sentence1
stringlengths
16
446
sentence2
stringlengths
14
436
for all methods , the tweets were tokenized with the cmu twitter nlp tool .
the tweets were tokenized and part-ofspeech tagged with the cmu ark twitter nlp tool and stanford corenlp .
it was shown by nederhof et al that prefix probabilities can also be effectively computed for probabilistic tree adjoining grammars .
nederhof et al , for instance , show that prefix probabilities , and therefore surprisal , can be estimated from tree adjoining grammars .
first , kikuchi et al proposed a new long short-term memory network to control the length of the sentence generated by an encoder-decoder model in a text summarization task .
first , kikuchi et al tried to control the length of the sentence generated by an encoder-decoder model in a text summarization task .
with word confusion networks further improves performance .
the complexity is dominated by the word confusion network construction and parsing .
fofe can model the word order in a sequence based on a simple ordinally-forgetting mechanism , which uses the position of each word in the sequence .
fofe can model the word order in a sequence using a simple ordinally-forgetting mechanism according to the positions of words .
we ’ ve demonstrated that the benefits of unsupervised multilingual learning increase steadily with the number of available languages .
we found that performance improves steadily as the number of available languages increases .
dependency parsing consists of finding the structure of a sentence as expressed by a set of directed links ( dependencies ) between words .
dependency parsing is a way of structurally analyzing a sentence from the viewpoint of modification .
for each task , we provide separate training , development , and test datasets for english , arabic , and spanish tweets .
for each task , we provided training , development , and test datasets for english , arabic , and spanish tweets .
a 3-gram language model was trained from the target side of the training data for chinese and arabic , using the srilm toolkit .
a 5-gram language model with kneser-ney smoothing was trained with srilm on monolingual english data .
c . ~ = { ( subj , 0 ) , < n , 0 ) , < v , 0 ) , < comp , 0 ) , ( bar , 0 ) , and a type 1feature successor to the feature system and . . . < agr , 1 ) , < slash , 1 ) } .
we add a type 0 feature 0e ( with p ( 0e ) = { 0 } ) c. ~= { ( subj , 0 ) , < n , 0 ) , < v , 0 ) , < comp,0 ) , ( bar , 0 ) , and a type 1feature successor to the feature system and ... < agr , 1 ) , < slash , 1 ) } use this to build the set of indices .
shared task is a new approach to time normalization based on the semantically compositional annotation of time expressions .
the parsing time normalization task is the first effort to extend time normalization to richer and more complex time expressions .
we derive 100-dimensional word vectors using word2vec skip-gram model trained over the domain corpus .
we use the word2vec cbow model with a window size of 5 and a minimum frequency of 5 to generate 200-dimensional vectors .
syntactic language models can become intolerantly slow to train .
in contrast , syntactic language models can be much slower to train due to rich features .
the learning rule was adam with default tensorflow parameters .
the learning rule was adam with standard parameters .
we embed all words and characters into low-dimensional real-value vectors which can be learned by language model .
we derive 100-dimensional word vectors using word2vec skip-gram model trained over the domain corpus .
semantic knowledge ( e . g . word-senses ) has been defined at the ibm scientific center .
semantic knowledge is represented in a very detailed form ( word_sense pragmatics ) .
we used the target side of the parallel corpus and the srilm toolkit to train a 5-gram language model .
we use sri language model toolkit to train a 5-gram model with modified kneser-ney smoothing on the target-side training corpus .
to obtain this , we used mcut proposed by ding et al which is a type of spectral clustering .
to obtain this , we perform min-max cut proposed by ding et al , which is a spectral clustering method .
part-of-speech tagging is a crucial preliminary process in many natural language processing applications .
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 .
information extraction ( ie ) is a main nlp aspects for analyzing scientific papers , which includes named entity recognition ( ner ) and relation extraction ( re ) .
information extraction ( ie ) is the process of finding relevant entities and their relationships within textual documents .
all systems are evaluated using case-insensitive bleu .
we adopted the case-insensitive bleu-4 as the evaluation metric .
automatic image captioning is a fundamental task that couples visual and linguistic learning .
automatic image captioning is a much studied topic in both the natural language processing ( nlp ) and computer vision ( cv ) areas of research .
in particular , the recent shared tasks of conll 2008 tackled joint parsing of syntactic and semantic dependencies .
the recent conll shared tasks have been focusing on semantic dependency parsing along with the traditional syntactic dependency parsing .
conditional random fields are undirected graphical models trained to maximize the conditional probability of the desired outputs given the corresponding inputs .
conditional random fields are discriminatively-trained undirected graphical models that find the globally optimal labeling for a given configuration of random variables .
additionally , a back-off 2-gram model with goodturing discounting and no lexical classes was built from the same training data , using the srilm toolkit .
a 5-gram language model was created with the sri language modeling toolkit and trained using the gigaword corpus and english sentences from the parallel data .
johnson and charniak proposed a tag-based noisy channel model , which showed great improvement over a boosting-based classifier .
johnson and charniak , 2004 ) proposed a tag-based noisy channel model for disfluency detection .
this is also in line with what has been previously observed in that a person may express the same stance towards a target by using negative or positive language .
as previously reported in , a person may express the same stance towards a target by using negative or positive language .
relation extraction ( re ) is a task of identifying typed relations between known entity mentions in a sentence .
relation extraction is a fundamental task in information extraction .
semantic difference is a ternary relation between two concepts ( apple , banana ) and a discriminative attribute ( red ) that characterizes the first concept but not the other .
semantic difference is a ternary relation between two concepts ( apple , banana ) and a discriminative feature ( red ) that characterizes the first concept but not the other .
ding and palmer propose a syntax-based translation model based on a probabilistic synchronous dependency insert grammar , a version of synchronous grammars defined on dependency trees .
ding and palmer introduce the notion of a synchronous dependency insertion grammar as a tree substitution grammar defined on dependency trees .
sentiment classification is a very domain-specific problem ; training a classifier using the data from one domain may fail when testing against data from another .
sentiment classification is a task to predict a sentiment label , such as positive/negative , for a given text and has been applied to many domains such as movie/product reviews , customer surveys , news comments , and social media .
bansal et al show the benefits of such modified-context embeddings in dependency parsing task .
bansal et al show that deps context is preferable to linear context on parsing task .
they have been useful as features in many nlp tasks .
others have found them useful in parsing and other tasks .
for example , faruqui and dyer use canonical component analysis to align the two embedding spaces .
more concretely , faruqui and dyer use canonical correlation analysis to project the word embeddings in both languages to a shared vector space .
the log-lineal combination weights were optimized using mert .
the minimum error rate training was used to tune the feature weights .
we train a secondorder crf model using marmot , an efficient higher-order crf implementation .
we model the sequence of morphological tags using marmot , a pruned higher-order crf .
word alignment is the process of identifying wordto-word links between parallel sentences .
word alignment is a fundamental problem in statistical machine translation .
sentiment analysis is a natural language processing ( nlp ) task ( cite-p-10-3-0 ) which aims at classifying documents according to the opinion expressed about a given subject ( federici and dragoni , 2016a , b ) .
sentiment analysis is a much-researched area that deals with identification of positive , negative and neutral opinions in text .
we use pre-trained glove embeddings to represent the words .
we use pre-trained vectors from glove for word-level embeddings .
so in most cases of irony , such features will be useful for detection .
given much of the irony in tweets is sarcasm , looking at some of these features may be useful .
that considers a word type and its allowed pos tags as a primary element of the model .
in this work , we take a more direct approach and treat a word type and its allowed pos tags as a primary element of the model .
we use wordsim-353 , which contains 353 english word pairs with human similarity ratings .
specifically , we used wordsim353 , a benchmark dataset , consisting of relatedness judgments for 353 word pairs .
mccarthy instead compares two semantic profiles in wordnet that contain the concepts corresponding to the nouns from the two argument positions .
in contrast to comparing head nouns directly , mccarthy instead compares the selectional preferences for each of the two slots .
the 50-dimensional pre-trained word embeddings are provided by glove , which are fixed during our model training .
we use the glove pre-trained word embeddings for the vectors of the content words .
mann and yarowsky use semantic information that is extracted from documents to inform a hierarchical agglomerative clustering algorithm .
mann and yarowsky used semantic information extracted from documents referring to the target person in an hierarchical agglomerative clustering algorithm .
twitter is a very popular micro blogging site .
twitter is a well-known social network service that allows users to post short 140 character status update which is called β€œ tweet ” .
the feature weights are tuned to optimize bleu using the minimum error rate training algorithm .
the parameter weights are optimized with minimum error rate training .
in this paper , we propose a forest-based tree sequence to string model , which is designed to integrate the strengths of the forest-based and the tree .
to integrate their strengths , in this paper , we propose a forest-based tree sequence to string translation model .
transliteration is a subtask in ne translation , which translates nes based on the phonetic similarity .
transliteration is often defined as phonetic translation ( cite-p-21-3-2 ) .
in this paper , we discuss methods for automatically creating models of dialog structure .
in future work , we will assess the performance of dialog structure prediction on recognized speech .
as a statistical significance test , we used bootstrap resampling .
we used bleu as our evaluation criteria and the bootstrapping method for significance testing .
we use the sri language model toolkit to train a 5-gram model with modified kneser-ney smoothing on the target-side training corpus .
we have used the srilm with kneser-ney smoothing for training a language model for the first stage of decoding .
the words in the document , question and answer are represented using pre-trained word embeddings .
the word embeddings are identified using the standard glove representations .
relation extraction is the task of detecting and classifying relationships between two entities from text .
relation extraction is a fundamental task in information extraction .
kobayashi et al identified opinion relations by searching for useful syntactic contextual clues .
kobayashi et al adopted a supervised learning technique to search for useful syntactic patterns as contextual clues .
neural models have shown great success on a variety of tasks , including machine translation , image caption generation , and language modeling .
various models for learning word embeddings have been proposed , including neural net language models and spectral models .
morphological disambiguation is the process of assigning one set of morphological features to each individual word in a text .
morphological disambiguation is the task of selecting the correct morphological parse for a given word in a given context .
case-insensitive bleu4 was used as the evaluation metric .
all systems are evaluated using case-insensitive bleu .
evaluation results show that our model clearly outperforms a number of baseline models in terms of both clustering posts .
the results show that our model can clearly outperform the baselines in terms of three evaluation metrics .
modified kneser-ney trigram models are trained using srilm upon the chinese portion of the training data .
gram language models are trained over the target-side of the training data , using srilm with modified kneser-ney discounting .
there are techniques for analyzing agreement when annotations involve segment boundaries , but our focus in this article is on words .
there are techniques for analyzing agreement when annotations involve segment boundaries , but our focus in this paper is on words .
for the tree-based system , we applied a 4-gram language model with kneserney smoothing using srilm toolkit trained on the whole monolingual corpus .
further , we apply a 4-gram language model trained with the srilm toolkit on the target side of the training corpus .
to reduce overfitting , we apply the dropout method to regularize our model .
to mitigate overfitting , we apply the dropout method to the inputs and outputs of the network .
we train a kn-smoothed 5-gram language model on the target side of the parallel training data with srilm .
for language model , we use a trigram language model trained with the srilm toolkit on the english side of the training corpus .
twitter is the medium where people post real time messages to discuss on the different topics , and express their sentiments .
twitter is a rich resource for information about everyday events – people post their tweets to twitter publicly in real-time as they conduct their activities throughout the day , resulting in a significant amount of mundane information about common events .
the neural embeddings were created using the word2vec software 3 accompanying .
those models were trained using word2vec skip-gram and cbow .
in this paper , we investigate unsupervised learning of field segmentation models .
in this work , we have examined the task of learning field segmentation models using unsupervised learning .
neural machine translation is currently the state-of-the art paradigm for machine translation .
neural machine translation has recently become the dominant approach to machine translation .
we use srilm toolkit to build a 5-gram language model with modified kneser-ney smoothing .
these language models were built up to an order of 5 with kneser-ney smoothing using the srilm toolkit .
the constituent context model for inducing constituency parses was the first unsupervised approach to surpass a right-branching baseline .
the constituent-context model is the first unsupervised constituency grammar induction system that achieves better performance than the trivial right branching baseline for english .
in this paper , we propose a procedure to train multi-domain , recurrent neural network-based ( rnn ) language generators via multiple adaptation .
the paper presents an incremental recipe for training multi-domain language generators based on a purely data-driven , rnn-based generation model .
we use pre-trained word2vec word vectors and vector representations by tilk et al to obtain word-level similarity information .
we also used word2vec to generate dense word vectors for all word types in our learning corpus .
the stochastic gradient descent with back-propagation is performed using adadelta update rule .
training is done through stochastic gradient descent over shuffled mini-batches with adadelta update rule .
in the n-coalescent , every pair of lineages merges independently with rate 1 , with parents chosen uniformly at random from the set of possible parents .
in the n-coalescent , every pair of lineages merges independently with rate 1 , with parents chosen uniformly at random from the set of possible parents at the previous time step .
in our approach is to allow highly flexible reordering operations , in combination with a discriminative model that can condition on rich features of the source-language input .
a critical difference in our work is to allow arbitrary reorderings of the source language sentence ( as in phrase-based systems ) , through the use of flexible parsing operations .
we measure the translation quality using a single reference bleu .
we evaluated the translation quality of the system using the bleu metric .
we use srilm toolkit to build a 5-gram language model with modified kneser-ney smoothing .
a trigram language model with modified kneser-ney discounting and interpolation was used as produced by the srilm toolkit .
sentence compression is a standard nlp task where the goal is to generate a shorter paraphrase of a sentence .
sentence compression is the task of compressing long , verbose sentences into short , concise ones .
wu presents a better-constrained grammar designed to only produce tail-recursive parses .
wu proposes a bilingual segmentation grammar extending the terminal rules by including phrase pairs .
although coreference resolution is a subproblem of natural language understanding , coreference resolution evaluation metrics have predominately been discussed in terms of abstract entities and hypothetical system errors .
coreference resolution is the process of determining whether two expressions in natural language refer to the same entity in the world .
we trained a tri-gram hindi word language model with the srilm tool .
we used the srilm toolkit to generate the scores with no smoothing .
sentiment analysis is a research area where does a computational analysis of people ’ s feelings or beliefs expressed in texts such as emotions , opinions , attitudes , appraisals , etc . ( cite-p-12-1-3 ) .
sentiment analysis is the task of automatically identifying the valence or polarity of a piece of text .
we adopt two standard metrics rouge and bleu for evaluation .
for the evaluation of the results we use the bleu score .
in this paper , we focus on designing a review generation model that is able to leverage both user and item information .
in this paper , we focus on the problem of building assistive systems that can help users to write reviews .
the syntactic feature set is extracted after dependency parsing using the maltparser .
all data is automatically annotated with syntactic tags using maltparser .
we used a bitext projection technique to transfer dependency-based opinion frames .
we propose a cross-lingual framework for fine-grained opinion mining using bitext projection .
knowledge of our native language provides an initial foundation for second language learning .
our native language ( l1 ) plays an essential role in the process of lexical choice .
semantic roles are approximated by propbank argument roles .
direction , manner , and purpose are propbank adjunctive argument labels .
in this paper , we study the problem of sentiment analysis on product reviews .
in this paper , we propose a novel and effective approach to sentiment analysis on product reviews .
in this paper we present an algorithmic framework which allows an automated acquisition of map-like information from the web , based on surface patterns .
in this paper we utilize a pattern-based lexical acquisition framework for the discovery of geographical information .
circles denote events , squares denote arguments , solid arrows represent event-event relations , and dashed arrows represent event-argument relations .
the circles denote fixations , and the lines are saccades .
semantic parsing is the task of mapping natural language sentences to a formal representation of meaning .
semantic parsing is the task of translating natural language utterances into a machine-interpretable meaning representation .
each context consists of approximately a paragraph of surrounding text , where the word to be discriminated ( the target word ) is found approximately in the middle of the context .
1 a context consists of all the patterns of n-grams within a certain window around the corresponding entity mention .
we used the pre-trained google embedding to initialize the word embedding matrix .
in this baseline , we applied the word embedding trained by skipgram on wiki2014 .
the word embeddings used in each neural network is initialized with the pre-trained glove with the dimension of 300 .
the word embeddings are initialized using the pre-trained glove , and the embedding size is 300 .
in spite of this broad attention , the open ie task definition has been lacking .
in spite of this wide attention , open ie ’ s formal definition is lacking .
neural network models have been exploited to learn dense feature representation for a variety of nlp tasks .
interestingly convolutional neural networks , widely used for image processing , have recently emerged as a strong class of models for nlp tasks .
reordering is a difficult task in translating between widely different languages such as japanese and english .
reordering is a common problem observed in language pairs of distant language origins .
we initialize our word vectors with 300-dimensional word2vec word embeddings .
our cdsm feature is based on word vectors derived using a skip-gram model .
we define a conditional random field for this task .
our model is a first order linear chain conditional random field .