal , we propose a corpus based study to examine doctor-patient conversation of antibiotic treatment negotiation .

in this proposal , we propose a corpus-based study of doctor-patient conversations of antibiotic treatment negotiation in pediatric consultations .

our model is effective in dealing with negation phrases ( a typical case of sentiment expressed by sequence ) .

an interesting case study on negation expression processing shows a promising potential of the architecture dealing with complex sentiment phrases .

sentiment analysis is a recent attempt to deal with evaluative aspects of text .

sentiment analysis is the process of identifying and extracting subjective information using natural language processing ( nlp ) .

tiedemann propose cache-based language and translation models , which are built on recently translated sentences .

tiedemann proposed a cache-model to enforce consistent translation of phrases across the document .

as well as temporal constraints , we are able to extract text-to-text .

we are able to achieve significantly better results than with a text-to-words wtmf model .

this paper has described an unsupervised method for inducing semantic frames from instances of each verb .

this paper presents a method for automatically building verb-specific semantic frames from a large raw corpus .

the results of automatic evaluation and manual assessment of title quality show that the output of our system is consistently ranked higher than that of non-hierarchical baselines .

the results of automatic evaluation and manual assessment confirm the benefits of this design : our system is consistently ranked higher than non-hierarchical baselines .

we parse each document using stanford corenlp in order to acquire both dependency , named entity , and coreference resolution features .

for these data , we preprocess the text including using stanford corenlp to split the review documents into sentences and tokenizing all words .

in addition , we use an english corpus of roughly 227 million words to build a target-side 5-gram language model with srilm in combination with kenlm .

to rerank the candidate texts , we used a 5-gram language model trained on the europarl corpus using kenlm .

more recently , watanabe et al and chiang et al presented a learning algorithm using the mira technique .

recently , watanabe et al and chiang et al have developed tuning methods using the mira algorithm as a nucleus .

in this paper , we propose an approach for identifying curatable articles .

in this paper , we demonstrate how our system is constructed .

the log-lineal combination weights were optimized using mert .

the decoding weights were optimized with minimum error rate training .

and the task includes resolving not just a certain type of noun phrase ( e . g . , pronouns ) .

the approach learns from a small , annotated corpus and the task includes resolving not just pronouns but general noun phrases .

keyphrases are useful for a variety of tasks such as summarization , information retrieval and document clustering .

keyphrases are useful in many tasks such as information retrieval , document summarization or document clustering .

mada-arz is an egyptian arabic extension of the morphological analysis and disambiguation of arabic tool .

mada-arz is an egyptian arabic extension of the morphological analysis and disambiguation of arabic .

from the business language testing service ( bulats ) , the proposed approach is found to outperform gps and dnns with mcd in uncertainty-based rejection .

this method outperforms gps and monte-carlo dropout in uncertainty based rejection for automatic assessment .

it includes a top-level ontology developed following the procedure outlined by russell and norvig and originally covered the tourism domain encoding knowledge about sights , historical persons and buildings .

it includes a top-level developed following the procedure outlined by russell and norvig and originally covered the tourism domain encoding knowledge about sights , historical persons and buildings .

part-of-speech tagging is the assignment of syntactic categories ( tags ) to words that occur in the processed text .

part-of-speech tagging is a key process for various tasks such as ` information extraction , text-to-speech synthesis , word sense disambiguation and machine translation .

for all experiments , we used a 4-gram language model with modified kneser-ney smoothing which was trained with the srilm toolkit .

for the language model , we used sri language modeling toolkit to train a trigram model with modified kneser-ney smoothing on the 31 , 149 english sentences .

the level of the agreement was then assessed using the kappa statistic .

the interannotator agreement was measured using the kappa statistic .

negation is a linguistic phenomenon that can alter the meaning of a textual segment .

negation is a complex phenomenon present in all human languages , allowing for the uniquely human capacities of denial , contradiction , misrepresentation , lying , and irony ( horn and wansing , 2015 ) .

wizard could also tell when the query results did not contain the requested title .

the most successful wizard could also tell when the query results did not contain the requested title .

with an empirical evaluation on the penn discourse treebank ( pdtb ) ( cite-p-11-3-7 ) dataset , which yields an f 1 score of 0 . 485 .

our approach achieves an f 1 score of 0.485 on the implicit relation labeling task for the penn discourse treebank .

in portantly , this type of evaluation can measure how well , fourth darpa speech and natural language .

in portantly , this type of evaluation can measure how well , fourth darpa speech and natural language worka structural theory of text can perform .

the tilde system is a moses phrase-based smt system that was trained on the tilde mt platform .

the smt system is a standard phrase-based system that was trained on the tilde mt platform with moses .

we implemented all models in python using the pytorch deep learning library .

we used pytorch to implement the embedding model and gurobi as our ilp solver .

the berkeley framenet project is an ongoing effort of building a semantic lexicon for english based on the theory of frame semantics .

the berkeley framenet is an ongoing project for building a large lexical resource for english with expert annotations based on frame semantics .

jacy is a hand-crafted japanese hpsg grammar that provides semantic information as well as linguistically motivated analysis of complex constructions .

jacy is a type of hand-crafted japanese grammar based on hpsg that can compute a detailed semantic representation .

the penn discourse treebank corpus is the best-known resource for obtaining english connectives .

the penn discourse treebank is the largest available discourseannotated resource in english .

they used stanford parser to create the parse trees for all sentences .

the stanford parser was used to generate the dependency parse information for each sentence .

gram language models are trained over the target-side of the training data , using srilm with modified kneser-ney discounting .

a 5-gram language model with kneser-ney smoothing is trained using s-rilm on the target language .

vector based word representation has powerful capability that captures the phenomenon that words having the similar meanings should appear together .

the word embeddings can provide word vector representation that captures semantic and syntactic information of words .

the annotation scheme leans on the universal stanford dependencies complemented with the google universal pos tagset and the interset interlingua for morphological tagsets .

the annotation scheme is based on an evolution of stanford dependencies , google universal part-ofspeech tags , and the interset interlingua for morphosyntactic tagsets .

to train our reranking models we used svm-light-tk 7 , which encodes structural kernels in svmlight solver .

to train our models , we adopted svm-light-tk 7 , which enables the use of structural kernels in svm-light , with default parameters .

collobert et al adjust the feature embeddings according to the specific task in a deep neural network architecture .

collobert et al apply generic neural network architectures to several sequence labelling tasks and obtain competitive results despite of the task-specific variations .

klementiev et al treat the task as a multi-task learning problem where each task corresponds to a single word , and task relatedness is derived from co-occurrence statistics in bilingual parallel data .

klementiev et al treated the task as a multi-task learning problem where each task corresponds to a single word , and the task relatedness is derived from cooccurrence statistics in bilingual parallel corpora .

later , miwa and bansal have implemented an end-to-end neural network to construct a context representation for joint entity and relation extraction .

miwa and bansal adopted a bidirectional tree lstm model to jointly extract named entities and relations under a dependency tree structure .

word sense disambiguation ( wsd ) is a difficult natural language processing task which requires that for every content word ( noun , adjective , verb or adverb ) the appropriate meaning is automatically selected from the available sense inventory 1 .

word sense disambiguation ( wsd ) is a widely studied task in natural language processing : given a word and its context , assign the correct sense of the word based on a predefined sense inventory ( cite-p-15-3-4 ) .

for language models , we use the srilm linear interpolation feature .

in the case of the trigram model , we expand the lattice with the aid of the srilm toolkit .

we successfully apply the attention scheme to detect word senses and learn representations according to contexts with the favor of the sememe annotation .

we also analyze several cases in wsd and wrl , which confirms our models are capable of selecting appropriate word senses with the favor of sememe attention .

word sense disambiguation is the task of determining the particular sense of a word from a given set of pre-defined senses .

word sense disambiguation is the task of computationally determining the meaning of a word in its context .

our base model is a transition-based neural parser of chen and manning .

our parsing model is built based on the work of chen and manning .

we use the pre-trained glove 50-dimensional word embeddings to represent words found in the glove dataset .

as input to the aforementioned model , we are going to use dense representations , and more specifically pre-trained word embeddings , such as glove .

we used the meetings from the icsi meeting data , which are recordings of naturally occurring meetings .

we used the icsi meeting corpus , which contains naturally occurring meetings , each about an hour long .

at a large scope , they facilitate tremendously the costly but unavoidable process of semi-automatic lexical acquisition .

nevertheless , large-scope lrs are justified because they facilitate the unavoidable process of large-scale semi-automatic lexical acquisition .

semantic role labeling ( srl ) is the task of labeling predicate-argument structure in sentences with shallow semantic information .

semantic role labeling ( srl ) is the process of extracting simple event structures , i.e. , “ who ” did “ what ” to “ whom ” , “ when ” and “ where ” .

we trained a trigram language model on the chinese side , with the srilm toolkit , using the modified kneser-ney smoothing option .

we used srilm to build a 4-gram language model with interpolated kneser-ney discounting .

our system is based on the phrase-based part of the statistical machine translation system moses .

our translation system is an in-house phrasebased system analogous to moses .

the discursive relations used in this work came from the penn discourse treebank .

they had shown that the penn discourse treebank style discourse relations are useful .

we employed the machine learning tool of scikit-learn 3 , for training the classifier .

for the feature-based system we used logistic regression classifier from the scikit-learn library .

the smt systems were built using the moses toolkit .

we use the popular moses toolkit to build the smt system .

socher et al , 2012 , uses a recursive neural network in relation extraction , and further use lstm .

socher et al train a composition function using a neural network-however their method requires annotated data .