id
stringlengths
1
4
example_id
stringlengths
1
3
question
stringlengths
12
167
context
listlengths
1
168
answer
stringlengths
0
1.61k
choices
null
question_type
stringclasses
3 values
200
57
Is text-to-image synthesis trained is suppervized or unsuppervized manner?
[ "Text-to-Image Synthesis Taxonomy and Categorization ::: Resolution Enhancement GANs ::: HDGAN\tHierarchically-nested adversarial network (HDGAN) is a method proposed by , and its main objective is to tackle the difficult problem of dealing with photographic images from semantic text descriptions. These semantic te...
unsupervised
null
extractive
201
57
What challenges remain unresolved?
[ "The taxonomy and the review outline the techniques and the evolution of different approaches, and eventually provide a clear roadmap to summarize the list of contemporaneous solutions that utilize GANs and DCNNs to generate enthralling results in categories such as human faces, birds, flowers, room interiors, obje...
give more independence to the several learning methods (e.g. less human intervention) involved in the studies, increasing the size of the output images
null
extractive
202
57
What is the conclusion of comparison of proposed solution?
[ "The taxonomy and the review outline the techniques and the evolution of different approaches, and eventually provide a clear roadmap to summarize the list of contemporaneous solutions that utilize GANs and DCNNs to generate enthralling results in categories such as human faces, birds, flowers, room interiors, obje...
HDGAN produced relatively better visual results on the CUB and Oxford datasets while AttnGAN produced far more impressive results than the rest on the more complex COCO dataset, In terms of inception score (IS), which is the metric that was applied to majority models except DC-GAN, the results in Table TABREF48 show th...
null
extractive
203
57
What is typical GAN architecture for each text-to-image synhesis group?
[ "A Survey and Taxonomy of Adversarial Neural Networks for Text-to-Image Synthesis\tText-to-image synthesis refers to computational methods which translate human written textual descriptions, in the form of keywords or sentences, into images with similar semantic meaning to the text. In earlier research, image synth...
Semantic Enhancement GANs: DC-GANs, MC-GAN Resolution Enhancement GANs: StackGANs, AttnGAN, HDGAN Diversity Enhancement GANs: AC-GAN, TAC-GAN etc. Motion Enhancement GAGs: T2S, T2V, StoryGAN
null
free_form
204
58
Where do they employ feature-wise sigmoid gating?
[ "Combining Character and Word-level Representations\tWe tested three different methods for combining INLINEFORM0 with INLINEFORM1 : simple concatenation, a learned scalar gate , and a learned vector gate (also referred to as feature-wise sigmoidal gate). Additionally, we compared these methods to two baselines: us...
gating mechanism acts upon each dimension of the word and character-level vectors
null
extractive
205
58
Which model architecture do they use to obtain representations?
[ "Obtaining Sentence Representations\tTo enable sentence-level classification we need to obtain a sentence representation from the word vectors INLINEFORM0 . We achieved this by using a BiLSTM with max pooling, which was shown to be a good universal sentence encoding mechanism . Let INLINEFORM0 , be an input senten...
BiLSTM with max pooling
null
extractive
206
58
Which downstream sentence-level tasks do they evaluate on?
[ "Relationship Between Word- and Sentence-level Evaluation Tasks\tIt is clear that the better performance the vector gate had in word similarity tasks did not translate into overall better performance in downstream tasks. This confirms previous findings indicating that intrinsic word evaluation metrics are not good ...
,
null
extractive
207
58
Which similarity datasets do they use?
[ "Gating Mechanisms for Combining Character and Word-level Word Representations: An Empirical Study\tIn this paper we study how different ways of combining character and word-level representations affect the quality of both final word and sentence representations. We provide strong empirical evidence that modeling c...
MEN, MTurk287, MTurk771, RG, RW, SimLex999, SimVerb3500, WS353, WS353R, WS353S
null
extractive
208
59
Are there datasets with relation tuples annotated, how big are datasets available?
[ "Experiments ::: Datasets\tWe focus on the task of extracting multiple tuples with overlapping entities from sentences. We choose the New York Times (NYT) corpus for our experiments. This corpus has multiple versions, and we choose the following two versions as their test dataset has significantly larger number of ...
Yes
null
yes_no
209
59
Which one of two proposed approaches performed better in experiments?
[ "Analysis and Discussion ::: Error Analysis\tThe relation tuples extracted by a joint model can be erroneous for multiple reasons such as: (i) extracted entities are wrong; (ii) extracted relations are wrong; (iii) pairings of entities with relations are wrong. To see the effects of the first two reasons, we analyz...
WordDecoding (WDec) model
null
extractive
210
59
What is previous work authors reffer to?
[ "Effective Modeling of Encoder-Decoder Architecture for Joint Entity and Relation Extraction\tA relation tuple consists of two entities and the relation between them, and often such tuples are found in unstructured text. There may be multiple relation tuples present in a text and they may share one or both entities...
SPTree, Tagging, CopyR, HRL, GraphR, N-gram Attention
null
extractive
211
59
How higher are F1 scores compared to previous work?
[ "This ensemble mechanism increases the precision significantly on both datasets with a small improvement in recall as well. In the ensemble scenario, compared to HRL, WDec achieves $4.2\\%$ and $3.5\\%$ higher F1 scores and PNDec achieves $4.2\\%$ and $2.9\\%$ higher F1 scores on the NYT29 and NYT24 datasets respec...
WordDecoding (WDec) model achieves F1 scores that are $3.9\%$ and $4.1\%$ higher than HRL on the NYT29 and NYT24 datasets respectively, PtrNetDecoding (PNDec) model achieves F1 scores that are $3.0\%$ and $1.3\%$ higher than HRL on the NYT29 and NYT24 datasets respectively
null
extractive
212
60
what were the baselines?
[ "However, we show that combining the domain-specific features with more traditional text-based features does improve the model's scores over simply using the domain-specific features themselves. Interestingly, citation impact and age of the citation are both negatively correlated with rank. We hypothesize that this...
Rank by the number of times a citation is mentioned in the document, Rank by the number of times the citation is cited in the literature (citation impact). , Rank using Google Scholar Related Articles., Rank by the TF*IDF weighted cosine similarity. , ank using a learning-to-rank model trained on text similarity rankin...
null
extractive
213
60
what is the supervised model they developed?
[ "Topic models have been used to distill a document into a set of topics used to form query ( ). generated queries using the top TF*IDF weighted terms in each document. suggested extracting phrasal concepts from a document, which are then used to generate queries. combined query extraction and pseudo-relevance f...
SVMRank
null
extractive
214
60
what is the size of this built corpus?
[ "However, we show that combining the domain-specific features with more traditional text-based features does improve the model's scores over simply using the domain-specific features themselves. Interestingly, citation impact and age of the citation are both negatively correlated with rank. We hypothesize that this...
90 annotated documents with 5 citations each ranked 1 to 5, where 1 is least relevant and 5 is most relevant for a total of 450 annotated citations
null
extractive
215
60
what crowdsourcing platform is used?
[ "Topic models have been used to distill a document into a set of topics used to form query ( ). generated queries using the top TF*IDF weighted terms in each document. suggested extracting phrasal concepts from a document, which are then used to generate queries. combined query extraction and pseudo-relevance f...
asked the authors to rank by closeness five citations we selected from their paper
null
extractive
216
61
Which deep learning model performed better?
[ "Exploiting Deep Learning for Persian Sentiment Analysis\tThe rise of social media is enabling people to freely express their opinions about products and services. The aim of sentiment analysis is to automatically determine subject's sentiment (e.g., positive, negative, or neutral) towards a particular aspect such ...
autoencoders
null
extractive
217
61
What was their performance on the dataset?
[ "Methodology and Experimental Results\tThe novel dataset used in this work was collected manually and includes Persian movie reviews from 2014 to 2016. A subset of dataset was used to train the neural network (60% training dataset) and rest of the data (40%) was used to test and validate the performance of the trai...
accuracy of 82.6%
null
extractive
218
62
Did the authors use crowdsourcing platforms?
[ "Tourist Generation Models\tAfter training the supervised tourist model (conditioned on observations and action from human expert trajectories), there are two ways to train an accompanying guide model. We can optimize a location prediction model on either (i) extracted human trajectories (as in the localization set...
Yes
null
yes_no
219
62
How was the dataset collected?
[ "Data Collection\tWe crowd-sourced the collection of the dataset on Amazon Mechanical Turk (MTurk). We use the MTurk interface of ParlAI to render 360 images via WebGL and dynamically display neighborhood maps with an HTML5 canvas. Detailed task instructions, which were also given to our workers before they starte...
crowd-sourced the collection of the dataset on Amazon Mechanical Turk (MTurk)
null
extractive
220
62
What language do the agents talk in?
[ "The guide has access to a map and knows the target location, but does not know where the tourist is; the tourist has a 360-degree view of the world, but knows neither the target location on the map nor the way to it. The agents need to work together through communication in order to successfully solve the task. An...
English
null
free_form
221
62
What evaluation metrics did the authors look at?
[ "Tourist: crossing the street now Tourist: ACTION:FORWARD ACTION:FORWARD ACTION:TURNLEFT Tourist: there is an I love new york shop across the street on the left from me now Tourist: ACTION:TURNRIGHT ACTION:FORWARD Guide: ok. I'll see if it's right. Guide: EVALUATE_LOCATION Guide: It's not right. Tourist: What shoul...
localization accuracy
null
extractive
222
62
What data did they use?
[ "Data Collection\tWe crowd-sourced the collection of the dataset on Amazon Mechanical Turk (MTurk). We use the MTurk interface of ParlAI to render 360 images via WebGL and dynamically display neighborhood maps with an HTML5 canvas. Detailed task instructions, which were also given to our workers before they starte...
dataset on Mechanical Turk involving human perception, action and communication
null
extractive
223
63
How is the accuracy of the system measured?
[ "In order to test whether these results generalised to our domain, we devised a test that would make use of what little data we had to evaluate. We had no original data on whether sentences were semantically similar, but we did have a corpus of articles clustered into stories. Working on the assumption that similar...
F1 score of 0.71 for this task without any specific training, simply by choosing a threshold below which all sentence pairs are considered duplicates, distances between duplicate and non-duplicate questions using different embedding systems
null
extractive
224
63
How is an incoming claim used to retrieve similar factchecked claims?
[ "Real-time Claim Detection from News Articles and Retrieval of Semantically-Similar Factchecks\tFactchecking has always been a part of the journalistic process. However with newsroom budgets shrinking it is coming under increasing pressure just as the amount of false information circulating is on the rise. We there...
text clustering on the embeddings of texts
null
extractive
225
63
What existing corpus is used for comparison in these experiments?
[ "Such is the speed of new development in the field that since then, several papers describing textual embeddings have been published. Due to the fact that we had already evaluated embeddings for clustering, and therefore knew our system would rely on Google USE Large , we decided to use this instead. We compared t...
Quora duplicate question dataset
null
extractive
226
64
What is the baseline?
[ "Related work ::: Analysis of RC models and datasets\tThere is a large body of work on analyzing the nature of RC datasets, motivated by the question to what degree RC models understand natural language , . Several studies suggest that current RC datasets have unintended bias, which enables RC systems to rely on a ...
path ranking-based KGC (PRKGC)
null
extractive
227
64
What dataset was used in the experiment?
[ "The developed crowdsourcing annotation framework can be used for annotating other QA datasets with derivations. Through an experiment using two baseline models, we highlight several challenges of RC-QED. We will make the corpus of reasoning annotations and the baseline system publicly available at https://naoya-i....
WikiHop
null
extractive
228
64
Did they use any crowdsourcing platform?
[ "The developed crowdsourcing annotation framework can be used for annotating other QA datasets with derivations. Through an experiment using two baseline models, we highlight several challenges of RC-QED. We will make the corpus of reasoning annotations and the baseline system publicly available at https://naoya-i....
Yes
null
yes_no
229
64
How was the dataset annotated?
[ "Related work ::: RC datasets with explanations\tThere exists few RC datasets annotated with explanations (Table TABREF50). The most similar work to ours is Science QA dataset , , , which provides a small set of NLDs annotated for analysis purposes. By developing the scalable crowdsourcing framework, our work provi...
True, Likely (i.e. Answerable), or Unsure (i.e. Unanswerable), why they are unsure from two choices (“Not stated in the article” or “Other”), The “summary” text boxes
null
extractive
230
65
How many label options are there in the multi-label task?
[ "Methodology ::: Machine Learning Models\tWe compare 4 different models for performing our task of sentiment classification. We then pick the best performing model for the task of outcome prediction. Here, we have two categories of algorithms: single-label and multi-label (We already discussed above why it is meani...
two labels
null
extractive
231
65
Who are the experts?
[ "Trends: Here, we analyze certain trends of the debates like the change in sentiments of the users towards the candidates over time (hours, days, months) and how the opinion of experts such as Washington Post affect the sentiments of the users. For the sentiment analysis, we look at our problem in a multi-label set...
political pundits of the Washington Post
null
extractive
232
65
Who is the crowd in these experiments?
[ "Our contributions are mainly: we want to see how predictive the sentiment/opinion of the users are in social media microblogs and how it compares to that of the experts. In essence, we find that the crowd wisdom in the microblog domain matches that of the experts in most cases. There are cases, however, where they...
peoples' sentiments expressed over social media
null
extractive
233
65
How do you establish the ground truth of who won a debate?
[ "Data Analysis\tIn this section, we analyze the presidential debates data and show some trends. First, we look at the trend of the tweet frequency. Figure FIGREF21 shows the trends of the tweet frequency and the number of TV viewers as the debates progress over time. We observe from Figures FIGREF21 and FIGREF21 th...
experts in Washington Post
null
extractive
234
66
How much better is performance of proposed method than state-of-the-art methods in experiments?
[ "Experiments ::: Experiments Setting\tIn evaluation, we compare our method with three types of models: 1) Typical Methods. Three typical knowledge graph embedding methods includes TransE, TransR and TransH are selected as baselines. For TransE, the dissimilarity measure is implemented with L1-norm, and relation as ...
Accuracy of best proposed method KANE (LSTM+Concatenation) are 0.8011, 0.8592, 0.8605 compared to best state-of-the art method R-GCN + LR 0.7721, 0.8193, 0.8229 on three datasets respectively.
null
free_form
235
66
What further analysis is done?
[ "Experiments ::: Entity Classification ::: Efficiency Evaluation.\tFigure FIGREF30 shows the test accuracy with increasing epoch on DBP24K and Game30K. We can see that test accuracy first rapidly increased in the first ten iterations, but reaches a stable stages when epoch is larger than 40. Figure FIGREF31 shows t...
we use t-SNE tool to visualize the learned embedding
null
extractive
236
66
What seven state-of-the-art methods are used for comparison?
[ "Conclusion and Future Work\tMany recent works have demonstrated the benefits of knowledge graph embedding in knowledge graph completion, such as relation extraction. However, We argue that knowledge graph embedding method still have room for improvement. First, TransE and its most extensions only take direct relat...
TransE, TransR and TransH, PTransE, and ALL-PATHS, R-GCN and KR-EAR
null
extractive
237
66
What three datasets are used to measure performance?
[ "Experiments ::: Experiments Setting\tIn evaluation, we compare our method with three types of models: 1) Typical Methods. Three typical knowledge graph embedding methods includes TransE, TransR and TransH are selected as baselines. For TransE, the dissimilarity measure is implemented with L1-norm, and relation as ...
FB24K, DBP24K, Game30K
null
extractive
238
66
How does KANE capture both high-order structural and attribute information of KGs in an efficient, explicit and unified manner?
[ "The main contributions of this study are as follows: 1) We highlight the importance of explicitly modeling the high-order structural and attribution information of KGs to provide better knowledge graph embedding. 2) We proposed a new method KANE, which achieves can capture both high-order structural and attribute ...
To capture both high-order structural information of KGs, we used an attention-based embedding propagation method.
null
extractive
239
66
What are recent works on knowedge graph embeddings authors mention?
[ "Related Work\tIn recent years, there are many efforts in Knowledge Graph Embeddings for KGs aiming to encode entities and relations into a continuous low-dimensional embedding spaces. Knowledge Graph Embedding provides a very simply and effective methods to apply KGs in various artificial intelligence applications...
entity types or concepts , relations paths , textual descriptions , , logical rules , deep neural network models
null
extractive
240
67
Do they report results only on English data?
[ "Dataset Creation\tWe use hashtag-based supervision to create our datasets, similar to tasks like emotion classification . The tweets are downloaded using Twitter API (https://dev.twitter.com/). We remove non-Unicode characters, and eliminate tweets that contain hyperlinks and also tweets that are shorter than 6 w...
Yes
null
yes_no
241
67
Do the authors mention any confounds to their study?
[ "Introduction\tThe ubiquity of communication devices has made social media highly accessible. The content on these media reflects a user's day-to-day activities. This includes content created under the influence of alcohol. In popular culture, this has been referred to as `drunk-texting'. In this paper, we introduc...
No
null
yes_no
242
67
What baseline model is used?
[ "To identify a more refined set of words that correspond to the two classes, we also estimated 20 topics for the dataset by estimating an LDA model . We then consider top 10 words per topic, for both classes. This results in 400 LDA-specific unigrams that are then used as features.", "Dataset Creation\tWe use ha...
Human evaluators
null
free_form
243
67
What stylistic features are used to detect drunk texts?
[ "A Computational Approach to Automatic Prediction of Drunk Texting\tAlcohol abuse may lead to unsociable behavior such as crime, drunk driving, or privacy leaks. We introduce automatic drunk-texting prediction as the task of identifying whether a text was written when under the influence of alcohol. We experiment w...
LDA unigrams (Presence/Count), POS Ratio, #Named Entity Mentions, #Discourse Connectors, Spelling errors, Repeated characters, Capitalisation, Length, Emoticon (Presence/Count ) and Sentiment Ratio
null
free_form
244
67
Is the data acquired under distant supervision verified by humans at any stage?
[ "A Computational Approach to Automatic Prediction of Drunk Texting\tAlcohol abuse may lead to unsociable behavior such as crime, drunk driving, or privacy leaks. We introduce automatic drunk-texting prediction as the task of identifying whether a text was written when under the influence of alcohol. We experiment w...
Yes
null
yes_no
245
67
Do the authors equate drunk tweeting with drunk texting?
[ "Introduction\tThe ubiquity of communication devices has made social media highly accessible. The content on these media reflects a user's day-to-day activities. This includes content created under the influence of alcohol. In popular culture, this has been referred to as `drunk-texting'. In this paper, we introduc...
Yes
null
yes_no
246
68
What corpus was the source of the OpenIE extractions?
[ "Experiments\tComparing our method with two state-of-the-art systems for 4th and 8th grade science exams, we demonstrate that (a) TupleInf with only automatically extracted tuples significantly outperforms TableILP with its original curated knowledge as well as with additional tuples, and (b) TupleInf's complementa...
domain-targeted $~$ 80K sentences and 280 GB of plain text extracted from web pages used by aristo2016:combining
null
extractive
247
68
What is the accuracy of the proposed technique?
[ "Answering Complex Questions Using Open Information Extraction\tWhile there has been substantial progress in factoid question-answering (QA), answering complex questions remains challenging, typically requiring both a large body of knowledge and inference techniques. Open Information Extraction (Open IE) provides a...
51.7 and 51.6 on 4th and 8th grade question sets with no curated knowledge. 47.5 and 48.0 on 4th and 8th grade question sets when both solvers are given the same knowledge
null
free_form
248
68
Is an entity linking process used?
[ "Using curated tables with TupleInf\tFor each question, we select the 7 best matching tables using the tf-idf score of the table w.r.t. the question tokens and top 20 rows from each table using the Jaccard similarity of the row with the question. (same as tableilp2016). We then convert the table rows into the tupl...
No
null
yes_no
249
68
Are the OpenIE extractions all triples?
[ "Using Open IE tuples with TableILP\tWe create an additional table in TableILP with all the tuples in $T$ . Since TableILP uses fixed-length $(subject; predicate; object)$ triples, we need to map tuples with multiple objects to this format. For each object, $O_i$ in the input Open IE tuple $(S; P; O_1; O_2 \\ldots ...
No
null
yes_no
250
68
What method was used to generate the OpenIE extractions?
[ "Answering Complex Questions Using Open Information Extraction\tWhile there has been substantial progress in factoid question-answering (QA), answering complex questions remains challenging, typically requiring both a large body of knowledge and inference techniques. Open Information Extraction (Open IE) provides a...
for each multiple-choice question $(q,A) \in Q_\mathit {tr}$ and each choice $a \in A$ , we use all non-stopword tokens in $q$ and $a$ as an ElasticSearch query against S, take the top 200 hits, run Open IE v4, and aggregate the resulting tuples over all $a \in A$ and over all questions in $Q_\mathit {tr}$
null
extractive
251
68
Can the method answer multi-hop questions?
[ "To avoid issues such as (Planet; orbit; Sun) matching the sample question in the introduction (“Which object $\\ldots $ orbits around a planet”), we also add an ordering constraint (third group in Table 1 ). Its worth mentioning that TupleInf only combines parallel evidence i.e. each tuple must connect words in th...
Yes
null
yes_no
252
68
What was the textual source to which OpenIE was applied?
[ "To avoid issues such as (Planet; orbit; Sun) matching the sample question in the introduction (“Which object $\\ldots $ orbits around a planet”), we also add an ordering constraint (third group in Table 1 ). Its worth mentioning that TupleInf only combines parallel evidence i.e. each tuple must connect words in th...
domain-targeted $~$ 80K sentences and 280 GB of plain text extracted from web pages used by aristo2016:combining
null
extractive
253
68
What OpenIE method was used to generate the extractions?
[ "Answering Complex Questions Using Open Information Extraction\tWhile there has been substantial progress in factoid question-answering (QA), answering complex questions remains challenging, typically requiring both a large body of knowledge and inference techniques. Open Information Extraction (Open IE) provides a...
for each multiple-choice question $(q,A) \in Q_\mathit {tr}$ and each choice $a \in A$ , we use all non-stopword tokens in $q$ and $a$ as an ElasticSearch query against S, take the top 200 hits, run Open IE v4, and aggregate the resulting tuples over all $a \in A$ and over all questions in $Q_\mathit {tr}$
null
extractive
254
68
Is their method capable of multi-hop reasoning?
[ "To avoid issues such as (Planet; orbit; Sun) matching the sample question in the introduction (“Which object $\\ldots $ orbits around a planet”), we also add an ordering constraint (third group in Table 1 ). Its worth mentioning that TupleInf only combines parallel evidence i.e. each tuple must connect words in th...
Yes
null
yes_no
255
69
Do the authors offer any hypothesis about why the dense mode outperformed the sparse one?
[ "An Unsupervised Word Sense Disambiguation System for Under-Resourced Languages\tIn this paper, we present Watasense, an unsupervised system for word sense disambiguation. Given a sentence, the system chooses the most relevant sense of each input word with respect to the semantic similarity between the given senten...
Yes
null
yes_no
256
69
What evaluation is conducted?
[ "Evaluation\tWe conduct our experiments using the evaluation methodology of SemEval 2010 Task 14: Word Sense Induction & Disambiguation . In the gold standard, each word is provided with a set of instances, i.e., the sentences containing the word. Each instance is manually annotated with the single sense identifie...
Word Sense Induction & Disambiguation
null
extractive
257
69
Which corpus of synsets are used?
[ "It uses the synsets induced in an unsupervised way by the Watset[CWnolog, MCL] method and the semantic relations from such dictionaries as Wiktionary referred as Joint INLINEFORM0 Exp INLINEFORM1 SWN in Ustalov:17:dialogue. This is the only automatically built inventory we use in the evaluation. RuThes, a large-s...
Wiktionary
null
extractive
258
69
What measure of semantic similarity is used?
[ "Quality Measure\tThe original SemEval 2010 Task 14 used the V-Measure external clustering measure . However, this measure is maximized by clustering each sentence into his own distinct cluster, i.e., a `dummy' singleton baseline. This is achieved by the system deciding that every ambiguous word in every sentence ...
cosine similarity
null
extractive
259
70
Which retrieval system was used for baselines?
[ "The GA model achieves $48.3\\%$ accuracy on the set of instances for which the answer is in context, however, a search accuracy of only $65\\%$ means its overall performance is lower. This can improve with improved retrieval. For Quasar-T, both the neural models significantly outperform the heuristic models, with ...
The dataset comes with a ranked set of relevant documents. Hence the baselines do not use a retrieval system.
null
free_form
260
71
What word embeddings were used?
[ "Data and model ::: Pre-trained word Embeddings\tWe use the word embeddings for Vietnamese that created by Kyubyong Park and Edouard Grave at al: Kyubyong Park: In his project, he uses two methods including fastText and word2vec to generate word embeddings from wikipedia database backup dumps. His word embedding is...
Kyubyong Park, Edouard Grave et al
null
extractive
261
71
What type of errors were produced by the BLSTM-CNN-CRF system?
[ "LSTM-CNN-CRF: vietnam⟨PER⟩ Ca_sĩ Nguyễn Văn A ⟨PER⟩ Annotator: vietnamCa_sĩ ⟨PER⟩ Nguyễn Văn A ⟨PER⟩ Wrong tag: The error where the model extracted tokens as an NE and only the tag type was wrong. LSTM-CNN-CRF: vietnamKhám phá ⟨PER⟩ Yangsuri ⟨PER⟩ Annotator: vietnamKhám phá ⟨LOC⟩ Yangsuri ⟨LOC⟩ Wrong range and tag...
No extraction, No annotation, Wrong range, Wrong tag, Wrong range and tag
null
extractive
262
71
How much better was the BLSTM-CNN-CRF than the BLSTM-CRF?
[ "Error Analysis for Vietnamese Named Entity Recognition on Deep Neural Network Models\tIn recent years, Vietnamese Named Entity Recognition (NER) systems have had a great breakthrough when using Deep Neural Network methods. This paper describes the primary errors of the state-of-the-art NER systems on Vietnamese la...
Best BLSTM-CNN-CRF had F1 score 86.87 vs 86.69 of best BLSTM-CRF
null
free_form
263
72
What supplemental tasks are used for multitask learning?
[ "Multitask Learning\tAs mentioned in Section \"Modeling Question-External Comments\" , we also explored a multitask learning framework that jointly learns to predict the relationships of all three tasks. We set $0.8$ for the main task (task C) and $0.1$ for the other auxiliary tasks. The MAP score did not improve, ...
Multitask learning is used for the task of predicting relevance of a comment on a different question to a given question, where the supplemental tasks are predicting relevance between the questions, and between the comment and the corresponding question
null
free_form
264
72
Is the improvement actually coming from using an RNN?
[ "By applying this method, the MAP score increased slightly from $0.5723$ to $0.5789$ but the F1 score improved from $0.4334$ to $0.5860$ . For task C, we used task A's data directly. The results are very similar with a slight improvement on MAP, but large improvement on F1 score from $0.1449$ to $0.2064$ .", "Rob...
No
null
yes_no
265
72
How much performance gap between their approach and the strong handcrafted method?
[ "Recurrent Neural Network Encoder with Attention for Community Question Answering\tWe apply a general recurrent neural network (RNN) encoder framework to community question answering (cQA) tasks. Our approach does not rely on any linguistic processing, and can be applied to different languages or domains. Further i...
0.007 MAP on Task A, 0.032 MAP on Task B, 0.055 MAP on Task C
null
free_form
266
72
Did they experimnet in other languages?
[ "Multitask Learning\tAs mentioned in Section \"Modeling Question-External Comments\" , we also explored a multitask learning framework that jointly learns to predict the relationships of all three tasks. We set $0.8$ for the main task (task C) and $0.1$ for the other auxiliary tasks. The MAP score did not improve, ...
Yes
null
yes_no
267
73
Do they use multi-attention heads?
[ "Attentional Encoder Layer\tThe attentional encoder layer is a parallelizable and interactive alternative of LSTM and is applied to compute the hidden states of the input embeddings. This layer consists of two submodules: the Multi-Head Attention (MHA) and the Point-wise Convolution Transformation (PCT). Multi-Head...
Yes
null
yes_no
268
73
How big is their model?
[ "We also apply pre-trained BERT to this task and show our model enhances the performance of basic BERT model. Experimental results on three benchmark datasets show that the proposed model achieves competitive performance and is a lightweight alternative of the best RNN based models. The main contributions of this ...
Proposed model has 1.16 million parameters and 11.04 MB.
null
free_form
269
73
How is their model different from BERT?
[ "Attentional Encoder Layer\tThe attentional encoder layer is a parallelizable and interactive alternative of LSTM and is applied to compute the hidden states of the input embeddings. This layer consists of two submodules: the Multi-Head Attention (MHA) and the Point-wise Convolution Transformation (PCT). Multi-Head...
overall architecture of the proposed Attentional Encoder Network (AEN), which mainly consists of an embedding layer, an attentional encoder layer, a target-specific attention layer, and an output layer.
null
extractive
270
74
What datasets were used?
[ "JESSI is trained using only the datasets given on the shared task, without using any additional external data. Despite this, JESSI performs second on Subtask A with an F1 score of 77.78% among 33 other team submissions. It also performs well on Subtask B with an F1 score of 79.59%.", "Organizers observed four ma...
datasets given on the shared task, without using any additional external data
null
extractive
271
74
How did they do compared to other teams?
[ "JESSI is trained using only the datasets given on the shared task, without using any additional external data. Despite this, JESSI performs second on Subtask A with an F1 score of 77.78% among 33 other team submissions. It also performs well on Subtask B with an F1 score of 79.59%.", "Introduction\tOpinion minin...
second on Subtask A with an F1 score of 77.78% among 33 other team submissions, performs well on Subtask B with an F1 score of 79.59%
null
extractive
272
75
Which tested technique was the worst performer?
[ "Benchmarks\tWe performed benchmark experiments on the dataset using several different algorithms. In all experiments, we have discarded the data labelled with Surprise and Disgust. We pre-processed the data by using the SpaCy pipeline. We masked out named entities with entity-type specific placeholders to reduce t...
Depeche + SVM
null
extractive
273
75
How many emotions do they look at?
[ "Dataset ::: Plutchik’s Wheel of Emotions\tThe dataset is annotated based on a modified Plutchik’s wheel of emotions. The original Plutchik’s wheel consists of 8 primary emotions: Joy, Sadness, Anger, Fear, Anticipation, Surprise, Trust, Disgust. In addition, more complex emotions can be formed by combing two basic...
9
null
free_form
274
75
What are the baseline benchmarks?
[ "DENS: A Dataset for Multi-class Emotion Analysis\tWe introduce a new dataset for multi-class emotion analysis from long-form narratives in English. The Dataset for Emotions of Narrative Sequences (DENS) was collected from both classic literature available on Project Gutenberg and modern online narratives available...
TF-IDF + SVM, Depeche + SVM, NRC + SVM, TF-NRC + SVM, Doc2Vec + SVM, Hierarchical RNN, BiRNN + Self-Attention, ELMo + BiRNN, Fine-tuned BERT
null
extractive
275
75
What is the size of this dataset?
[ "Benchmarks ::: Bi-directional RNN and Self-Attention (BiRNN + Self-Attention)\tOne challenge with RNN-based solutions for text classification is finding the best way to combine word-level representations into higher-level representations. Self-attention , , has been adapted to text classification, providing impro...
9710 passages, with an average of 6.24 sentences per passage, 16.16 words per sentence, and an average length of 86 words
null
extractive
276
75
How many annotators were there?
[ "Dataset ::: Passage Selection\tWe selected both classic and modern narratives in English for this dataset. The modern narratives were sampled based on popularity from Wattpad. We parsed selected narratives into passages, where a passage is considered to be eligible for annotation if it contained between 40 and 200...
3
null
extractive
277
76
Can SCRF be used to pretrain the model?
[ "Conclusion\tWe investigated multitask learning with CTC and SCRF for speech recognition in this paper. Using an RNN encoder for feature extraction, both CTC and SCRF can be trained end-to-end, and the two models can be trained together by interpolating the two loss functions. From experiments on the TIMIT dataset,...
No
null
yes_no
278
77
What conclusions are drawn from the syntactic analysis?
[ "Conclusions\tWe highlight the problem of translating between languages with different morphological systems, in which the target translation must contain gender and number information that is not available in the source. We propose a method for injecting such information into a pre-trained NMT model in a black-box...
our method enables to control the morphological realization of first and second-person pronouns, together with verbs and adjectives related to them
null
extractive
279
77
What type of syntactic analysis is performed?
[ "Qualitative Results\tThe BLEU score is an indication of how close the automated translation is to the reference translation, but does not tell us what exactly changed concerning the gender and number properties we attempt to control. We perform a finer-grained analysis focusing on the relation between the injected...
Speaker's Gender Effects, Interlocutors' Gender and Number Effects
null
extractive
280
77
How is it demonstrated that the correct gender and number information is injected using this system?
[ "Conclusions\tWe highlight the problem of translating between languages with different morphological systems, in which the target translation must contain gender and number information that is not available in the source. We propose a method for injecting such information into a pre-trained NMT model in a black-box...
correct information substantially improves it - we see an increase of up to 2.3 BLEU over the baseline, Finally, the “She said” prefixes substantially increase the number of feminine-marked verbs, bringing the proportion much closer to that of the reference
null
extractive
281
77
Which neural machine translation system is used?
[ "Filling Gender&Number Gaps in Neural Machine Translation with Black-box Context Injection\tWhen translating from a language that does not morphologically mark information such as gender and number into a language that does, translation systems must\"guess\"this missing information, often leading to incorrect trans...
Google's machine translation system (GMT)
null
extractive
282
77
What are the components of the black-box context injection system?
[ "Filling Gender&Number Gaps in Neural Machine Translation with Black-box Context Injection\tWhen translating from a language that does not morphologically mark information such as gender and number into a language that does, translation systems must\"guess\"this missing information, often leading to incorrect trans...
supply an NMT system with knowledge regarding the speaker and interlocutor of first-person sentences
null
extractive
283
78
What normalization techniques are mentioned?
[ "Using different features\tWe explored different normalization techniques. FBanks with cepstral mean normalization (CMN) perform better than raw FBanks. We found using variance with mean normalization (CMVN) unnecessary for the task. Using deltas and delta-deltas improves model, so we used them in other experiments...
FBanks with cepstral mean normalization (CMN), variance with mean normalization (CMVN)
null
extractive
284
78
What features do they experiment with?
[ "Basic setup\tFor all experiments we used conversational speech from IARPA Babel Turkish Language Pack (LDC2016S10). This corpus contains about 80 hours of transcribed speech for training and 10 hours for development. The dataset is rather small compared to widely used benchmarks for conversational speech: English ...
40 mel-scaled log filterbank enegries (FBanks) computed every 10 ms with 25 ms window, deltas and delta-deltas (120 features in vector), spectrogram
null
extractive
285
78
Which architecture is their best model?
[ "Exploring End-to-End Techniques for Low-Resource Speech Recognition\tIn this work we present simple grapheme-based system for low-resource speech recognition using Babel data for Turkish spontaneous speech (80 hours). We have investigated different neural network architectures performance, including fully-convolut...
6-layer bLSTM with 1024 hidden units
null
extractive
286
79
What approach did previous models use for multi-span questions?
[ "Tag-based Multi-Span Extraction in Reading Comprehension\tWith models reaching human performance on many popular reading comprehension datasets in recent years, a new dataset, DROP, introduced questions that were expected to present a harder challenge for reading comprehension models. Among these new types of ques...
Only MTMSM specifically tried to tackle the multi-span questions. Their approach consisted of two parts: first train a dedicated categorical variable to predict the number of spans to extract and the second was to generalize the single-span head method of extracting a span
null
free_form
287
79
How they use sequence tagging to answer multi-span questions?
[ "Model ::: Objective and Training ::: Dealing with too Many Correct Tag Sequences\tThe number of correct tag sequences can be expressed by where $s$ is the number of spans in the answer and $\\#_i$ is the number of times the $i^\\text{th}$ span appears in the text. For questions with a reasonable amount of correct ...
To model an answer which is a collection of spans, the multi-span head uses the $\mathtt {BIO}$ tagging format : $\mathtt {B}$ is used to mark the beginning of a span, $\mathtt {I}$ is used to mark the inside of a span and $\mathtt {O}$ is used to mark tokens not included in a span
null
extractive
288
79
What is difference in peformance between proposed model and state-of-the art on other question types?
[ "Tag-based Multi-Span Extraction in Reading Comprehension\tWith models reaching human performance on many popular reading comprehension datasets in recent years, a new dataset, DROP, introduced questions that were expected to present a harder challenge for reading comprehension models. Among these new types of ques...
For single-span questions, the proposed LARGE-SQUAD improve performance of the MTMSNlarge baseline for 2.1 EM and 1.55 F1. For number type question, MTMSNlarge baseline have improvement over LARGE-SQUAD for 3,11 EM and 2,98 F1. For date question, LARGE-SQUAD have improvements in 2,02 EM but MTMSNlarge have improv...
null
free_form
289
79
What is the performance of proposed model on entire DROP dataset?
[ "Introduction\tThe task of reading comprehension, where systems must understand a single passage of text well enough to answer arbitrary questions about it, has seen significant progress in the last few years. With models reaching human performance on the popular SQuAD dataset , and with much of the most popular re...
The proposed model achieves EM 77,63 and F1 80,73 on the test and EM 76,95 and F1 80,25 on the dev
null
free_form
290
79
What is the previous model that attempted to tackle multi-span questions as a part of its design?
[ "Tag-based Multi-Span Extraction in Reading Comprehension\tWith models reaching human performance on many popular reading comprehension datasets in recent years, a new dataset, DROP, introduced questions that were expected to present a harder challenge for reading comprehension models. Among these new types of ques...
MTMSN
null
extractive
291
80
Does the system trained only using XR loss outperform the fully supervised neural system?
[ "Transfer Learning Between Related Tasks Using Expected Label Proportions\tDeep learning systems thrive on abundance of labeled training data but such data is not always available, calling for alternative methods of supervision. One such method is expectation regularization (XR) (Mann and McCallum, 2007), where mod...
Yes
null
yes_no
292
80
How accurate is the aspect based sentiment classifier trained only using the XR loss?
[ "Transfer Learning Between Related Tasks Using Expected Label Proportions\tDeep learning systems thrive on abundance of labeled training data but such data is not always available, calling for alternative methods of supervision. One such method is expectation regularization (XR) (Mann and McCallum, 2007), where mod...
BiLSTM-XR-Dev Estimation accuracy is 83.31 for SemEval-15 and 87.68 for SemEval-16. BiLSTM-XR accuracy is 83.31 for SemEval-15 and 88.12 for SemEval-16.
null
free_form
293
80
How is the expectation regularization loss defined?
[ "Introduction\tData annotation is a key bottleneck in many data driven algorithms. Specifically, deep learning models, which became a prominent tool in many data driven tasks in recent years, require large datasets to work well. However, many tasks require manual annotations which are relatively hard to obtain at s...
DISPLAYFORM0
null
extractive
294
81
What were the non-neural baselines used for the task?
[ "Baselines ::: Task 2 Baselines ::: Non-neural\t: The Lemming model is a log-linear model that performs joint morphological tagging and lemmatization. The model is globally normalized with the use of a second order linear-chain CRF. To efficiently calculate the partition function, the choice of lemmata are pruned w...
The Lemming model in
null
free_form
295
82
Which publicly available NLU dataset is used?
[ "Some of them also use bigger, not publicly available datasets, which appear to be similar to the NLU-BM in terms of number of intents and slots, but they cover no more than three or four domains. Our work stands out for its more challenging NLU setting, since we are dealing with a higher number of domains/scenario...
ROMULUS dataset, NLU-Benchmark dataset
null
extractive
296
82
What metrics other than entity tagging are compared?
[ "Following , we then evaluated a metric that combines intent and entities, computed by simply summing up the two confusion matrices (Table TABREF23). Results highlight the contribution of the entity tagging task, where HERMIT outperforms the other approaches. Paired-samples t-tests were conducted to compare the HER...
We also report the metrics in for consistency, we report the span F1, Exact Match (EM) accuracy of the entire sequence of labels, metric that combines intent and entities
null
extractive
297
83
Do they provide decision sequences as supervision while training models?
[ "Related Works\tSkip-reading , , is an existing setting in which MRC models read partial documents. Concretely, these methods assume that not all tokens in the input sequence are useful, and therefore learn to skip irrelevant tokens based on the current input and their internal memory. Since skipping decisions are...
No
null
yes_no
298
83
What are the models evaluated on?
[ "iMRC: Making MRC Interactive ::: Evaluation Metric\tSince iMRC involves both MRC and RL, we adopt evaluation metrics from both settings. First, as a question answering task, we use $\\text{F}_1$ score to compare predicted answers against ground-truth, as in previous works. When there exist multiple ground-truth an...
They evaluate F1 score and agent's test performance on their own built interactive datasets (iSQuAD and iNewsQA)
null
free_form
299
83
How do they train models in this setup?
[ "Interactive Machine Comprehension with Information Seeking Agents\tExisting machine reading comprehension (MRC) models do not scale effectively to real-world applications like web-level information retrieval and question answering (QA). We argue that this stems from the nature of MRC datasets: most of these are st...
Thus, our task requires models to `feed themselves' rather than spoon-feeding them with information. This casts MRC as a sequential decision-making problem amenable to reinforcement learning (RL).
null
extractive