text
stringlengths
82
736
label
int64
0
1
we present a general framework for comparing multiple groups of documents---in this work , we present a general framework to perform such comparisons
1
the characters themselves are often composed of subcharacter components which are also semantically informative---characters are often composed of subcharacter components which are also semantically informative
1
through comparative experiments , we show that emotion recognition can be performed using either textual or musical features , and that the joint use of lyrics and music can improve significantly over classifiers that use only one dimension at a time---on the dataset of 100 songs , we showed that emotion recognition ca...
1
for our logistic regression classifier we use the implementation included in the scikit-learn toolkit 2---we use the logistic regression classifier in the skll package , which is based on scikit-learn , optimizing for f 1 score
1
triviaqa , which has wikipedia entities as answers , makes it possible to leverage structured kbs like freebase , which we leave to future work---as answers , makes it possible to leverage structured kbs like freebase , which we leave to future work
1
for the contextual polarity disambiguation subtask , we described a very efficient and robust method based on a sentiment lexicon associated with a polarity shift detector and a tree based classification---for the contextual polarity disambiguation subtask , covered in section 2 , we use a system that combines a lexico...
1
the target fourgram language model was built with the english part of training data using the sri language modeling toolkit---we presented the first neural network based shift-reduce parsers for ccg
0
we implement the pbsmt system with the moses toolkit---we use an in-house implementation of a pbsmt system similar to moses
1
we discuss an interactive approach to robust interpretation in a large scale speech-to-speech translation system---we discuss rose , an interactive approach to robust interpretation developed in the context of the janus speech-to-speech translation system
1
we trained a standard 5-gram language model with modified kneser-ney smoothing using the kenlm toolkit on 4 billion running words---for tagging , we use the stanford pos tagger package
0
chang and teng extends the work in chang and lai to automatically extract the relations between full-form phrases and their abbreviations , where both the full-form phrase and its abbreviation are not given---chang and teng extends the work in chang and lai to automatically extract the relations between full-form phras...
1
in the second step , we propose a relational adaptive bootstrapping ( rap ) algorithm to expand the seeds in the target domain---in the second step , we propose a novel relational adaptive bootstrapping ( rap ) algorithm to expand the seeds in the target domain by exploiting the labeled source domain
1
coreference resolution is the problem of identifying which mentions ( i.e. , noun phrases ) refer to which real-world entities---coreference resolution is the task of determining which mentions in a text refer to the same entity
1
albrecht and hwa , 2007 ) presented a regression based method for developing automatic evaluation metrics for machine translation systems without directly relying on human reference translations---albrecht and hwa proposed a method to evaluate mt outputs with pseudo references using support vector regression as the lea...
1
we used the svm implementation provided within scikit-learn---we used the svm implementation of scikit learn
1
we used a phrase-based smt model as implemented in the moses toolkit---we use the moses software package 5 to train a pbmt model
1
continuous-valued vector representation of words has been one of the key components in neural architectures for natural language processing---one of the most useful neural network techniques for nlp is the word embedding , which learns vector representations of words
1
ambiguity is the task of building up multiple alternative linguistic structures for a single input ( cite-p-13-1-8 )---we build a model of all unigrams and bigrams in the gigaword corpus using the c-mphr method , srilm , irstlm , and randlm 3 toolkits
0
relation extraction is a traditional information extraction task which aims at detecting and classifying semantic relations between entities in text ( cite-p-10-1-18 )---relation extraction is the task of finding semantic relations between entities from text
1
in 2003 , bengio et al proposed a neural network architecture to train language models which produced word embeddings in the neural network---introduced by bengio et al , the authors proposed a statistical language model based on shallow neural networks
1
we use logistic regression with l2 regularization , implemented using the scikit-learn toolkit---we train and evaluate a l2-regularized logistic regression classifier with the liblin-ear solver as implemented in scikit-learn
1
sentence compression is the task of producing a summary at the sentence level---in section 4 , we show that this result still holds for multimodal ccg
0
in this run , we use a sentence vector derived from word embeddings obtained from word2vec---named entity disambiguation ( ned ) is the task of resolving ambiguous mentions of entities to their referent entities in a knowledge base ( kb ) ( e.g. , wikipedia )
0
we adapted the moses phrase-based decoder to translate word lattices---the language model is trained and applied with the srilm toolkit
0
stance detection has been defined as automatically detecting whether the author of a piece of text is in favor of the given target or against it---we propose a reinforcement learning based approach that integrates target information and generates target-specific tree structures
0
moreover , as we will show in experiment section , a preprocessing method does not work well when only source information is available---however , as we will show below , existing smt systems do not deal well with the measure word generation in general due to data
1
our baseline system is phrase-based moses with feature weights trained using mert---we employ widely used and standard machine translation tool moses to train the phrasebased smt system
1
"socher et al used recursive neural networks to model sentences for different tasks , including paraphrase detection and sentence classification---socher et al utilized parsing to model the hierarchical structure of sentences and uses unfolding recursive autoencoders to learn representations for single words and phra...
1
krause generalized the work by khuller et alon budgeted maximum cover problem to the submodular framework , and showed a 1 2 -approximation algorithm---khuller et al studied the maximum coverage problem with a knapsack constraint , and proved that the greedy algorithm achieves -approximation
1
melamud et al use word embeddings generated using the word2vec skip-gram model---in 2013 , mikolov et al generated phrase representation using the same method used for word representation in word2vec
1
syntactic analysis for syntactic features , we trained an arabic dependency parser using maltparser on the columbia arabic treebank version of the patb ,---to avoid this problem , tromble et al propose linear bleu , an approximation to the bleu score to efficiently perform mbr decoding when the search space is represen...
0
wordnet is a large semantic lexicon database of english words , where nouns , verbs , adjectives and adverbs are grouped into sets of cognitive synonyms---wordnet is a large lexical database of english , where open class words are grouped into concepts represented by synonyms that are linked to each other by semantic r...
1
our model builds on word2vec , a neural network based language model that learns word embeddings by maximizing the probability of raw text---we use a popular word2vec neural language model to learn the word embeddings on an unsupervised tweet corpus
1
relation extraction is the task of finding semantic relations between two entities from text---relation extraction is the task of finding relationships between two entities from text
1
to speed up training using parallel processing , we use the iterative parameter mixing approach of mcdonald et al , where training data are split into several parts and weight updates are averaged after each pass through the training data---we furthermore use the distributed learning technique of iterative parameter mi...
1
levin has in fact proposed a well-known classification of verbs based on their range of syntactic alternations---levin provides a classification of over 3000 verbs according to their participation in alternations involving np and pp constituents
1
phrase-based translation systems prove to be the stateof-the-art as they have delivered translation performance in recent machine translation evaluations---in recent years , phrase-based systems for statistical machine translation have delivered state-of-the-art performance on standard translation tasks
1
eurowordnet is a multilingual semantic lexicon with wordnets for several european languages , which are structured as the princeton wordnet---eurowordnet is a multilingual lexical knowledge base comprised of hierarchical representations of lexical items for several european languages
1
we use a set of 318 english function words from the scikit-learn package---we use the svm implementation from scikit-learn , which in turn is based on libsvm
0
aw et al and kaufmann and kalita consider normalisation as a machine translation task from lexical variants to standard forms using off-theshelf tools---aw et al , kobus et al viewed the text message normalization as a statistical machine translation process from the texting language to standard english
1
a pun is a means of expression , the essence of which is in the given context the word or phrase can be understood in two meanings simultaneously ( cite-p-22-3-7 )---a pun is a word used in a context to evoke two or more distinct senses for humorous effect
1
we use minimal error rate training to maximize bleu on the complete development data---we used minimum error rate training to optimize the feature weights
1
we use 5-grams for all language models implemented using the srilm toolkit---we train trigram language models on the training set using the sri language modeling tookit
1
pitler and nenkova show that the entity transition features extracted from the entity grid model on its own do not significantly predict human readability ratings---human-annotated image and video descriptions allow us to investigate what types of verb ¨c noun relations are in principle present in the visual data
0
each sentence in the documents is firstly assigned a salience score---a salience score is computed for each phrase by exploiting redundancy of the document content
1
semantic role labeling ( srl ) is a task of automatically identifying semantic relations between predicate and its related arguments in the sentence---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-...
1
relation extraction is the task of detecting and characterizing semantic relations between entities from free text---the language model is trained on the target side of the parallel training corpus using srilm
0
other work extracts hypernym relations from encyclopedias but has limited coverage---other work tries to extract hypernym relations from large-scale encyclopedias like wikipedia and achieves high precision
1
word sense disambiguation ( wsd ) is the problem of assigning a sense to an ambiguous word , using its context---in natural language , a word often assumes different meanings , and the task of determining the correct meaning , or sense , of a word in different contexts is known as word sense disambiguation ( wsd )
1
we presented an approach of using esa for sentiment classification---in this work , we investigated the use of esa for the given task of sentiment analysis
1
our machine translation system is a phrase-based system using the moses toolkit---we use the moses toolkit to train various statistical machine translation systems
1
coreference resolution is a challenging task , that involves identification and clustering of noun phrases mentions that refer to the same real-world entity---coreference resolution is the process of linking together multiple expressions of a given entity
1
we performed paired bootstrap sampling to test the significance in bleu score differences---we performed significance testing using paired bootstrap resampling
1
this sampler sidesteps the intractability issues of previous models which required inference over derivation forests---sampler over synchronous derivation trees can efficiently draw samples from the posterior , overcoming the limitations of previous models
1
the optimisation of the feature weights of the model is done with minimum error rate training against the bleu evaluation metric---the feature weights for the log-linear combination of the features are tuned using minimum error rate training on the devset in terms of bleu
1
both files are concatenated and learned by word2vec---the word embeddings are word2vec of dimension 300 pre-trained on google news
1
for capturing the semantics of words , we again derive features from the pre-trained fasttext word vectors---to alleviate issues with out-of-vocabulary words , we use both character-and subwordbased word embeddings computed with fasttext
1
we then perform training by using an expectation-maximization algorithm that iteratively maximizes fto reach a local optimal solution---for training the trigger-based lexicon model , we apply the expectation-maximization algorithm
1
we used weka to experiment with several classifiers---an lm is trained on 462 million words in english using the srilm toolkit
0
the mt performance is measured with the widely adopted bleu and ter metrics---semantic role labeling ( srl ) is the task of identifying semantic arguments of predicates in text
0
we use the word2vec framework in the gensim implementation to generate the embedding spaces---we pre-train the word embedding via word2vec on the whole dataset
1
the language models were created using the srilm toolkit on the standard training sections of the ccgbank , with sentenceinitial words uncapitalized---the language models are 4-grams with modified kneser-ney smoothing which have been trained with the srilm toolkit
1
we selected target verbs by choosing classes from levin that are expected to undergo the causative alternation---afterwards , user and product information is considered via attentions over different semantic levels
0
supervised systems based on neural networks achieve the most promising results---exploiting neural networks on unlabeled corpora achieve promising results , surpassing this hard baseline
1
the multi-label setting is common and useful in the real world---multi-label text categorization is a common and useful
1
our cdsm feature is based on word vectors derived using a skip-gram model---as embedding vectors , we used the publicly available representations obtained from the word2vec cbow model
1
first , we examine three subproblems that play a role in coreference resolution : named entity recognition , anaphoricity determination , and coreference element detection---our submission to the english-french task was a phrase-based statistical machine translation based on the moses decoder
0
word embeddings are low-dimensional vector representations of words such as word2vec that recently gained much attention in various semantic tasks---word embeddings are distributed representations of words learned on large scale corpus using neural networks
1
by incorporating textual information , rcm can effectively deal with data sparseness problem---we take fully advantage of questions ’ textual descriptions to address data sparseness problem and cold-start problem
1
a semantic parser is learned given a set of sentences and their correct logical forms using smt methods---however , obtaining labeled data is a big challenge in many real-world problems
0
in this paper , we propose to compress neural language models by sparse word representations---we propose an approach to represent uncommon words ¡¯ embeddings by a sparse linear combination of common ones
1
text categorization is the task of classifying documents into a certain number of predefined categories---text categorization is the task of assigning a text document to one of several predefined categories
1
we use the glove word vector representations of dimension 300---for word embeddings , we consider word2vec and glove
1
this approach can be used for word alignment in language pairs like english-hindi---approach has been proposed as an alternative strategy for word alignment
1
coreference resolution is the task of partitioning a set of mentions ( i.e . person , organization and location ) into entities---coreference resolution is the task of partitioning the set of mentions of discourse referents in a text into classes ( or ‘ chains ’ ) corresponding to those referents ( cite-p-12-3-14 )
1
latent dirichlet allocation is one of the widely adopted generative models for topic modeling---for this experiment , we train a standard phrase-based smt system over the entire parallel corpus
0
we use the same evaluation metrics as described in , which is similar to those in---we use the same metrics as described in wu et al , which is similar to those in
1
all the feature weights and the weight for each probability factor are tuned on the development set with minimumerror-rate training---unreliable scores does not result in a reliable one
0
a 4-gram language model was trained on the target side of the parallel data using the srilm toolkit---a trigram language model with modified kneser-ney discounting and interpolation was used as produced by the srilm toolkit
1
this has shown to be effective for numerous nlp tasks as it can capture word morphology and reduce out-of-vocabulary---stance detection is the task of automatically determining from text whether the author of the text is in favor of , against , or neutral towards a proposition or target
0
as a result , bus request cycle may conceivably be understood either as a corn- * when a sequence has length three or more the order of modification may vary---as a result , bus request cycle may conceivably be understood either as a corn- * when a sequence has length three or more
1
for building our statistical ape system , we used maximum phrase length of 7 and a 5-gram language model trained using kenlm---blei et al proposed lda as a general bayesian framework and gave a variational model for learning topics from data
0
in table 6 , we list the rtm test results for tasks and subtasks that predict hter or meteor from qet15 , qet14 , and qet13---we use the logistic regression classifier as implemented in the skll package , which is based on scikitlearn , with f1 optimization
0
using word2vec , we compute word embeddings for our text corpus---then we train word2vec to represent each entity with a 100-dimensional embedding vector
1
that is , since the morphological analysis is the first-step in most nlp applications , the sentences with incorrect word spacing must be corrected for their further processing---we assume that a morphological analysis consists of three processes : tokenization , dictionary lookup , and disambiguation
1
thus , in this paper , we assume that there is a relationship between the canonical word order and the proportion of each word order in a large corpus and present a corpus-based analysis of canonical word order of japanese double object constructions---we have already used our pos-based model to rescore word-graphs , w...
0
mining parallel data from web is a promising method to overcome the knowledge bottleneck faced by machine translation---web mining for parallel data becomes a promising solution to this knowledge acquisition problem
1
results show approximately 6-10 % cer reduction of the acms in comparison with the word trigram models , even when the acms are slightly smaller---experimental results show substantial improvements of the acm in comparison with classical cluster models and word n-gram models
1
mikolov et al proposed a computationally efficient method for learning distributed word representation such that words with similar meanings will map to similar vectors---by an unsupervised one , we may raise the question as to whether the end of supervised nlp comes in sight
0
another stream of work tries to identify domain-specific words to improve crossdomain classification---another line of work tries to derive domain-specific sentiment words
1
goldwasser et al presented a confidence-driven approach to semantic parsing based on self-training---in contrast , goldwasser et al proposed a self-supervised approach , which iteratively chose high-confidence parses to retrain the parser
1
since coreference resolution is a pervasive discourse phenomenon causing performance impediments in current ie systems , we considered a corpus of aligned english and romanian texts to identify coreferring expressions---coreference resolution is the task of clustering a set of mentions in the text such that all mention...
1
in this paper , we show that using well calibrated probabilities to estimate sense priors is important---named entity recognition ( ner ) is the task of identifying and classifying phrases that denote certain types of named entities ( nes ) , such as persons , organizations and locations in news articles , and genes , ...
0
in this paper we described the system submitted for the semeval 2014 task 9 ( sentiment analysis in twitter )---in this paper we describe the system submitted for the semeval 2014 sentiment analysis in twitter task ( task 9
1
it also outperforms related models on similarity tasks and named entity recognition---we use srilm toolkit to train a trigram language model with modified kneser-ney smoothing on the target side of training corpus
0
the accuracy was measured using the bleu score and the string edit distance by comparing the generated sentences with the original sentences---the overall mt system is evaluated both with and without function guessing on 500 held-out sentences , and the quality of the translation is measured using the bleu metric
1
for all experiments , we used a 4-gram language model with modified kneser-ney smoothing which was trained with the srilm toolkit---for all data sets , we trained a 5-gram language model using the sri language modeling toolkit
1
furthermore , we propose a way to generate onthe-fly knowledge in logical inference , by combining our framework with the idea of tree transformation---in practical inference , we combine our framework with the idea of tree transformation ( cite-p-26-1-2 ) , to propose a way of generating knowledge in logical represent...
1
multi-task joint modeling has been shown to effectively improve individual tasks---we present the first approach for applying distant supervision to cross-sentence relation extraction
0
there are numerous theoretical approaches describing is and its semantics and the terminology used is diverse for an overview )---we adopted a novel formulation that models dependency edges in argument paths and jointly predicts them along with events and arguments
0