id stringlengths 10 10 | title stringlengths 12 156 | abstract stringlengths 279 2.02k | full_text dict | qas dict | figures_and_tables dict | text stringlengths 0 170k |
|---|---|---|---|---|---|---|
1811.04604 | Learning Personalized End-to-End Goal-Oriented Dialog | Most existing works on dialog systems only consider conversation content while neglecting the personality of the user the bot is interacting with, which begets several unsolved issues. In this paper, we present a personalized end-to-end model in an attempt to leverage personalization in goal-oriented dialogs. We first ... | {
"section_name": [
"Introduction",
"Related Work",
"End-to-End Memory Network",
"Personalized Dialog System",
"Notation",
"Profile Model",
"Preference Model",
"Combined Model",
"Dataset",
"Baselines",
"Experiment Settings",
"Results",
"Analysis",
"Analysis of P... | {
"question": [
"What datasets did they use?"
],
"question_id": [
"43878a6a8fc36aaae29d95815355aaa7d25c3b53"
],
"nlp_background": [
""
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
""
],
"question_writer": [
"c1fbdd7a261021041f75fbe00a55b4c3... | {
"caption": [
"Figure 1: Examples to show the common issues with content-based models. We can see that the content-based model (1) is incapable of adjusting appellations and language styles, (2) fails to provide the best candidate, and (3) fails to choose the correct answer when facing ambiguities. (a) Three dia... | Introduction
There has been growing research interest in training dialog systems with end-to-end models BIBREF0 , BIBREF1 , BIBREF2 in recent years. These models are directly trained on past dialogs, without assumptions on the domain or dialog state structure BIBREF3 . One of their limitations is that they select respo... |
1909.05890 | Determining the Scale of Impact from Denial-of-Service Attacks in Real Time Using Twitter | Denial of Service (DoS) attacks are common in on-line and mobile services such as Twitter, Facebook and banking. As the scale and frequency of Distributed Denial of Service (DDoS) attacks increase, there is an urgent need for determining the impact of the attack. Two central challenges of the task are to get feedback f... | {
"section_name": [
"Introduction",
"Related Work",
"Approach",
"Approach ::: Data Collection",
"Approach ::: Preprocessing",
"Approach ::: Create LDA Models",
"Approach ::: The attack topics",
"Approach ::: The attack tweets",
"Approach ::: Optional Classifier Layer",
"Approac... | {
"question": [
"Do twitter users tend to tweet about the DOS attack when it occurs? How much data supports this assumption?",
"What is the training and test data used?",
"Was performance of the weakly-supervised model compared to the performance of a supervised model?"
],
"question_id": [
"68ff2a... | {
"caption": [
"Figure 1: Workflow to process tweets gathered and build a model to rank future tweets that likely to be related to a DoS attack. The ranked tweets are used to measure the severity of the attack.",
"Figure 2: Plate notation of LDA [4]. The outer box denotes documents in the corpus andM is the n... | Introduction
Denial of Service attacks are explicit attempts to stop legitimate users from accessing specific network systems BIBREF0. Attackers try to exhaust network resources like bandwidth, or server resources like CPU and memory. As a result, the targeted system slows down or becomes unusable BIBREF1. On-line serv... |
1912.06927 | #MeTooMA: Multi-Aspect Annotations of Tweets Related to the MeToo Movement | In this paper, we present a dataset containing 9,973 tweets related to the MeToo movement that were manually annotated for five different linguistic aspects: relevance, stance, hate speech, sarcasm, and dialogue acts. We present a detailed account of the data collection and annotation processes. The annotations have a ... | {
"section_name": [
"Introduction",
"Related Datasets",
"Dataset ::: Data Collection",
"Dataset ::: Annotation Task",
"Dataset ::: Annotation Task ::: Task 1: Relevance",
"Dataset ::: Annotation Task ::: Task 2: Stance",
"Dataset ::: Annotation Task ::: Task 3: Hate Speech",
"Dataset :... | {
"question": [
"Do the tweets come from a specific region?"
],
"question_id": [
"a4422019d19f9c3d95ce8dc1d529bf3da5edcfb1"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
"metoo"
],
"question_writer": [
... | {
"caption": [
"Table 1: Summary of related datasets.",
"Table 2: Distribution of tweets by the country.",
"Figure 1: Choropleth world map recording tweet frequency.",
"Figure 2: Geographical distribution of various class labels.",
"Figure 3: Word cloud representation of the dataset: font size is ... | Introduction
Over the last couple of years, the MeToo movement has facilitated several discussions about sexual abuse. Social media, especially Twitter, was one of the leading platforms where people shared their experiences of sexual harassment, expressed their opinions, and also offered support to victims. A large por... |
1909.01247 | Introducing RONEC -- the Romanian Named Entity Corpus | We present RONEC - the Named Entity Corpus for the Romanian language. The corpus contains over 26000 entities in ~5000 annotated sentences, belonging to 16 distinct classes. The sentences have been extracted from a copy-right free newspaper, covering several styles. This corpus represents the first initiative in the Ro... | {
"section_name": [
"Introduction",
"Introduction ::: Related corpora",
"Introduction ::: Related corpora ::: ROCO corpus",
"Introduction ::: Related corpora ::: ROMBAC corpus",
"Introduction ::: Related corpora ::: CoRoLa corpus",
"Corpus Description",
"Corpus Description ::: BRAT format"... | {
"question": [
"Did they experiment with the corpus?",
"What writing styles are present in the corpus?",
"How did they determine the distinct classes?"
],
"question_id": [
"bb169a0624aefe66d3b4b1116bbd152d54f9e31b",
"0d7de323fd191a793858386d7eb8692cc924b432",
"ca8e023d142d89557714d67739e1... | {
"caption": [
"Table 1: Stylistic domains and examples (bold marks annotated entities)",
"Table 2: Corpus statistics: Each entity is marked with a class and can span one or more words",
"Table 3: CoNLL-U Plus format for the first 20 tokens of sentence ”Tot ı̂n cadrul etapei a 2-a, a avut loc ı̂ntâlnirea... | Introduction
Language resources are an essential component in entire R&D domains. From the humble but vast repositories of monolingual texts that are used by the newest language modeling approaches like BERT and GPT, to parallel corpora that allows our machine translation systems to inch closer to human performance, to... |
1706.01723 | A General-Purpose Tagger with Convolutional Neural Networks | We present a general-purpose tagger based on convolutional neural networks (CNN), used for both composing word vectors and encoding context information. The CNN tagger is robust across different tagging tasks: without task-specific tuning of hyper-parameters, it achieves state-of-the-art results in part-of-speech taggi... | {
"section_name": [
"Introduction",
"Model",
"Character Composition Model",
"Context Encoding Model",
"Hyper-parameters",
"Data",
"Tasks",
"Setups",
"Results",
"Unnormalized Text",
"Conclusion"
],
"paragraphs": [
[
"Recently, character composition models have ... | {
"question": [
"Do they jointly tackle multiple tagging problems?",
"How many parameters does their CNN have?",
"How do they confirm their model working well on out-of-vocabulary problems?"
],
"question_id": [
"3fddd9f6707b9e40e35518dae7f6da7c4cb77d16",
"676c874266ee0388fe5b9a75e1006796c68c3c... | {
"caption": [
"Figure 1: Diagram of the CNN tagger.",
"Table 1: Tagging accuracies of the three taggers in the three tasks on the test set of UD-1.2, the highest accuracy for each task on each language is marked in bold face.",
"Figure 2: POS tagging accuracies on the dev set with the three modifications... | Introduction
Recently, character composition models have shown great success in many NLP tasks, mainly because of their robustness in dealing with out-of-vocabulary (OOV) words by capturing sub-word informations. Among the character composition models, bidirectional long short-term memory (LSTM) models and convolutiona... |
1608.06378 | Towards Machine Comprehension of Spoken Content: Initial TOEFL Listening Comprehension Test by Machine | Multimedia or spoken content presents more attractive information than plain text content, but it's more difficult to display on a screen and be selected by a user. As a result, accessing large collections of the former is much more difficult and time-consuming than the latter for humans. It's highly attractive to deve... | {
"section_name": [
"Introduction",
"Task Definition and Contributions",
"Proposed Approach",
"Question Representation",
"Story Attention Module",
"Hopping",
"Answer Selection",
"Experimental Setup",
"Baselines",
"Results",
"Analysis on a typical example",
"Conclusions"... | {
"question": [
"What approach does this work propose for the new task?",
"What is the new task proposed in this work?"
],
"question_id": [
"a53683d1a0647c80a4398ff8f4a03e11c0929be2",
"0fd7d12711dfe0e35467a7ee6525127378a1bacb"
],
"nlp_background": [
"two",
"two"
],
"topic_backgroun... | {
"caption": [
"Figure 2: The overall structure of the proposed Attention-based Multi-hop Recurrent Neural Network (AMRNN) model.",
"Figure 3: (A) The Question Vector Representation and (B) The Attention Mechanism.",
"Table 1: Accuracy results of different models",
"Figure 4: Visualization of the atte... | Introduction
With the popularity of shared videos, social networks, online course, etc, the quantity of multimedia or spoken content is growing much faster beyond what human beings can view or listen to. Accessing large collections of multimedia or spoken content is difficult and time-consuming for humans, even if thes... |
1808.02022 | Principles for Developing a Knowledge Graph of Interlinked Events from News Headlines on Twitter | The ever-growing datasets published on Linked Open Data mainly contain encyclopedic information. However, there is a lack of quality structured and semantically annotated datasets extracted from unstructured real-time sources. In this paper, we present principles for developing a knowledge graph of interlinked events u... | {
"section_name": [
"Introduction",
"Notation and Problem Statement",
"Outline of The Required Steps",
"Background Data Model",
"Using Existing Data Models",
"Developing a Data Model",
"Using Singleton Property",
"Event Annotation",
"Entity Annotation",
"Interlinking Events",
... | {
"question": [
"Which news organisations are the headlines sourced from?"
],
"question_id": [
"5dc2f79cd8078d5976f2df9ab128d4517e894257"
],
"nlp_background": [
"five"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
"twitter"
],
"quest... | {
"caption": [
"Table 1: Samples of news headlines from different publishers on Twitter.",
"Figure 1: The pipeline of the required steps for developing a knowledge graph of interlinked events.",
"Figure 2: Schematic representation of LODE.",
"Figure 4: Schematic representation of SEM.",
"Figure 3:... | Introduction
Several successful efforts have led to publishing huge RDF (Resource Description Framework) datasets on Linked Open Data (LOD) such as DBpedia BIBREF0 and LinkedGeoData BIBREF1 . However, these sources are limited to either structured or semi-structured data. So far, a significant portion of the Web conten... |
1909.01515 | Meta Relational Learning for Few-Shot Link Prediction in Knowledge Graphs | Link prediction is an important way to complete knowledge graphs (KGs), while embedding-based methods, effective for link prediction in KGs, perform poorly on relations that only have a few associative triples. In this work, we propose a Meta Relational Learning (MetaR) framework to do the common but challenging few-sh... | {
"section_name": [
"Introduction",
"Related Work",
"Knowledge Graph Embedding",
"Meta-Learning",
"Task Formulation",
"Method",
"Relation-Meta Learner",
"Embedding Learner",
"Training Objective",
"Experiments",
"Datasets and Evaluation Metrics",
"Implementation",
"R... | {
"question": [
"What meta-information is being transferred?",
"What datasets are used to evaluate the approach?"
],
"question_id": [
"4226a1830266ed5bde1b349205effafe7a0e2337",
"5fb348b2d7b012123de93e79fd46a7182fd062bd"
],
"nlp_background": [
"five",
"five"
],
"topic_background": ... | {
"caption": [
"Figure 1: An example of 3-shot link prediction in KGs. One task represents observing only three instances of one specific relation and conducting link prediction on this relation. Our model focuses on extracting relationspecific meta information by a kind of relational learner which is shared acro... | Introduction
A knowledge graph is composed by a large amount of triples in the form of $(head\; entity,\, relation,\, tail\; entity)$ ( $(h, r, t)$ in short), encoding knowledge and facts in the world. Many KGs have been proposed BIBREF0 , BIBREF1 , BIBREF2 and applied to various applications BIBREF3 , BIBREF4 , BIBREF... |
1811.01299 | SimplerVoice: A Key Message&Visual Description Generator System for Illiteracy | We introduce SimplerVoice: a key message and visual description generator system to help low-literate adults navigate the information-dense world with confidence, on their own. SimplerVoice can automatically generate sensible sentences describing an unknown object, extract semantic meanings of the object usage in the f... | {
"section_name": [
"Introduction",
"Related Work",
"System Design",
"Overview",
"Object2Text",
"Text2Visual",
"Evaluation",
"Case Study",
"Prototype System",
"Experiment",
"Conclusion",
"Acknowledgments"
],
"paragraphs": [
[
"Illiteracy has been one of th... | {
"question": [
"Does their solution involve connecting images and text?",
"Which model do they use to generate key messages?"
],
"question_id": [
"7ff48fe5b7bd6b56553caacc891ce3d7e0070440",
"54a2c08aa55c3db9b30ae2922c96528d3f4fc733"
],
"nlp_background": [
"infinity",
"infinity"
],
... | {
"caption": [
"Figure 1. Overall system structure & workflow of SimplerVoice, with 4 main components: input retrieval, object2text, text2visual, and output display.",
"Figure 2. The process of generating S-V-O in Object2Text.",
"Figure 3. An example of ontology sub-tree for \"Bagel\" built by crawled dat... | Introduction
Illiteracy has been one of the most serious pervasive problems all over the world. According to the U. S. Department of Education, the National Center for Education Statistics, approximately 32 million adults in the United States are not able to read, which is about 14% of the entire adult population BIBRE... |
1912.01220 | Modelling Semantic Categories using Conceptual Neighborhood | While many methods for learning vector space embeddings have been proposed in the field of Natural Language Processing, these methods typically do not distinguish between categories and individuals. Intuitively, if individuals are represented as vectors, we can think of categories as (soft) regions in the embedding spa... | {
"section_name": [
"Introduction",
"Related Work",
"Model Description",
"Model Description ::: Step 1: Predicting Conceptual Neighborhood from Embeddings",
"Model Description ::: Step 1: Predicting Conceptual Neighborhood from Embeddings ::: Generating Distant Supervision Labels",
"Model Desc... | {
"question": [
"What experiments they perform to demonstrate that their approach leads more accurate region based representations?",
"How they indentify conceptual neighbours?"
],
"question_id": [
"ecb680d79e847beb7c1aa590d288a7313908d64a",
"b622f57c4e429b458978cb8863978d7facab7cfe"
],
"nlp_b... | {
"caption": [
"Figure 1: Using conceptual neighborhood for estimating category boundaries.",
"Table 1: Cross-validation results on the training split of the text classifier (accuracy and macro-average F1, precision and recall).",
"Figure 2: Instances of three BabelNet categories which intuitively can be ... | Introduction
Vector space embeddings are commonly used to represent entities in fields such as machine learning (ML) BIBREF0, natural language processing (NLP) BIBREF1, information retrieval (IR) BIBREF2 and cognitive science BIBREF3. An important point, however, is that such representations usually represent both indi... |
1908.06151 | The Transference Architecture for Automatic Post-Editing | In automatic post-editing (APE) it makes sense to condition post-editing (pe) decisions on both the source (src) and the machine translated text (mt) as input. This has led to multi-source encoder based APE approaches. A research challenge now is the search for architectures that best support the capture, preparation a... | {
"section_name": [
"Introduction",
"Related Research",
"Transference Model for APE",
"Experiments",
"Experiments ::: Data",
"Experiments ::: Experiment Setup",
"Experiments ::: Hyper-parameter Setup",
"Results",
"Results ::: Baselines",
"Results ::: Single-Encoder Transformer ... | {
"question": [
"What experiment result led to conclussion that reducing the number of layers of the decoder does not matter much?",
"How much is performance hurt when using too small amount of layers in encoder?",
"What was previous state of the art model for automatic post editing?"
],
"question_id"... | {
"caption": [
"Figure 1: The transference model architecture for APE ({src,mt}tr → pe).",
"Table 1: Evaluation results on the WMT APE test set 2016, and test set 2017 for the PBSMT task; (±X) value is the improvement over wmt18smtbest (x4). The last section of the table shows the impact of increasing and dec... | Introduction
The performance of state-of-the-art MT systems is not perfect, thus, human interventions are still required to correct machine translated texts into publishable quality translations BIBREF0. Automatic post-editing (APE) is a method that aims to automatically correct errors made by MT systems before perform... |
1802.00273 | Emerging Language Spaces Learned From Massively Multilingual Corpora | Translations capture important information about languages that can be used as implicit supervision in learning linguistic properties and semantic representations. In an information-centric view, translated texts may be considered as semantic mirrors of the original text and the significant variations that we can obser... | {
"section_name": [
"Introduction and Motivation",
"Multilingual Neural Machine Translation",
"Experiments and Results",
"Conclusions",
"Acknowledgements"
],
"paragraphs": [
[
"Our primary goal is to learn meaning representations of sentences and sentence fragments by looking at the ... | {
"question": [
"What neural machine translation models can learn in terms of transfer learning?"
],
"question_id": [
"41e300acec35252e23f239772cecadc0ea986071"
],
"nlp_background": [
"two"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
... | {
"caption": [
"Fig. 1. Conceptual illustrations of neural machine translation and abstractions to meaning representations.",
"Fig. 2. Multilingual Neural MT and training data with language flags.",
"Fig. 3. Experimental setup: Bible translations paired with English as either source or target language are... | Introduction and Motivation
Our primary goal is to learn meaning representations of sentences and sentence fragments by looking at the distributional information that is available in parallel corpora of human translations. The basic idea is to use translations into other languages as “semantic mirrors” of the original ... |
1911.03514 | An Annotation Scheme of A Large-scale Multi-party Dialogues Dataset for Discourse Parsing and Machine Comprehension | In this paper, we propose the scheme for annotating large-scale multi-party chat dialogues for discourse parsing and machine comprehension. The main goal of this project is to help understand multi-party dialogues. Our dataset is based on the Ubuntu Chat Corpus. For each multi-party dialogue, we annotate the discourse ... | {
"section_name": [
"Introduction",
"Ubuntu Corpus",
"Annotation for discourse parsing in multi-party dialogues",
"Annotation for discourse parsing in multi-party dialogues ::: Edges between utterances",
"Annotation for discourse parsing in multi-party dialogues ::: Sense of discourse relations",
... | {
"question": [
"Did they experiment on the proposed task?",
"Is annotation done manually?",
"How large is the proposed dataset?"
],
"question_id": [
"e70236c876c94dbecd9a665d9ba8cefe7301dcfd",
"aa1f605619b2487cc914fc2594c8efe2598d8555",
"9f2634c142dc4ad2c68135dbb393ecdfd23af13f"
],
"n... | {
"caption": [
"Figure 1: The discourse dependency structure and relations for Example 1.",
"Table 1: Overview of STAC corpus."
],
"file": [
"3-Figure1-1.png",
"4-Table1-1.png"
]
} | Introduction
There are more and more NLP scholars focusing on the research of multi-party dialogues, such as multi-party dialogues discourse parsing and multi-party meeting summarization BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5. However, the scale of the STAC dataset has limited the research of discourse pa... |
1805.05581 | Unsupervised Learning of Style-sensitive Word Vectors | This paper presents the first study aimed at capturing stylistic similarity between words in an unsupervised manner. We propose extending the continuous bag of words (CBOW) model (Mikolov et al., 2013) to learn style-sensitive word vectors using a wider context window under the assumption that the style of all the word... | {
"section_name": [
"Introduction",
"Style-sensitive Word Vector",
"Notation",
"Baseline Model (CBOW-near-ctx)",
"Learning Style with Utterance-size Context Window (CBOW-all-ctx)",
"Learning the Style and Syntactic/Semantic Separately",
"Experiments",
"Settings",
"Stylistic Similar... | {
"question": [
"How large is the dataset?",
"How is the dataset created?"
],
"question_id": [
"77e57d19a0d48f46de8cbf857f5e5284bca0df2b",
"50c8b821191339043306fd28e6cda2db400704f9"
],
"nlp_background": [
"two",
"two"
],
"topic_background": [
"familiar",
"familiar"
],
"... | {
"caption": [
"Figure 1: Word vector capturing stylistic and syntactic/semantic similarity.",
"Figure 2: The architecture of CBOW-SEP-CTX.",
"Table 1: Results of the quantitative evaluations.",
"Table 2: The top similar words for the style-sensitive and syntactic/semantic vectors learned with propose... | Introduction
Analyzing and generating natural language texts requires the capturing of two important aspects of language: what is said and how it is said. In the literature, much more attention has been paid to studies on what is said. However, recently, capturing how it is said, such as stylistic variations, has also ... |
1708.00077 | Bayesian Sparsification of Recurrent Neural Networks | Recurrent neural networks show state-of-the-art results in many text analysis tasks but often require a lot of memory to store their weights. Recently proposed Sparse Variational Dropout eliminates the majority of the weights in a feed-forward neural network without significant loss of quality. We apply this technique ... | {
"section_name": [
"Introduction",
"Bayesian Neural Networks",
"Sparse Variational Dropout",
"Dropout for Recurrent Neural Networks",
"Variational Dropout for RNN sparsification",
"Experiments",
"Sentiment Analysis",
"Character Level Language Modeling",
"Regularization of RNNs",
... | {
"question": [
"What is binary variational dropout?"
],
"question_id": [
"dee7383a92c78ea49859a2d5ff2a9d0a794c1f0f"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"50d8b4a94... | {
"caption": [
"Table 1. Results on sentiment regression task. Prediction quality is reported in MSE (the lower the better). Sparsity levels reported for W x and Wh separately in percents of zero weights. For Sparse VD methods initialization types are reported in brackets.",
"Table 2. Results on character lev... | Introduction
Recurrent neural networks (RNNs) are among the most powerful models for natural language processing, speech recognition, question-answering systems and other problems with sequential data BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 , BIBREF6 . For complex tasks such as machine translation BIBREF5 or speech recog... |
1902.07285 | Towards a Robust Deep Neural Network in Text Domain A Survey | Deep neural networks (DNNs) have shown an inherent vulnerability to adversarial examples which are maliciously crafted on real examples by attackers, aiming at making target DNNs misbehave. The threats of adversarial examples are widely existed in image, voice, speech, and text recognition and classification. Inspired ... | {
"section_name": [
"Introduction",
"Background",
"Adversarial Example Formulation",
"Types of Adversarial Attack",
"Metric",
"Datasets in Text",
"Adversarial Attacks in Text",
"Non-target attacks for classification",
"Target attacks for classification",
"Adversarial examples o... | {
"question": [
"Which strategies show the most promise in deterring these attacks?"
],
"question_id": [
"a458c649a793588911cef4c421f95117d0b9c472"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"que... | {
"caption": [],
"file": []
} | Introduction
Nowadays, DNNs have solved masses of significant practical problems in various areas like computer vision BIBREF0 , BIBREF1 , audio BIBREF2 , BIBREF3 , natural language processing (NLP) BIBREF4 , BIBREF5 etc. Due to the great success, systems based on DNN are widely deployed in physical world, including so... |
1912.01679 | Deep Contextualized Acoustic Representations For Semi-Supervised Speech Recognition | We propose a novel approach to semi-supervised automatic speech recognition (ASR). We first exploit a large amount of unlabeled audio data via representation learning, where we reconstruct a temporal slice of filterbank features from past and future context frames. The resulting deep contextualized acoustic representat... | {
"section_name": [
"Introduction",
"Related work",
"DEep COntextualized Acoustic Representations ::: Representation learning from unlabeled data",
"DEep COntextualized Acoustic Representations ::: End-to-end ASR training with labeled data",
"Experimental Setup ::: Data",
"Experimental Setup :... | {
"question": [
"What are baseline models on WSJ eval92 and LibriSpeech test-clean?"
],
"question_id": [
"04cab3325e20c61f19846674bf9a2c46ea60c449"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"qu... | {
"caption": [
"Fig. 1. Illustration of our semi-supervised speech recognition system.",
"Table 1. Semi-supervised LibriSpeech results.",
"Table 2. Semi-supervised WSJ results. Unlabeled indicates the amount of unlabeled data used for acoustic representation learning, and Labeled indicates the amount of l... | Introduction
Current state-of-the-art models for speech recognition require vast amounts of transcribed audio data to attain good performance. In particular, end-to-end ASR models are more demanding in the amount of training data required when compared to traditional hybrid models. While obtaining a large amount of lab... |
1705.00861 | Deep Neural Machine Translation with Linear Associative Unit | Deep Neural Networks (DNNs) have provably enhanced the state-of-the-art Neural Machine Translation (NMT) with their capability in modeling complex functions and capturing complex linguistic structures. However NMT systems with deep architecture in their encoder or decoder RNNs often suffer from severe gradient diffusio... | {
"section_name": [
"Introduction",
"Neural machine translation",
"Model Description",
"Recurrent Layers",
"Gated Recurrent Unit",
"Training details",
"Results on Chinese-English Translation",
"Results on English-German Translation",
"Results on English-French Translation",
"An... | {
"question": [
"Do they use the same architecture as LSTM-s and GRUs with just replacing with the LAU unit?"
],
"question_id": [
"76c8aac84152fc4bbc0d5faa7b46e40438353e77"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"searc... | {
"caption": [
"Figure 1: DEEPLAU: a neural machine translation model with deep encoder and decoder.",
"Table 1: Case-insensitive BLEU scores on Chinese-English translation.",
"Table 2: Case-sensitive BLEU scores on German-English translation.",
"Table 4: BLEU scores of DEEPLAU and DEEPGRU with differ... | Introduction
Neural Machine Translation (NMT) is an end-to-end learning approach to machine translation which has recently shown promising results on multiple language pairs BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 . Unlike conventional Statistical Machine Translation (SMT) systems ... |
1902.00756 | Graph Neural Networks with Generated Parameters for Relation Extraction | Recently, progress has been made towards improving relational reasoning in machine learning field. Among existing models, graph neural networks (GNNs) is one of the most effective approaches for multi-hop relational reasoning. In fact, multi-hop relational reasoning is indispensable in many natural language processing ... | {
"section_name": [
"Introduction",
"Graph Neural Networks (GNNs)",
"Relational Reasoning",
"Graph Neural Network with Generated Parameters (GP-GNNs)",
"Encoding Module",
"Propagation Module",
"Classification Module",
"Relation Extraction with GP-GNNs",
"Experiments",
"Experime... | {
"question": [
"So this paper turns unstructured text inputs to parameters that GNNs can read?"
],
"question_id": [
"6916596253d67f74dba9222f48b9e8799581bad9"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""... | {
"caption": [
"Figure 1: An example of relation extraction from plain text. Given a sentence with several entities marked, we model the interaction between these entities by generating the weights of graph neural networks. Modeling the relationship between “Léon” and “English” as well as “Luc Besson” helps disc... | Introduction
Recent years, graph neural networks (GNNs) have been applied to various fields of machine learning, including node classification BIBREF0 , relation classification BIBREF1 , molecular property prediction BIBREF2 , few-shot learning BIBREF3 , and achieve promising results on these tasks. These works have de... |
1602.03661 | On the emergence of syntactic structures: quantifying and modelling duality of patterning | The complex organization of syntax in hierarchical structures is one of the core design features of human language. Duality of patterning refers for instance to the organization of the meaningful elements in a language at two distinct levels: a combinatorial level where meaningless forms are combined into meaningful fo... | {
"section_name": [
"Introduction",
"Quantifying duality of patterning",
"Modelling the emergence of duality of patterning",
"Discussion and perspectives",
"Acknowledgements"
],
"paragraphs": [
[
"In a seminal paper, Charles Hockett BIBREF0 identified duality of patterning as one of ... | {
"question": [
"What other models are compared to the Blending Game?",
"What empirical data are the Blending Game predictions compared to?"
],
"question_id": [
"7ccf2392422b44ede35a3fbd85bbb1da25adf795",
"4d60e9494a412d581bd5e85f4e78881914085afc"
],
"nlp_background": [
"five",
"five"
... | {
"caption": [
"Figure 1 . Left The normalized histogram of the lengths for the words of the six lexica considered. In the inset, a zoom of the region with percentages above the 5%. Center The phoneme frequency-rank plot for the words of the considered lexica. On the x-axis we report the rank of individual phonem... | Introduction
In a seminal paper, Charles Hockett BIBREF0 identified duality of patterning as one of the core design features of human language. A language exhibits duality of patterning when it is organized at two distinct levels. At a first level, meaningless forms (typically referred to as phonemes) are combined into... |
2003.09520 | TArC: Incrementally and Semi-Automatically Collecting a Tunisian Arabish Corpus | This article describes the constitution process of the first morpho-syntactically annotated Tunisian Arabish Corpus (TArC). Arabish, also known as Arabizi, is a spontaneous coding of Arabic dialects in Latin characters and arithmographs (numbers used as letters). This code-system was developed by Arabic-speaking users ... | {
"section_name": [
"Introduction",
"Related Work",
"Characteristics of Tunisian Arabic and Tunisian Arabish",
"Characteristics of Tunisian Arabic and Tunisian Arabish ::: Tunisian Arabic",
"Characteristics of Tunisian Arabic and Tunisian Arabish ::: Tunisian Arabish",
"Tunisian Arabish Corpus... | {
"question": [
"How does the semi-automatic construction process work?",
"Does the paper report translation accuracy for an automatic translation model for Tunisian to Arabish words?"
],
"question_id": [
"cf63a4f9fe0f71779cf5a014807ae4528279c25a",
"8829f738bcdf05b615072724223dbd82463e5de6"
],
... | {
"caption": [
"Table 1: Arabish code-system for TUN",
"Table 2: Example of the fifteen thematic categories",
"Table 3: An Excerpt of the TArC structure. In the column Var, \"Bnz\" stands for \"Bizerte\" a northern city in Tunisia. Glosses: w1:how, w2:do you(pl) see, w3-4:the life, w5-6:at the, w7:outside... | Introduction
Arabish is the romanization of Arabic Dialects (ADs) used for informal messaging, especially in social networks. This writing system provides an interesting ground for linguistic research, computational as well as sociolinguistic, mainly due to the fact that it is a spontaneous representation of the ADs, a... |
1807.09000 | Speakers account for asymmetries in visual perspective so listeners don't have to | Debates over adults' theory of mind use have been fueled by surprising failures of visual perspective-taking in simple communicative tasks. Motivated by recent computational models of context-sensitive language use, we reconsider the evidence in light of the nuanced Gricean pragmatics of these tasks: the differential i... | {
"section_name": [
"Introduction",
"Experiment 1: Speaker behavior under uncertainty",
"Methods",
"Behavioral results",
"Model comparison",
"Experiment 2: Comparing confederates to natural speakers",
"Results",
"General Discussion",
"Acknowledgements",
"Author contributions",
... | {
"question": [
"Did participants behave unexpectedly?",
"Was this experiment done in a lab?"
],
"question_id": [
"4b624064332072102ea674254d7098038edad572",
"65ba7304838eb960e3b3de7c8a367d2c2cd64c54"
],
"nlp_background": [
"infinity",
"infinity"
],
"topic_background": [
"unfam... | {
"caption": [
"Figure 1: Critical trial of director-matcher task using the ambiguous utterance “the tape”: a cassette tape is in view of both players, but a roll of tape is occluded from the speaker’s view.",
"Figure 2: Design used in Exp. 1 (from speaker’s view; grey square indicates target).",
"Figure ... | Introduction
Our success as a social species depends on our ability to understand, and be understood by, different communicative partners across different contexts. Theory of mind—the ability to represent and reason about others' mental states—is considered to be the key mechanism that supports such context-sensitivity... |
2001.01798 | Domain Adaptation via Teacher-Student Learning for End-to-End Speech Recognition | Teacher-student (T/S) has shown to be effective for domain adaptation of deep neural network acoustic models in hybrid speech recognition systems. In this work, we extend the T/S learning to large-scale unsupervised domain adaptation of an attention-based end-to-end (E2E) model through two levels of knowledge transfer:... | {
"section_name": [
"Introduction",
"Attention-Based Encoder-Decoder (AED) Model",
"T/S Learning for Unsupervised Domain Adaptation of AED",
"Adaptive T/S (AT/S) Learning for Supervised Domain Adaptation of AED",
"Experiments",
"Experiments ::: Data Preparation",
"Experiments ::: AED Basel... | {
"question": [
"How long is new model trained on 3400 hours of data?"
],
"question_id": [
"a60030cfd95d0c10b1f5116c594d50cb96c87ae6"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer"... | {
"caption": [
"Fig. 1. T/S learning for unsupervised domain adaptation of AED model for E2E ASR. The two orange lines signify the two-level knowledge transfer.",
"Fig. 2. Adaptive T/S (AT/S) learning for supervised domain adaptation of AED model for E2E ASR.",
"Table 1. The ASR WER (%) of far-field AEDs ... | Introduction
Recently, with the advancement of deep learning, great progress has been made in end-to-end (E2E) automatic speech recognition (ASR). With the goal of directly mapping a sequence of speech frames to a sequence of output tokens, an E2E ASR system incorporates the acoustic model, language model and pronuncia... |
1901.03866 | HAS-QA: Hierarchical Answer Spans Model for Open-domain Question Answering | This paper is concerned with open-domain question answering (i.e., OpenQA). Recently, some works have viewed this problem as a reading comprehension (RC) task, and directly applied successful RC models to it. However, the performances of such models are not so good as that in the RC task. In our opinion, the perspectiv... | {
"section_name": [
"Introduction",
"Related Works",
"Probabilistic Views of OpenQA",
"HAS-QA Model",
"Question Aware Context Encoder",
"Conditional Span Predictor",
"Multiple Spans Aggregator",
"Paragraph Quality Estimator",
"Datasets",
"Experimental Settings",
"Overall Re... | {
"question": [
"How much does HAS-QA improve over baselines?"
],
"question_id": [
"efe49829725cfe54de01405c76149a4fe4d18747"
],
"nlp_background": [
"two"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"50... | {
"caption": [
"Figure 1: Examples of RC task and OpenQA task.",
"Figure 2: The three hierarchical levels of OpenQA task.",
"Table 1: The negative paragraph ratio and average answer span count are statistic on three datasets, in order to illustrate the problems mentioned above in OpenQA task.",
"Figur... | Introduction
Open-domain question answering (OpenQA) aims to seek answers for a broad range of questions from a large knowledge sources, e.g., structured knowledge bases BIBREF0 , BIBREF1 and unstructured documents from search engine BIBREF2 . In this paper we focus on the OpenQA task with the unstructured knowledge so... |
1706.02027 | Question Answering and Question Generation as Dual Tasks | We study the problem of joint question answering (QA) and question generation (QG) in this paper. Our intuition is that QA and QG have intrinsic connections and these two tasks could improve each other. On one side, the QA model judges whether the generated question of a QG model is relevant to the answer. On the other... | {
"section_name": [
"Introduction",
"The Proposed Framework",
"Task Definition and Notations",
"Algorithm Description",
"Relationships with Existing Studies",
"The Question Answering Model",
"The Question Generation Model",
"Experiment",
"Experimental Setting",
"Implementation ... | {
"question": [
"What does \"explicitly leverages their probabilistic correlation to guide the training process of both models\" mean?"
],
"question_id": [
"3d49b678ff6b125ffe7fb614af3e187da65c6f65"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
... | {
"caption": [
"Table 1: Statistics of the MARCO, SQUAD and WikiQA datasets for answer sentence selection.",
"Table 2: QA Performance on the MARCO and SQUAD datasets.",
"Table 3: QA performance on the WikiQA dataset.",
"Table 4: Sampled examples from the SQUAD dataset.",
"Table 5: QG performance (... | Introduction
Question answering (QA) and question generation (QG) are two fundamental tasks in natural language processing BIBREF0 , BIBREF1 . Both tasks involve reasoning between a question sequence $q$ and an answer sentence $a$ . In this work, we take answer sentence selection BIBREF2 as the QA task, which is a fund... |
1704.08424 | Multimodal Word Distributions | Word embeddings provide point representations of words containing useful semantic information. We introduce multimodal word distributions formed from Gaussian mixtures, for multiple word meanings, entailment, and rich uncertainty information. To learn these distributions, we propose an energy-based max-margin objective... | {
"section_name": [
"Introduction",
"Related Work",
"Methodology",
"Word Representation",
"Skip-Gram",
"Energy-based Max-Margin Objective",
"Energy Function",
"Experiments",
"Hyperparameters",
"Similarity Measures",
"Qualitative Evaluation",
"Word Similarity",
"Word... | {
"question": [
"How does this compare to contextual embedding methods?"
],
"question_id": [
"b686e10a725254695821e330a277c900792db69f"
],
"nlp_background": [
"two"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
"multimodal"
],
"questio... | {
"caption": [
"Figure 1: Top: A Gaussian Mixture embedding, where each component corresponds to a distinct meaning. Each Gaussian component is represented by an ellipsoid, whose center is specified by the mean vector and contour surface specified by the covariance matrix, reflecting subtleties in meaning and unc... | Introduction
To model language, we must represent words. We can imagine representing every word with a binary one-hot vector corresponding to a dictionary position. But such a representation contains no valuable semantic information: distances between word vectors represent only differences in alphabetic ordering. Mode... |
1702.06700 | Task-driven Visual Saliency and Attention-based Visual Question Answering | Visual question answering (VQA) has witnessed great progress since May, 2015 as a classic problem unifying visual and textual data into a system. Many enlightening VQA works explore deep into the image and question encodings and fusing methods, of which attention is the most effective and infusive mechanism. Current at... | {
"section_name": [
"Introduction",
"Saliency Detection Modeling",
"Attention in VQA Models",
"Proposed Method",
"Model",
"Training",
"Experiments",
"Implementation Details",
"Datasets",
"Compared Models",
"Results and Analysis",
"Conclusion"
],
"paragraphs": [
... | {
"question": [
"Does the new system utilize pre-extracted bounding boxes and/or features?",
"To which previous papers does this work compare its results?"
],
"question_id": [
"40f87db3a8d1ac49b888ce3358200f7d52903ce7",
"36383971a852d1542e720d3ea1f5adeae0dbff18"
],
"nlp_background": [
"two... | {
"caption": [
"Figure 1. (Best view in color and zoom in.) The flow of our proposed VQA model. Q and A represent question and answer related to the image. EWM denotes element-wise multiplication operation.",
"Figure 2. (Best view in color and zoom in.) Frameworks of our proposed VQA model. Bars with frame li... | Introduction
Visual question answering (VQA) comes as a classic task which combines visual and textual modal data into a unified system. Taking an image and a natural language question about it as input, a VQA system is supposed to output the corresponding natural language answer. VQA problem requires image and text un... |
1909.02151 | KagNet: Knowledge-Aware Graph Networks for Commonsense Reasoning | Commonsense reasoning aims to empower machines with the human ability to make presumptions about ordinary situations in our daily life. In this paper, we propose a textual inference framework for answering commonsense questions, which effectively utilizes external, structured commonsense knowledge graphs to perform exp... | {
"section_name": [
"Introduction",
"Overview",
"Problem statement",
"Reasoning Workflow",
"Schema Graph Grounding",
"Concept Recognition",
"Schema Graph Construction",
"Path Pruning via KG Embedding",
"Knowledge-Aware Graph Network",
"Graph Convolutional Networks",
"Relati... | {
"question": [
"Do they consider other tasks?"
],
"question_id": [
"1d941d390c0ee365aa7d7c58963e646eea74cbd6"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"research"
],
"paper_read": [
"somewhat"
],
"search_query": [
"commonsense"
],
"question_writer": [
... | {
"caption": [
"Figure 1: An example of using external commonsense",
"Figure 2: The overall workflow of the proposed framework with knowledge-aware graph network module.",
"Figure 3: Illustration of the GCN-LSTM-HPA architecture for the proposed KAGNET module.",
"Table 1: Comparisons with large pre-tr... | Introduction
Human beings are rational and a major component of rationality is the ability to reason. Reasoning is the process of combining facts and beliefs to make new decisions BIBREF0 , as well as the ability to manipulate knowledge to draw inferences BIBREF1 . Commonsense reasoning utilizes the basic knowledge tha... |
2003.04973 | Localized Flood DetectionWith Minimal Labeled Social Media Data Using Transfer Learning | Social media generates an enormous amount of data on a daily basis but it is very challenging to effectively utilize the data without annotating or labeling it according to the target application. We investigate the problem of localized flood detection using the social sensing model (Twitter) in order to provide an eff... | {
"section_name": [
"Introduction",
"Related Work",
"Data Collection and Processing",
"Methodology",
"Methodology ::: Universal Language Model Fine-tuning (ULMFiT)",
"Methodology ::: ULMFiT adaptation for Flood Tweet Classification",
"Experimental Results and Discussion",
"Limitation a... | {
"question": [
"What were the model's results on flood detection?",
"What dataset did they use?"
],
"question_id": [
"3ee976add83e37339715d4ae9d8aa328dd54d052",
"ef04182b6ae73a83d52cb694cdf4d414c81bf1dc"
],
"nlp_background": [
"",
""
],
"topic_background": [
"unfamiliar",
... | {
"caption": [],
"file": []
} | Introduction
There are various forms of a natural disaster such as flood, earthquake, volcano eruptions, storms, etc. but the flood is one of the lethal and prominent forms of natural disaster according to World Meteorological Organization (WMO) for most of the countries. National Weather Services (NWS) reported 28,826... |
1610.01030 | Applications of Online Deep Learning for Crisis Response Using Social Media Information | During natural or man-made disasters, humanitarian response organizations look for useful information to support their decision-making processes. Social media platforms such as Twitter have been considered as a vital source of useful information for disaster response and management. Despite advances in natural language... | {
"section_name": [
"Introduction",
"Deep Neural Network",
"Convolutional Neural Network",
"Online Learning",
"Word Embedding and Fine-tuning",
"Dataset and Experimental Settings",
"Dataset and Preprocessing",
"Online Training Settings",
"Results",
"Binary Classification",
... | {
"question": [
"What exactly is new about this stochastic gradient descent algorithm?"
],
"question_id": [
"decb07f9be715de024236e50dc7011a132363480"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
"social"... | {
"caption": [
"Figure 1: Convolutional neural network on a sample tweet: “guys if know any medical emergency around balaju area you can reach umesh HTTP doctor at HTTP HTTP”.",
"Table 1: Description of the classes in the dataset. Column Labels shows the total number of labeled examples in each class",
"F... | Introduction
Emergency events such as natural or man-made disasters bring unique challenges for humanitarian response organizations. Particularly, sudden-onset crisis situations demand officials to make fast decisions based on minimum information available to deploy rapid crisis response. However, information scarcity ... |
1909.00100 | Small and Practical BERT Models for Sequence Labeling | We propose a practical scheme to train a single multilingual sequence labeling model that yields state of the art results and is small and fast enough to run on a single CPU. Starting from a public multilingual BERT checkpoint, our final model is 6x smaller and 27x faster, and has higher accuracy than a state-of-the-ar... | {
"section_name": [
"Multilingual Models for Sequence Labeling",
"Meta-LSTM",
"Multilingual BERT",
"Small and Practical Models",
"Size and speed",
"Distillation",
"Data",
"Tuning",
"Multilingual Models",
"Low Resource Languages",
"Codemixed Input",
"Conclusion",
"Tr... | {
"question": [
"What codemixed language pairs are evaluated?",
"How do they compress the model?",
"What is the multilingual baseline?"
],
"question_id": [
"63eb31f613a41a3ddd86f599e743ed10e1cd07ba",
"d2804ac0f068e9c498e33582af9c66906b26cac3",
"e24fbcc8be922c43f6b6037cdf2bfd4c0a926c08"
]... | {
"caption": [
"Table 1: Macro-averaged F1 comparison of per-language models and multilingual models over 48 languages. For non-multilingual models, F1 is the average over each per-language model trained.",
"Table 2: The number of parameters of each model. Tokens refers to the number of tokens of the embeddin... | Multilingual Models for Sequence Labeling
We discuss two core models for addressing sequence labeling problems and describe, for each, training them in a single-model multilingual setting: (1) the Meta-LSTM BIBREF0 , an extremely strong baseline for our tasks, and (2) a multilingual BERT-based model BIBREF1 .
Meta-LST... |
1711.05568 | Dialogue Act Recognition via CRF-Attentive Structured Network | Dialogue Act Recognition (DAR) is a challenging problem in dialogue interpretation, which aims to attach semantic labels to utterances and characterize the speaker's intention. Currently, many existing approaches formulate the DAR problem ranging from multi-classification to structured prediction, which suffer from han... | {
"section_name": [
"Introduction",
"CRF-attentive Structured Network ",
"The problem",
"Hierarchical Semantic Network",
"Structured CRF-Attention Network",
"End-to-End Training",
"Experiments",
"Data Preparation",
"Evaluation Criteria",
"Implemental Details",
"Performance ... | {
"question": [
"Which features do they use?",
"By how much do they outperform state-of-the-art solutions on SWDA and MRDA?"
],
"question_id": [
"e8c0fabae0d29491471e37dec34f652910302928",
"cafa6103e609acaf08274a2f6d8686475c6b8723"
],
"nlp_background": [
"infinity",
"infinity"
],
"... | {
"caption": [
"Table 1: A snippet of a conversation sample. Each utterance has related dialogue act label.",
"Figure 1: An illustration of the hierarchical conversation structure. The input of the model is a conversation which consist of n utterances u1,u2, ...,un with corresponding dialogue act labels a1,a2... | Introduction
Dialogue Act Recognition (DAR) is an essential problem in modeling and detecting discourse structure. The goal of DAR is to attach semantic labels to each utterance in a conversation and recognize the speaker's intention, which can be regarded as a sequence labeling task. Many applications have benefited f... |
1810.08732 | Named Entity Recognition on Twitter for Turkish using Semi-supervised Learning with Word Embeddings | Recently, due to the increasing popularity of social media, the necessity for extracting information from informal text types, such as microblog texts, has gained significant attention. In this study, we focused on the Named Entity Recognition (NER) problem on informal text types for Turkish. We utilized a semi-supervi... | {
"section_name": [
"Introduction",
"Related Work",
"NER for Turkish Tweets using Semi-supervised Learning",
"Unsupervised Stage",
"Supervised Stage",
"Unlabeled Data",
"Labeled Data",
"Experiments and Results",
"NER Models Trained on News",
"NER Models Trained on Tweets",
... | {
"question": [
"What type and size of word embeddings were used?",
"What data was used to build the word embeddings?"
],
"question_id": [
"7f2fd7ab968de720082133c42c2052d351589a67",
"369b0a481a4b75439ade0ec4f12b44414c4e5164"
],
"nlp_background": [
"five",
"five"
],
"topic_backgrou... | {
"caption": [
"Figure 1: Skip-grammodel architecture to learn continuous vector representation of words in order to predict surrounding words (Mikolov et al., 2013).",
"Table 2: Phrase-level overall F-score performance results of the NER models trained on news.",
"Table 1: Number of PLOs in Turkish Twitt... | Introduction
Microblogging environments, which allow users to post short messages, have gained increased popularity in the last decade. Twitter, which is one of the most popular microblogging platforms, has become an interesting platform for exchanging ideas, following recent developments and trends, or discussing any ... |
1906.05012 | BiSET: Bi-directional Selective Encoding with Template for Abstractive Summarization | The success of neural summarization models stems from the meticulous encodings of source articles. To overcome the impediments of limited and sometimes noisy training data, one promising direction is to make better use of the available training data by applying filters during summarization. In this paper, we propose a ... | {
"section_name": [
"Introduction",
"The Framework",
"Retrieve",
"Fast Rerank",
"Traditional Methodologies",
"BiSET",
"Training",
"Experiments",
"Dataset and Implementation",
"Evaluation Metrics",
"Results and Analysis",
"Performance of Retrieve",
"Interaction Appro... | {
"question": [
"How are templates discovered from training data?"
],
"question_id": [
"e97545f4a5e7bc96515e60f2f9b23d8023d1eed9"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer"... | {
"caption": [
"Figure 1: Overview of the Fast Rerank Module.",
"Figure 2: The structure of the proposed model: (a) the Bi-Directional Selective Encoding with Template model (BiSET) and (b) the bi-directional selective layer.",
"Figure 3: Quality of candidate templates under different ranges.",
"Figur... | Introduction
Abstractive summarization aims to shorten a source article or paragraph by rewriting while preserving the main idea. Due to the difficulties in rewriting long documents, a large body of research on this topic has focused on paragraph-level article summarization. Among them, sequence-to-sequence models have... |
1910.07134 | Efficiency through Auto-Sizing: Notre Dame NLP's Submission to the WNGT 2019 Efficiency Task | This paper describes the Notre Dame Natural Language Processing Group's (NDNLP) submission to the WNGT 2019 shared task (Hayashi et al., 2019). We investigated the impact of auto-sizing (Murray and Chiang, 2015; Murray et al., 2019) to the Transformer network (Vaswani et al., 2017) with the goal of substantially reduci... | {
"section_name": [
"Introduction",
"Auto-sizing",
"Auto-sizing the Transformer",
"Experiments",
"Experiments ::: Settings",
"Experiments ::: Auto-sizing sub-components",
"Experiments ::: Results",
"Conclusion",
"Acknowledgements"
],
"paragraphs": [
[
"The Transformer... | {
"question": [
"What is WNGT 2019 shared task?"
],
"question_id": [
"aaed6e30cf16727df0075b364873df2a4ec7605b"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"258ee4069f740... | {
"caption": [
"Figure 1: Architecture of the Transformer (Vaswani et al., 2017). We apply the auto-sizing method to the feed-forward (blue rectangles) and multi-head attention (orange rectangles) in all N layers of the encoder and decoder. Note that there are residual connections that can allow information and g... | Introduction
The Transformer network BIBREF3 is a neural sequence-to-sequence model that has achieved state-of-the-art results in machine translation. However, Transformer models tend to be very large, typically consisting of hundreds of millions of parameters. As the number of parameters directly corresponds to second... |
1606.00189 | Neural Network Translation Models for Grammatical Error Correction | Phrase-based statistical machine translation (SMT) systems have previously been used for the task of grammatical error correction (GEC) to achieve state-of-the-art accuracy. The superiority of SMT systems comes from their ability to learn text transformations from erroneous to corrected text, without explicitly modelin... | {
"section_name": [
"Introduction",
"Related Work",
"A Machine Translation Framework for Grammatical Error Correction",
"Neural Network Global Lexicon Model ",
"Model",
"Training",
"Rescaling",
"Neural Network Joint Model",
"Experiments",
"Setup",
"Results and Discussion",
... | {
"question": [
"Do they use pretrained word representations in their neural network models?",
"How do they combine the two proposed neural network models?",
"Which dataset do they evaluate grammatical error correction on?"
],
"question_id": [
"66f0dee89f084fe0565539a73f5bbe65f3677814",
"8f882... | {
"caption": [
"Figure 1: A single hidden layer neural network global lexicon model",
"Figure 2: A single hidden layer neural network joint model",
"Table 1: Statistics of training and development data",
"Table 2: Results of our experiments with NNGLM and NNJM on the CoNLL 2014 test set (* indicates s... | Introduction
Grammatical error correction (GEC) is a challenging task due to the variability of the type of errors and the syntactic and semantic dependencies of the errors on the surrounding context. Most of the grammatical error correction systems use classification and rule-based approaches for correcting specific e... |
1909.04493 | Context-aware Deep Model for Entity Recommendation in Search Engine at Alibaba | Entity recommendation, providing search users with an improved experience via assisting them in finding related entities for a given query, has become an indispensable feature of today's search engines. Existing studies typically only consider the queries with explicit entities. They usually fail to handle complex quer... | {
"section_name": [
"Introduction",
"Related Work",
"System Overview",
"Preliminaries",
"Preliminaries ::: Knowledge Graph",
"Preliminaries ::: Cognitive Concept Graph",
"Deep Collaborative Match",
"Deep Collaborative Match ::: Recommendation as Classification",
"Deep Collaborative... | {
"question": [
"How many users/clicks does their search engine have?"
],
"question_id": [
"a33ab5ce8497ff63ca575a80b03e0ed9c6acd273"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_write... | {
"caption": [
"Figure 1: Example of entity recommendation results for the query \"what food is good for cold weather.\"",
"Figure 2: System overview of entity recommendation in ShenMa search engine at Alibaba, the red part is the focus of this paper.",
"Figure 3: Base deep match model.",
"Figure 4: E... | Introduction
Over the past few years, major commercial search engines have enriched and improved the user experience by proactively presenting related entities for a query along with the regular web search results. Figure FIGREF3 shows an example of Alibaba ShenMa search engine's entity recommendation results presented... |
1608.03902 | Rapid Classification of Crisis-Related Data on Social Networks using Convolutional Neural Networks | The role of social media, in particular microblogging platforms such as Twitter, as a conduit for actionable and tactical information during disasters is increasingly acknowledged. However, time-critical analysis of big crisis data on social media streams brings challenges to machine learning techniques, especially the... | {
"section_name": [
"Introduction",
"Related Work",
"Convolutional Neural Network",
"Word Embedding and Fine-tuning",
"Incorporating Other Features",
"Experimental Settings",
"Datasets",
"Non-neural Model Settings",
"Settings for Convolutional Neural Network",
"Results",
"B... | {
"question": [
"what was their baseline comparison?"
],
"question_id": [
"8fcbae7c3bd85034ae074fa58a35e773936edb5b"
],
"nlp_background": [
"two"
],
"topic_background": [
"familiar"
],
"paper_read": [
"somewhat"
],
"search_query": [
"social"
],
"question_writer": [
... | {
"caption": [
"Figure 1: Convolutional neural network on a tweet.",
"Table I: Description of the classes in the datasets. Column Labels shows the total number of annotations for each class",
"Table II: Class distribution of events under consideration and all other crises (i.e. data used as part of out-of... | Introduction
Time-critical analysis of social media data streams is important for many application areas. For instance, responders to humanitarian disasters (e.g., earthquake, flood) need information about the disasters to determine what help is needed and where. This information usually breaks out on social media befo... |
2004.03061 | Information-Theoretic Probing for Linguistic Structure | The success of neural networks on a diverse set of NLP tasks has led researchers to question how much do these networks actually know about natural language. Probes are a natural way of assessing this. When probing, a researcher chooses a linguistic task and trains a supervised model to predict annotation in that lingu... | {
"section_name": [
"Introduction",
"Word-Level Syntactic Probes for Contextual Embeddings",
"Word-Level Syntactic Probes for Contextual Embeddings ::: Notation",
"Word-Level Syntactic Probes for Contextual Embeddings ::: Probing as Mutual Information",
"Word-Level Syntactic Probes for Contextual ... | {
"question": [
"Was any variation in results observed based on language typology?",
"Does the work explicitly study the relationship between model complexity and linguistic structure encoding?"
],
"question_id": [
"cbbcafffda7107358fa5bf02409a01e17ee56bfd",
"1e59263f7aa7dd5acb53c8749f627cf68683ad... | {
"caption": [
"Table 1: Amount of information BERT, fastText or one-hot embeddings share with a POS probing task. H(T ) is estimated with a plug-in estimator from same treebanks we use to train the POS labelers.",
"Table 2: Amount of information BERT, fastText or one-hot embeddings share with a dependency ar... | Introduction
Neural networks are the backbone of modern state-of-the-art Natural Language Processing (NLP) systems. One inherent by-product of training a neural network is the production of real-valued representations. Many speculate that these representations encode a continuous analogue of discrete linguistic propert... |
1601.04012 | Detecting and Extracting Events from Text Documents | Events of various kinds are mentioned and discussed in text documents, whether they are books, news articles, blogs or microblog feeds. The paper starts by giving an overview of how events are treated in linguistics and philosophy. We follow this discussion by surveying how events and associated information are handled... | {
"section_name": [
"Introduction",
"Events in Linguistics and Philosophy",
"Classifying Events",
"Parameters of Event Classes",
"Events in Logical Representation of Semantics",
" Event structure",
"Semantic Arguments and Syntactic Positions",
"Lexical Resources for Action or Event Rep... | {
"question": [
"Which datasets are used in this work?"
],
"question_id": [
"eac042734f76e787cb98ba3d0c13a916a49bdfb3"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"research"
],
"paper_read": [
"no"
],
"search_query": [
"events"
],
"question_writer": [
... | {
"caption": [
"Table I. Nomenclatures used by linguists to classify events",
"Table II. Features used by linguists to classify events",
"Fig. 2. Sample Comlex Subcategorization Frames",
"Table III. PropBank entry for attack.01, a sense of the verb attack",
"Table IV. Roles for the verbs purchase,... | Introduction
Among the several senses that The Oxford English Dictionary, the most venerable dictionary of English, provides for the word event, are the following.
Although an event may refer to anything that happens, we are usually interested in occurrences that are of some importance. We want to extract such events... |
1909.00574 | A Sketch-Based System for Semantic Parsing | This paper presents our semantic parsing system for the evaluation task of open domain semantic parsing in NLPCC 2019. Many previous works formulate semantic parsing as a sequence-to-sequence(seq2seq) problem. Instead, we treat the task as a sketch-based problem in a coarse-to-fine(coarse2fine) fashion. The sketch is a... | {
"section_name": [
"Introduction",
"Data Analyzation",
"Proposed Approach ::: Sketch Classification",
"Proposed Approach ::: Entity Labeling",
"Proposed Approach ::: Multi-Task Model",
"Proposed Approach ::: Pattern Pair Matching Network",
"Proposed Approach ::: Predicate-Entity Pair Matc... | {
"question": [
"Does the training dataset provide logical form supervision?",
"What is the difference between the full test set and the hard test set?"
],
"question_id": [
"9595bf228c9e859b0dc745e6c74070be2468d2cf",
"94c5f5b1eb8414ad924c3568cedd81dc35f29c48"
],
"nlp_background": [
"infini... | {
"caption": [
"Table 1. Examples demonstrating sketches of logical forms. P represents predicate and E represents entity. Subscripts are applied to distinguish different ones.",
"Table 2. An sample of MSParS.",
"Table 3. An example for question pattern and logical form pattern.",
"Fig. 1. An overview... | Introduction
Open domain semantic parsing aims to map natural language utterances to structured meaning representations. Recently, seq2seq based approaches have achieved promising performance by structure-aware networks, such as sequence-to-actionBIBREF0 and STAMPBIBREF1.
However, this kind of approach mixes up low-le... |
1707.07048 | Progressive Joint Modeling in Unsupervised Single-channel Overlapped Speech Recognition | Unsupervised single-channel overlapped speech recognition is one of the hardest problems in automatic speech recognition (ASR). Permutation invariant training (PIT) is a state of the art model-based approach, which applies a single neural network to solve this single-input, multiple-output modeling problem. We propose ... | {
"section_name": [
"Introduction",
"Unsupervised Single-channel Overlapped Speech Recognition",
"Methods",
"Modularization",
"Transfer Learning Based Joint Training",
"Multi-output Sequence Discriminative Training",
"Experiment",
"Experimental Setup",
"Separate Optimization v.s. J... | {
"question": [
"How is the discriminative training formulation different from the standard ones?",
"How are the two datasets artificially overlapped?"
],
"question_id": [
"ba05a53f5563b9dd51cc2db241c6e9418bc00031",
"7bf3a7d19f17cf01f2c9fa16401ef04a3bef65d8"
],
"nlp_background": [
"two",
... | {
"caption": [
"Fig. 1. The Proposed System Framework. The single monolithic structure (the dashed line box) that predicts independent targets for each speaker, proposed in [14], is improved through modularization (three solid line boxes) and pretraining. Self-transfer learning and multi-output sequence discrimin... | Introduction
The cocktail party problem BIBREF0 , BIBREF1 , referring to multi-talker overlapped speech recognition, is critical to enable automatic speech recognition (ASR) scenarios such as automatic meeting transcription, automatic captioning for audio/video recordings, and multi-party human-machine interactions, wh... |
1804.00520 | NIHRIO at SemEval-2018 Task 3: A Simple and Accurate Neural Network Model for Irony Detection in Twitter | This paper describes our NIHRIO system for SemEval-2018 Task 3"Irony detection in English tweets". We propose to use a simple neural network architecture of Multilayer Perceptron with various types of input features including: lexical, syntactic, semantic and polarity features. Our system achieves very high performance... | {
"section_name": [
"Introduction",
"Dataset",
"Our modeling approach",
"Neural network model",
"Features",
"Implementation details",
"Metrics",
"Results for subtask 1",
"Results for subtask 2",
"Discussions",
"Conclusion",
"Acknowledgments"
],
"paragraphs": [
[... | {
"question": [
"What baseline system is used?",
"What type of lexical, syntactic, semantic and polarity features are used?"
],
"question_id": [
"20f7b359f09c37e6aaaa15c2cdbb52b031ab4809",
"3efc0981e7f959d916aa8bb32ab1c347b8474ff8"
],
"nlp_background": [
"five",
"five"
],
"topic_ba... | {
"caption": [
"Figure 1: Overview of our model architecture for irony detection in tweets.",
"Table 1: Basic statistics of the provided dataset.",
"Figure 2: The training mechanism.",
"Table 2: Number of features used in our model",
"Table 3: Example of clusters produced by the Brown clustering a... | Introduction
Mining Twitter data has increasingly been attracting much research attention in many NLP applications such as in sentiment analysis BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 and stock market prediction BIBREF6 , BIBREF7 , BIBREF8 , BIBREF9 , BIBREF10 . Recently, Davidov2010 and Reyes2013 ha... |
1901.03859 | What comes next? Extractive summarization by next-sentence prediction | Existing approaches to automatic summarization assume that a length limit for the summary is given, and view content selection as an optimization problem to maximize informativeness and minimize redundancy within this budget. This framework ignores the fact that human-written summaries have rich internal structure whic... | {
"section_name": [
"Introduction",
"Related work",
"NextSum model overview",
"Predicting the next summary sentence",
"Summary generation",
"Implementing NextSum",
"Candidate selection",
"Features for next sentence prediction",
"Data",
"Length of articles and summaries",
"O... | {
"question": [
"How does nextsum work?"
],
"question_id": [
"10f560fe8e1c0c7dea5e308ee4cec16d07874f1d"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
"summarization"
],
"question_writer": [
"50d8b4... | {
"caption": [
"Table 2: Number of article-summary pairs in our data.",
"Table 3: Min, max and average lengths (in words) of source articles and abstracts. Tau is the Kendall Tau correlation between the length of source and abstract.",
"Table 4: Results on next sentence prediction task.",
"Table 5: RO... | Introduction
Writing a summary is a different task compared to producing a longer article. As a consequence, it is likely that the topic and discourse moves made in summaries differ from those in regular articles. In this work, we present a powerful extractive summarization system which exploits rich summary-internal s... |
1704.04521 | Translation of Patent Sentences with a Large Vocabulary of Technical Terms Using Neural Machine Translation | Neural machine translation (NMT), a new approach to machine translation, has achieved promising results comparable to those of traditional approaches such as statistical machine translation (SMT). Despite its recent success, NMT cannot handle a larger vocabulary because training complexity and decoding complexity propo... | {
"section_name": [
"Introduction",
"Japanese-Chinese Patent Documents",
"Neural Machine Translation (NMT)",
"NMT Training after Replacing Technical Term Pairs with Tokens",
"NMT Decoding and SMT Technical Term Translation",
"NMT Rescoring of 1,000-best SMT Translations",
"Training and Tes... | {
"question": [
"Can the approach be generalized to other technical domains as well? "
],
"question_id": [
"07580f78b04554eea9bb6d3a1fc7ca0d37d5c612"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
... | {
"caption": [
"Figure 1: Example of translation errors when translating patent sentences with technical terms using NMT",
"Figure 2: NMT training after replacing technical term pairs with technical term tokens “TTi” (i = 1, 2, . . .)",
"Figure 3: NMT decoding with technical term tokens “TTi” (i = 1, 2, .... | Introduction
Neural machine translation (NMT), a new approach to solving machine translation, has achieved promising results BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 , BIBREF6 . An NMT system builds a simple large neural network that reads the entire input source sentence and generates an output transl... |
1804.02233 | Forex trading and Twitter: Spam, bots, and reputation manipulation | Currency trading (Forex) is the largest world market in terms of volume. We analyze trading and tweeting about the EUR-USD currency pair over a period of three years. First, a large number of tweets were manually labeled, and a Twitter stance classification model is constructed. The model then classifies all the tweets... | {
"section_name": [
"Abstract",
"Introduction",
"Twitter stance model",
"Twitter user groups",
"Event study",
"Reputation manipulation",
"Deleting tweets to increase CARs",
"Analyzing trading companies",
"Conclusions",
"Acknowledgements"
],
"paragraphs": [
[
"Curr... | {
"question": [
"How many tweets were manually labelled? "
],
"question_id": [
"dc28ac845602904c2522f5349374153f378c42d3"
],
"nlp_background": [
"five"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
"twitter"
],
"question_writer": [
... | {
"caption": [
"Table 1. Evaluation results of the Twitter stance model.",
"Fig 1. Proportions of Twitter accounts and tweets for different user groups.",
"Fig 2. Twitter stance distribution of different user groups (bars show the proportion of tweets). Trading robots produce almost exclusively polarized ... | Abstract
Currency trading (Forex) is the largest world market in terms of volume. We analyze trading and tweeting about the EUR-USD currency pair over a period of three years. First, a large number of tweets were manually labeled, and a Twitter stance classification model is constructed. The model then classifies all t... |
1908.08566 | Unsupervised Text Summarization via Mixed Model Back-Translation | Back-translation based approaches have recently lead to significant progress in unsupervised sequence-to-sequence tasks such as machine translation or style transfer. In this work, we extend the paradigm to the problem of learning a sentence summarization system from unaligned data. We present several initial models wh... | {
"section_name": [
"Introduction",
"Related Work",
"Mixed Model Back-Translation",
"Mixed Model Back-Translation ::: Initialization Models for Summarization",
"Mixed Model Back-Translation ::: Initialization Models for Summarization ::: Procrustes Thresholded Alignment (Pr-Thr)",
"Mixed Model... | {
"question": [
"What dataset they use for evaluation?"
],
"question_id": [
"ac148fb921cce9c8e7b559bba36e54b63ef86350"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
"summarization"
],
"question_writer"... | {
"caption": [
"Table 1: Full text sequences generated by f (Pr-Thr),1 S→F , f (DBAE),1 S→F , and f (µ:1),1 S→F during the first back-translation loop.",
"Table 2: Test ROUGE for trivial baseline and initialization systems. 1(Wang and Lee, 2018).",
"Table 3: Comparison of full systems. The best scores for... | Introduction
Machine summarization systems have made significant progress in recent years, especially in the domain of news text. This has been made possible among other things by the popularization of the neural sequence-to-sequence (seq2seq) paradigm BIBREF0, BIBREF1, BIBREF2, the development of methods which combine... |
1708.04120 | Putting Self-Supervised Token Embedding on the Tables | Information distribution by electronic messages is a privileged means of transmission for many businesses and individuals, often under the form of plain-text tables. As their number grows, it becomes necessary to use an algorithm to extract text and numbers instead of a human. Usual methods are focused on regular expre... | {
"section_name": [
"INTRODUCTION",
"Information Extraction on Semi-Structured Data",
"Natural Language Processing",
"THE SC2T EMBEDDING",
"The Architecture",
"Alternative Model",
"Tokens and Lines Clustering",
"EMPIRICAL RESULTS",
"The Dataset",
"Labeling of tokens using the S... | {
"question": [
"What is the source of the tables?"
],
"question_id": [
"094ce2f912aa3ced9eb97b171745d38f58f946dd"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
"semi-structured"
],
"question_writer": [
... | {
"caption": [
"Fig. 1. An example of the type of ASCII table we want to extract, and the target extraction. The goal is to find what each token means, and each color corresponds to a type of token. We see that there are different line patterns, and this is only one type of message among thousands.",
"Fig. 2.... | INTRODUCTION
Today most of business-related information is transmitted in an electronic form, such as emails. Therefore, converting these messages into an easily analyzable representation could open numerous business opportunities, as a lot of them are not used fully because of the difficulty to build bespoke parsing m... |
1803.09745 | English verb regularization in books and tweets | The English language has evolved dramatically throughout its lifespan, to the extent that a modern speaker of Old English would be incomprehensible without translation. One concrete indicator of this process is the movement from irregular to regular (-ed) forms for the past tense of verbs. In this study we quantify the... | {
"section_name": [
"Introduction",
"Description of data sets",
"Verb regularization using Ngrams and Twitter",
"American and British English",
"Regularization by US county",
"Concluding remarks",
"Table of Verb Forms",
"Details on User Location Matching"
],
"paragraphs": [
[
... | {
"question": [
"Which regions of the United States do they consider?",
"Why did they only consider six years of published books?"
],
"question_id": [
"b5bfa6effdeae8ee864d7d11bc5f3e1766171c2d",
"bf00808353eec22b4801c922cce7b1ec0ff3b777"
],
"nlp_background": [
"five",
"five"
],
"to... | {
"caption": [
"Fig 1. Relative word frequencies for the irregular and regular past verb forms for ‘burn’ during the 19th and 20th centuries, using the Google Ngram Online Viewer with the English Fiction 2012 corpus. Google Ngram trends can be misleading but capture basic shifts in a language’s lexicon [7, 8]. Th... | Introduction
Human language reflects cultural, political, and social evolution. Words are the atoms of language. Their meanings and usage patterns reveal insight into the dynamical process by which society changes. Indeed, the increasing frequency with which electronic text is used as a means of communicating, e.g., th... |
1910.03771 | HuggingFace's Transformers: State-of-the-art Natural Language Processing | Recent advances in modern Natural Language Processing (NLP) research have been dominated by the combination of Transfer Learning methods with large-scale language models, in particular based on the Transformer architecture. With them came a paradigm shift in NLP with the starting point for training a model on a downstr... | {
"section_name": [
"Introduction",
"Introduction ::: Sharing is caring",
"Introduction ::: Easy-access and high-performance",
"Introduction ::: Interpretability and diversity",
"Introduction ::: Pushing best practices forward",
"Introduction ::: From research to production",
"Community",
... | {
"question": [
"What state-of-the-art general-purpose pretrained models are made available under the unified API? "
],
"question_id": [
"ec62c4cdbeaafc875c695f2d4415bce285015763"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"... | {
"caption": [
"Figure 1: Write With Transformer"
],
"file": [
"6-Figure1-1.png"
]
} | Introduction
In the past 18 months, advances on many Natural Language Processing (NLP) tasks have been dominated by deep learning models and, more specifically, the use of Transfer Learning methods BIBREF0 in which a deep neural network language model is pretrained on a web-scale unlabelled text dataset with a general-... |
1904.06941 | A framework for streamlined statistical prediction using topic models | In the Humanities and Social Sciences, there is increasing interest in approaches to information extraction, prediction, intelligent linkage, and dimension reduction applicable to large text corpora. With approaches in these fields being grounded in traditional statistical techniques, the need arises for frameworks whe... | {
"section_name": [
"Introduction",
"Definitions",
"LDA regression model",
"Regression model and number of topics",
"Introducing new documents",
"sLDA regression model",
"HMTM regression model",
"Testing the topic regression models",
"Word count model",
"Topic regression models... | {
"question": [
"How is performance measured?"
],
"question_id": [
"405964517f372629cda4326d8efadde0206b7751"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"50d8b4a941c26b89... | {
"caption": [
"Figure 1: Histograms of the maximum likelihood estimates of θ1 for corpora of two topics, given relative true values of 0.2 and 0.4.",
"Figure 2: Histograms of the maximum likelihood estimates of {θ1, θ2} for corpora of three topics, given relative true values of {0.1, 0.1} and {0.2, 0.3}.",
... | Introduction
For the past 20 years, topic models have been used as a means of dimension reduction on text data, in order to ascertain underlying themes, or `topics', from documents. These probabilistic models have frequently been applied to machine learning problems, such as web spam filtering BIBREF0 , database sortin... |
1906.01512 | LeafNATS: An Open-Source Toolkit and Live Demo System for Neural Abstractive Text Summarization | Neural abstractive text summarization (NATS) has received a lot of attention in the past few years from both industry and academia. In this paper, we introduce an open-source toolkit, namely LeafNATS, for training and evaluation of different sequence-to-sequence based models for the NATS task, and for deploying the pre... | {
"section_name": [
"Introduction",
"LeafNATS Toolkithttps://github.com/tshi04/LeafNATS",
"A Live System Demonstrationhttp://dmkdt3.cs.vt.edu/leafNATS",
"Architecture",
"Design of Frontend",
"The Proposed Model",
"Conclusion",
"Acknowledgments"
],
"paragraphs": [
[
"Being... | {
"question": [
"What models are included in the toolkit?"
],
"question_id": [
"ae95a7d286cb7a0d5bc1a8283ecbf803e9305951"
],
"nlp_background": [
""
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
""
],
"question_writer": [
"c1fbdd7a261021041f7... | {
"caption": [
"Figure 1: The framework of LeafNATS toolkit.",
"Table 1: Basic statistics of the datasets used.",
"Figure 2: The architecture of the live system.",
"Table 2: Performance of our implemented pointergenerator network on different datasets. NewsroomS and -H represent Newsroom summary and h... | Introduction
Being one of the prominent natural language generation tasks, neural abstractive text summarization (NATS) has gained a lot of popularity BIBREF0 , BIBREF1 , BIBREF2 . Different from extractive text summarization BIBREF3 , BIBREF4 , BIBREF5 , NATS relies on modern deep learning models, particularly sequenc... |
1711.06288 | Language-Based Image Editing with Recurrent Attentive Models | We investigate the problem of Language-Based Image Editing (LBIE). Given a source image and a natural language description, we want to generate a target image by editing the source image based on the description. We propose a generic modeling framework for two sub-tasks of LBIE: language-based image segmentation and im... | {
"section_name": [
"Introduction",
"Related Work",
"Recurrent attentive models",
"Segmentation from language expressions",
"Conditional GANs in image generation",
"Experiments",
"Experiments on CoSaL",
"Experiments on ReferIt",
"Experiments on Oxford-102 Flower Dataset",
"Conc... | {
"question": [
"Is there any human evaluation involved in evaluating this famework?"
],
"question_id": [
"0be0c8106df5fde4b544af766ec3d4a3d7a6c8a2"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"qu... | {
"caption": [
"Figure 1. In an interactive design interface, a sketch of shoes is presented to a customer, who then gives a verbal instruction on how to modify the design: “The insole of the shoes should be brown. The vamp and the heel should be purple and shining”. The system colorizes the sketch following the ... | Introduction
In this work, we aim to develop an automatic Language-Based Image Editing (LBIE) system. Given a source image, which can be a sketch, a grayscale image or a natural image, the system will automatically generate a target image by editing the source image following natural language instructions provided by ... |
2003.14026 | MULTEXT-East | MULTEXT-East language resources, a multilingual dataset for language engineering research, focused on the morphosyntactic level of linguistic description. The MULTEXT-East dataset includes the EAGLES-based morphosyntactic specifications, morphosyntactic lexicons, and an annotated multilingual corpora. The parallel corp... | {
"section_name": [
"Introduction",
"Resource encoding",
"The morphosyntactic specifications",
"The morphosyntactic specifications ::: Common specifications",
"The morphosyntactic specifications ::: Specifications for individual languages",
"The morphosyntactic specifications ::: PoS tags, MSD... | {
"question": [
"How big is multilingual dataset?"
],
"question_id": [
"959490ba72bd02f742db1e7b19525d4b6c419772"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"258ee4069f7... | {
"caption": [
"Table 1 MULTEXT-East resources by language and resource type.",
"Table 2 MULTEXT categories with the number of MULTEXT-East defined attributes, values and languages",
"Fig. 1 Example of the encoding for the common tables.",
"Fig. 2 Encoding of language particular tables.",
"Fig. 3 ... | Introduction
The MULTEXT-East project, (Multilingual Text Tools and Corpora for Central and Eastern European Languages) ran from ’95 to ’97 and developed standardised language resources for six Central and Eastern European languages, as well as for English, the “hub” language of the project BIBREF0. The project was a s... |
1911.00473 | BERT Goes to Law School: Quantifying the Competitive Advantage of Access to Large Legal Corpora in Contract Understanding | Fine-tuning language models, such as BERT, on domain specific corpora has proven to be valuable in domains like scientific papers and biomedical text. In this paper, we show that fine-tuning BERT on legal documents similarly provides valuable improvements on NLP tasks in the legal domain. Demonstrating this outcome is ... | {
"section_name": [
"Introduction",
"Background",
"Datasets ::: BERT Fine Tuning Dataset",
"Datasets ::: Classification Dataset",
"Experiments ::: Methodology",
"Experiments ::: Baseline",
"Experiments ::: Freezing the BERT layers",
"Results",
"Conclusion"
],
"paragraphs": [
... | {
"question": [
"How big is dataset used for fine-tuning BERT?"
],
"question_id": [
"504a069ccda21580ccbf18c34f5eefc0088fa105"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
... | {
"caption": [
"Figure 1: Screenshot of end user application showing an agreement with a \"fixed\" term.",
"Figure 2: Fine-Tuning BERT Loss Figure 3: BERT vs FT BERT Train Loss",
"Figure 4: BERT based architecture Figure 5: Benchmark architecture",
"Table 1: Pre-trained vs Fine-tuned BERT with no end ... | Introduction
Businesses rely on contracts to capture critical obligations with other parties, such as: scope of work, amounts owed, and cancellation policies. Various efforts have gone into automatically extracting and classifying these terms. These efforts have usually been modeled as: classification, entity and relat... |
1909.05016 | Proposal Towards a Personalized Knowledge-powered Self-play Based Ensemble Dialog System | This is the application document for the 2019 Amazon Alexa competition. We give an overall vision of our conversational experience, as well as a sample conversation that we would like our dialog system to achieve by the end of the competition. We believe personalization, knowledge, and self-play are important component... | {
"section_name": [
"Vision",
"Sample Conversation",
"Architecture",
"Architecture ::: Dialog State Manager.",
"Architecture ::: NLP.",
"Architecture ::: Response Candidates.",
"Architecture ::: Dialog Manager.",
"Novelty",
"Related Work",
"Ensuring an engaging experience",
... | {
"question": [
"How big are datasets for 2019 Amazon Alexa competition?",
"What is novel in author's approach?"
],
"question_id": [
"d76ecdc0743893a895bc9dc3772af47d325e6d07",
"2a6469f8f6bf16577b590732d30266fd2486a72e"
],
"nlp_background": [
"zero",
"zero"
],
"topic_background": [... | {
"caption": [
"Figure 1: System architecture. Components with gray background are provided by Amazon. Components marked core, core+, and core++ are to be completed by the end of phase 5, 7, and 9, denoted by a solid, long dotted, and short dotted outline, respectively."
],
"file": [
"4-Figure1-1.png"
]... | Vision
Prompt: What is your team’s vision for your Socialbot? How do you want your customers to feel at the end of an interaction with your socialbot? How would your team measure success in competition?
Our vision is made up of the following main points:
1. A natural, engaging, and knowledge-powered conversational ex... |
1605.07683 | Learning End-to-End Goal-Oriented Dialog | Traditional dialog systems used in goal-oriented applications require a lot of domain-specific handcrafting, which hinders scaling up to new domains. End-to-end dialog systems, in which all components are trained from the dialogs themselves, escape this limitation. But the encouraging success recently obtained in chit-... | {
"section_name": [
"Introduction",
"Related Work",
"Goal-Oriented Dialog Tasks",
"Models",
"Acknowledgments"
],
"paragraphs": [
[],
[],
[
"All our tasks involve a restaurant reservation system, where the goal is to book a table at a restaurant. The first five tasks are gener... | {
"question": [
"How large is the Dialog State Tracking Dataset?"
],
"question_id": [
"a02696d4ab728ddd591f84a352df9375faf7d1b4"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
"dialog"
],
"question_wr... | {
"caption": [
"Figure 1: Goal-oriented dialog tasks. A user (in green) chats with a bot (in blue) to book a table at a restaurant. Models must predict bot utterances and API calls (in dark red). Task 1 tests the capacity of interpreting a request and asking the right questions to issue an API call. Task 2 checks... | Introduction
Related Work
Goal-Oriented Dialog Tasks
All our tasks involve a restaurant reservation system, where the goal is to book a table at a restaurant. The first five tasks are generated by a simulation, the last one uses real human-bot dialogs. The data for all tasks is available at http://fb.ai/babi. We also g... |
1912.00955 | Dynamic Prosody Generation for Speech Synthesis using Linguistics-Driven Acoustic Embedding Selection | Recent advances in Text-to-Speech (TTS) have improved quality and naturalness to near-human capabilities when considering isolated sentences. But something which is still lacking in order to achieve human-like communication is the dynamic variations and adaptability of human speech. This work attempts to solve the prob... | {
"section_name": [
"Introduction",
"Proposed Systems",
"Proposed Systems ::: Systems",
"Proposed Systems ::: Systems ::: Syntactic",
"Proposed Systems ::: Systems ::: BERT",
"Proposed Systems ::: Systems ::: BERT Syntactic",
"Proposed Systems ::: Applications to LFR",
"Experimental Pr... | {
"question": [
"What dataset is used for train/test of this method?"
],
"question_id": [
"78577fd1c09c0766f6e7d625196adcc72ddc8438"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer":... | {
"caption": [
"Fig. 1: Constituency parse tree",
"Fig. 2: Acoustic Embedding Distance Matrix using Syntactic Distance as Linguistic Similarity Measure",
"Fig. 4: Schematic of the implemented TTS system",
"Fig. 3: Acoustic Distance (solid line) vs Linguistic Distance (dashed line) as a function of LSW... | Introduction
Corresponding author email: tshubhi@amazon.com. Paper submitted to IEEE ICASSP 2020
Recent advances in TTS have improved the achievable synthetic speech naturalness to near human-like capabilities BIBREF0, BIBREF1, BIBREF2, BIBREF3. This means that for simple sentences, or for situations in which we can c... |
1711.00106 | DCN+: Mixed Objective and Deep Residual Coattention for Question Answering | Traditional models for question answering optimize using cross entropy loss, which encourages exact answers at the cost of penalizing nearby or overlapping answers that are sometimes equally accurate. We propose a mixed objective that combines cross entropy loss with self-critical policy learning. The objective uses re... | {
"section_name": [
"Introduction",
null,
"Deep residual coattention encoder",
"Mixed objective using self-critical policy learning",
"Experiments",
"Results",
"Related work",
"Conclusion"
],
"paragraphs": [
[
"Existing state-of-the-art question answering models are train... | {
"question": [
"How much is the gap between using the proposed objective and using only cross-entropy objective?"
],
"question_id": [
"1f63ccc379f01ecdccaa02ed0912970610c84b72"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"somewhat"
],
"... | {
"caption": [
"Figure 1: Deep residual coattention encoder.",
"Figure 2: Computation of the mixed objective.",
"Table 1: Test performance on SQuAD. The papers are as follows: rnet (Microsoft Asia Natural Language Computing Group, 2017), SEDT (Liu et al., 2017), BiDAF (Seo et al., 2017), DCN w/ CoVe (McCa... | Introduction
Existing state-of-the-art question answering models are trained to produce exact answer spans for a question and a document. In this setting, a ground truth answer used to supervise the model is defined as a start and an end position within the document. Existing training approaches optimize using cross en... |
1804.06385 | Bootstrapping Generators from Noisy Data | A core step in statistical data-to-text generation concerns learning correspondences between structured data representations (e.g., facts in a database) and associated texts. In this paper we aim to bootstrap generators from large scale datasets where the data (e.g., DBPedia facts) and related texts (e.g., Wikipedia ab... | {
"section_name": [
"Introduction",
"Related Work",
"Bidirectional Content Selection",
"Generator Training",
"Encoder-Decoder Base Generator",
"Predicting Alignment Labels",
"Reinforcement Learning Training",
"Results",
"Conclusions",
"Acknowledgments"
],
"paragraphs": [
... | {
"question": [
"What is the multi-instance learning?"
],
"question_id": [
"736c74d2f61ac8d3ac31c45c6510a36c767a5d6d"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"50d8b4a9... | {
"caption": [
"Figure 1: Property-value pairs (a), related biographic abstract (b) for the Wikipedia entity Robert Flaherty, and model verbalisation in italics (c).",
"Table 1: Example of word-property alignments for the Wikipedia abstract and facts in Figure 1.",
"Table 2: Dataset statistics.",
"Tab... | Introduction
A core step in statistical data-to-text generation concerns learning correspondences between structured data representations (e.g., facts in a database) and paired texts BIBREF0 , BIBREF1 , BIBREF2 . These correspondences describe how data representations are expressed in natural language (content realisat... |
1811.09529 | Competency Questions and SPARQL-OWL Queries Dataset and Analysis | Competency Questions (CQs) are natural language questions outlining and constraining the scope of knowledge represented by an ontology. Despite that CQs are a part of several ontology engineering methodologies, we have observed that the actual publication of CQs for the available ontologies is very limited and even sca... | {
"section_name": [
"Introduction",
"Analysis of Competency Questions",
"Materials and Methods",
"Generating SPARQL-OWL queries from CQs",
"Acknowledgments"
],
"paragraphs": [
[
"Within the field of ontology engineering, Competency Questions (CQs) BIBREF0 are natural language questio... | {
"question": [
"How many domains of ontologies do they gather data from?"
],
"question_id": [
"b2254f9dd0e416ee37b577cef75ffa36cbcb8293"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question... | {
"caption": [
"Table 1: Competency questions dataset summary",
"Table 3: Replacing complex entity expressions with single identifier.",
"Table 2: Normalization of words into common forms. REMOVED means that given text is deleted from pattern.",
"Table 4: Number of pattern candidates and actual patter... | Introduction
Within the field of ontology engineering, Competency Questions (CQs) BIBREF0 are natural language questions outlining the scope of knowledge represented by an ontology. They represent functional requirements in the sense that the developed ontology or an ontology-based information system should be able to ... |
1604.06076 | Question Answering via Integer Programming over Semi-Structured Knowledge | Answering science questions posed in natural language is an important AI challenge. Answering such questions often requires non-trivial inference and knowledge that goes beyond factoid retrieval. Yet, most systems for this task are based on relatively shallow Information Retrieval (IR) and statistical correlation techn... | {
"section_name": [
"Introduction",
"Related Work",
"QA as Subgraph Optimization",
"Semi-Structured Knowledge as Tables",
"QA as a Search for Desirable Support Graphs",
"ILP Formulation",
"Evaluation",
"Solvers",
"Results",
"Ablation Study",
"Question Perturbation",
"Co... | {
"question": [
"How is the semi-structured knowledge base created?"
],
"question_id": [
"cb1126992a39555e154bedec388465b249a02ded"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"somewhat"
],
"search_query": [
"semi-structured"
],
... | {
"caption": [
"Figure 1: TableILP searches for the best support graph (chains of reasoning) connecting the question to an answer, in this case June. Constraints on the graph define what constitutes valid support and how to score it (Section 3.3).",
"Table 1: Notation for the ILP formulation.",
"Table 2: ... | Introduction
Answering questions posed in natural language is a fundamental AI task, with a large number of impressive QA systems built over the years. Today's Internet search engines, for instance, can successfully retrieve factoid style answers to many natural language queries by efficiently searching the Web. Inform... |
1808.04314 | Comparing morphological complexity of Spanish, Otomi and Nahuatl | We use two small parallel corpora for comparing the morphological complexity of Spanish, Otomi and Nahuatl. These are languages that belong to different linguistic families, the latter are low-resourced. We take into account two quantitative criteria, on one hand the distribution of types over tokens in a corpus, on th... | {
"section_name": [
"Introduction",
"The type-token relationship (TTR)",
"Entropy and Perplexity",
"The corpus",
"Morphological analysis tools",
"Complexity measures",
"TTR as a measure of morphological complexity",
"Predictability",
"Conclusions",
"Future work",
"Acknowled... | {
"question": [
"what is the practical application for this paper?"
],
"question_id": [
"d5256d684b5f1b1ec648d996c358e66fe51f4904"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
... | {
"caption": [
"Table 1: Size of the parallel corpus",
"Table 2: TTR for Nahuatl-Spanish corpus",
"Table 3: TTR for Otomi-Spanish corpus",
"Table 4: Perplexity obtained in the different parallel corpora",
"Table 5: Entropy obtained in the different parallel corpora"
],
"file": [
"4-Table1-... | Introduction
Morphology deals with the internal structure of words BIBREF0 , BIBREF1 . Languages of the world have different word production processes. Morphological richness vary from language to language, depending on their linguistic typology. In natural language processing (NLP), taking into account the morphologic... |
1707.06878 | Unsupervised, Knowledge-Free, and Interpretable Word Sense Disambiguation | Interpretability of a predictive model is a powerful feature that gains the trust of users in the correctness of the predictions. In word sense disambiguation (WSD), knowledge-based systems tend to be much more interpretable than knowledge-free counterparts as they rely on the wealth of manually-encoded elements repres... | {
"section_name": [
"Introduction",
"Related Work",
"Unsupervised Knowledge-Free Interpretable WSD",
"Induction of the WSD Models",
"WSD API",
"User Interface for Interpretable WSD",
"Evaluation",
"Dataset",
"Evaluation Metrics",
"Discussion of Results",
"Conclusion",
"... | {
"question": [
"Do they use a neural model for their task?"
],
"question_id": [
"2a1069ae3629ae8ecc19d2305f23445c0231dc39"
],
"nlp_background": [
"infinity"
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
""
],
"question_writer": [
"50d8b4a94... | {
"caption": [
"Figure 1: Software and functional architecture of the WSD system.",
"Figure 2: Single word disambiguation mode: results of disambiguation of the word “Jaguar” (B) in the sentence “Jaguar is a large spotted predator of tropical America similar to the leopard.” (A) using the WSD disambiguation m... | Introduction
The notion of word sense is central to computational lexical semantics. Word senses can be either encoded manually in lexical resources or induced automatically from text. The former knowledge-based sense representations, such as those found in the BabelNet lexical semantic network BIBREF0 , are easily int... |
1909.08752 | Summary Level Training of Sentence Rewriting for Abstractive Summarization | As an attempt to combine extractive and abstractive summarization, Sentence Rewriting models adopt the strategy of extracting salient sentences from a document first and then paraphrasing the selected ones to generate a summary. However, the existing models in this framework mostly rely on sentence-level rewards or sub... | {
"section_name": [
"Introduction",
"Background ::: Sentence Rewriting",
"Background ::: Learning Sentence Selection",
"Background ::: Pre-trained Transformers",
"Model",
"Model ::: Extractor Network",
"Model ::: Extractor Network ::: Leveraging Pre-trained Transformers",
"Model ::: Ex... | {
"question": [
"What's the method used here?"
],
"question_id": [
"0b411f942c6e2e34e3d81cc855332f815b6bc123"
],
"nlp_background": [
""
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
""
],
"question_writer": [
"50d8b4a941c26b89482c94ab324b5a2... | {
"caption": [
"Figure 1: The overview architecture of the extractor netwrok",
"Table 1: Performance on CNN/Daily Mail test set using the full length ROUGE F1 score. R-AVG calculates average score of ROUGE-1, ROUGE-2 and ROUGE-L.",
"Table 2: Comparison of extractor networks.",
"Table 3: Comparison of ... | Introduction
The task of automatic text summarization aims to compress a textual document to a shorter highlight while keeping salient information of the original text. In general, there are two ways to do text summarization: Extractive and Abstractive BIBREF0. Extractive approaches generate summaries by selecting sali... |
1905.10247 | Contextual Out-of-Domain Utterance Handling With Counterfeit Data Augmentation | Neural dialog models often lack robustness to anomalous user input and produce inappropriate responses which leads to frustrating user experience. Although there are a set of prior approaches to out-of-domain (OOD) utterance detection, they share a few restrictions: they rely on OOD data or multiple sub-domains, and th... | {
"section_name": [
"Introduction",
"METHODS",
"HCN",
"AE-HCN",
"AE-HCN-CNN",
"Counterfeit Data Augmentation",
"DATASETS",
"EXPERIMENTAL SETUP AND EVALUATION",
"CONCLUSION"
],
"paragraphs": [
[
"Recently, there has been a surge of excitement in developing chatbots for... | {
"question": [
"By how much does their method outperform state-of-the-art OOD detection?"
],
"question_id": [
"01123a39574bdc4684aafa59c52d956b532d2e53"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
... | {
"caption": [
"Fig. 1. The architecture of AE-HCN which is the same as HCN except for the autoencoder component.",
"Table 2. Data statistics. The numbers of distinct system actions are 58, 247, and 194 for bAbI6, GR, and GM, respectively.",
"Table 3. Evaluation results. P@K means Precision@K. OOD F1 deno... | Introduction
Recently, there has been a surge of excitement in developing chatbots for various purposes in research and enterprise. Data-driven approaches offered by common bot building platforms (e.g. Google Dialogflow, Amazon Alexa Skills Kit, Microsoft Bot Framework) make it possible for a wide range of users to eas... |
1811.07684 | Efficient keyword spotting using dilated convolutions and gating | We explore the application of end-to-end stateless temporal modeling to small-footprint keyword spotting as opposed to recurrent networks that model long-term temporal dependencies using internal states. We propose a model inspired by the recent success of dilated convolutions in sequence modeling applications, allowin... | {
"section_name": [
"Introduction",
"System description",
"Neural network architecture",
"Streaming inference",
"End-of-keyword labeling",
"Open dataset",
"Experimental setup",
"Results",
"Conclusion",
"Acknowledgements"
],
"paragraphs": [
[
"Keyword spotting (KWS... | {
"question": [
"What are dilated convolutions?"
],
"question_id": [
"954c4756e293fd5c26dc50dc74f505cc94b3f8cc"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"50d8b4a941c26b... | {
"caption": [
"Fig. 2: Dilated convolution layers with an exponential dilation rate of 1, 2, 4, 8 and filter size of 2. Blue nodes are input frame vectors, orange nodes are cached intermediate vectors used for streaming inference, green nodes are output vectors which are actually computed. refers to background."... | Introduction
Keyword spotting (KWS) aims at detecting a pre-defined keyword or set of keywords in a continuous stream of audio. In particular, wake-word detection is an increasingly important application of KWS, used to initiate an interaction with a voice interface. In practice, such systems run on low-resource device... |
1906.08382 | An Open-World Extension to Knowledge Graph Completion Models | We present a novel extension to embedding-based knowledge graph completion models which enables them to perform open-world link prediction, i.e. to predict facts for entities unseen in training based on their textual description. Our model combines a regular link prediction model learned from a knowledge graph with wor... | {
"section_name": [
"Introduction",
"Approach",
"Link Prediction Models",
"Word Embeddings and Aggregation",
"Transformation Functions",
"Experiments",
"Datasets",
"Experimental Setup",
"Comparison with State of the Art",
"Analysis of Different Link Prediction Models and Transf... | {
"question": [
"what was the evaluation metrics studied in this work?"
],
"question_id": [
"ee279ace5bc69d15e640da967bd4214fe264aa1a"
],
"nlp_background": [
"five"
],
"topic_background": [
"research"
],
"paper_read": [
"no"
],
"search_query": [
"knowledge graph completion ... | {
"caption": [
"Figure 1: Our approach first trains a KGC model on the graph without using textual information (bottom left). For every entity we can obtain a text-based embedding v by aggregating the word embeddings for tokens in the name and description (top left). A transformation Ψmap is learned on the traini... | Introduction
Knowledge graphs are a vital source for disambiguation and discovery in various tasks such as question answering BIBREF0 , information extraction BIBREF1 and search BIBREF2 . They are, however, known to suffer from data quality issues BIBREF3 . Most prominently, since formal knowledge is inherently sparse,... |
1608.04207 | Fine-grained Analysis of Sentence Embeddings Using Auxiliary Prediction Tasks | There is a lot of research interest in encoding variable length sentences into fixed length vectors, in a way that preserves the sentence meanings. Two common methods include representations based on averaging word vectors, and representations based on the hidden states of recurrent neural networks such as LSTMs. The s... | {
"section_name": [
"Encoder Decoder",
"Prediction Tasks",
"Additional Experiments - Content Task",
"Appendix III: Significance Tests"
],
"paragraphs": [
[
"Parameters of the encoder-decoder were tuned on a dedicated validation set. We experienced with different learning rates (0.1, 0.01... | {
"question": [
"Do they analyze ELMo?"
],
"question_id": [
"beda007307c76b8ce7ffcd159a8280d2e8c7c356"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"5053f146237e8fc8859ed398... | {
"caption": [
"Figure 1: Task accuracy vs. embedding size for different models; ED BLEU scores given for reference.",
"Figure 3: Order accuracy w/ and w/o sentence representation for ED and CBOW models.",
"Figure 4: Results for length, content and order tests on natural and permuted sentences.",
"Tab... | Encoder Decoder
Parameters of the encoder-decoder were tuned on a dedicated validation set. We experienced with different learning rates (0.1, 0.01, 0.001), dropout-rates (0.1, 0.2, 0.3, 0.5) BIBREF11 and optimization techniques (AdaGrad BIBREF6 , AdaDelta BIBREF30 , Adam BIBREF15 and RMSprop BIBREF29 ). We also experi... |
1911.08976 | Red Dragon AI at TextGraphs 2019 Shared Task: Language Model Assisted Explanation Generation | The TextGraphs-13 Shared Task on Explanation Regeneration asked participants to develop methods to reconstruct gold explanations for elementary science questions. Red Dragon AI's entries used the language of the questions and explanation text directly, rather than a constructing a separate graph-like representation. Ou... | {
"section_name": [
"Introduction",
"Introduction ::: Dataset Review",
"Introduction ::: Problem Review",
"Preliminary Steps",
"Model Architectures",
"Model Architectures ::: Optimized TF-IDF",
"Model Architectures ::: Iterated TF-IDF",
"Model Architectures ::: BERT Re-ranking",
"D... | {
"question": [
"what are the three methods presented in the paper?"
],
"question_id": [
"dac2591f19f5bbac3d4a7fa038ff7aa09f6f0d96"
],
"nlp_background": [
""
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
""
],
"question_writer": [
"c1fbdd7a2... | {
"caption": [
"Table 1: Base MAP scoring - where the Python Baseline1e9 is the same as the original Python Baseline, but with the evaluate.py code updated to assume missing explanations have rank of 109",
"Table 2: MAP scoring of new methods. The timings are in seconds for the whole dev-set, and the BERT Re-... | Introduction
The Explanation Regeneration shared task asked participants to develop methods to reconstruct gold explanations for elementary science questions BIBREF1, using a new corpus of gold explanations BIBREF2 that provides supervision and instrumentation for this multi-hop inference task.
Each explanation is rep... |
1812.01704 | Impact of Sentiment Detection to Recognize Toxic and Subversive Online Comments | The presence of toxic content has become a major problem for many online communities. Moderators try to limit this problem by implementing more and more refined comment filters, but toxic users are constantly finding new ways to circumvent them. Our hypothesis is that while modifying toxic content and keywords to fool ... | {
"section_name": [
"Introduction",
"Related Work",
"Lexicons",
"Message Preprocessing",
"Message Sentiment",
"Experimental Results",
"Toxicity Detection",
"Correlation",
"Subversive Toxicity Detection",
"Conclusion",
"Acknowledgements"
],
"paragraphs": [
[
"O... | {
"question": [
"what datasets did the authors use?"
],
"question_id": [
"f62c78be58983ef1d77049738785ec7ab9f2a3ee"
],
"nlp_background": [
""
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
""
],
"question_writer": [
"c1fbdd7a261021041f75fbe00... | {
"caption": [
"Table 1: Sentiment of words per lexicon",
"Table 2: Comparison between fixed window and syntactic dependencies negation detection algorithms",
"Table 3: Average sentiment scores of negative and positive (respectively) labeled sentences, and their overlap.",
"Table 4: Sentiment scores u... | Introduction
Online communities abound today, forming on social networks, on webforums, within videogames, and even in the comments sections of articles and videos. While this increased international contact and exchange of ideas has been a net positive, it has also been matched with an increase in the spread of high-r... |
2003.02639 | Phase transitions in a decentralized graph-based approach to human language | Zipf's law establishes a scaling behavior for word-frequencies in large text corpora. The appearance of Zipfian properties in human language has been previously explained as an optimization problem for the interests of speakers and hearers. On the other hand, human-like vocabularies can be viewed as bipartite graphs. T... | {
"section_name": [
"Introduction",
"The model ::: Key concepts on (bipartite) graphs",
"The model ::: Basic elements of the language game",
"The model ::: Language game rules",
"Methods",
"Results ::: Three structural phases in language formation",
"Results ::: Bipartite graphs to visuali... | {
"question": [
"What are three possible phases for language formation?"
],
"question_id": [
"639c145f0bcb1dd12d08108bc7a02f9ec181552e"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_write... | {
"caption": [
"Figure 1: Average population clustering 〈cc〉 versus ambiguity parameter ℘. It is described the behavior of the average population clustering versus ℘. 〈cc〉 denotes the average over 10 realizations. As shown in the figure, three phases for the evolution of bipartite word-meaning mappings tend to ap... | Introduction
This letter arises from two intriguing questions about human language. The first question is: To what extent language, and also language evolution, can be viewed as a graph-theoretical problem? Language is an amazing example of a system of interrelated units at different organization scales. Several recent... |
1909.12440 | Improving Pre-Trained Multilingual Models with Vocabulary Expansion | Recently, pre-trained language models have achieved remarkable success in a broad range of natural language processing tasks. However, in multilingual setting, it is extremely resource-consuming to pre-train a deep language model over large-scale corpora for each language. Instead of exhaustively pre-training monolingu... | {
"section_name": [
"META-REVIEW",
"REVIEWER #1 ::: What is this paper about, what contributions does it make, what are the main strengths and weaknesses?",
"REVIEWER #1 ::: Reasons to accept",
"REVIEWER #1 ::: Reasons to reject",
"REVIEWER #1 ::: Reviewer's Scores",
"REVIEWER #1 ::: Questions... | {
"question": [
"How many parameters does the model have?"
],
"question_id": [
"ab3737fbf17b7a0e790e1315fffe46f615ebde64"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
"5053f... | {
"caption": [
"Figure 1: Left: fine-tuning BERT on different kinds of end tasks. Right: illustration of joint and mixture mapping (in this example, during mixture mapping, we represent e(cer) = 0.7 ∗ e(er) + 0.2 ∗ e(or) + 0.1 ∗ e(ch)).",
"Table 1: Alignment from Independent Mapping.",
"Table 2: POS taggi... | META-REVIEW
Comments: An approach to handle the OOV issue in multilingual BERT is proposed. A great deal of nice experiments were done but ultimately (and in message board discussions) the reviewers agreed there wasn't enough novelty or result here to justify acceptance.
REVIEWER #1 ::: What is this paper about, what ... |
1606.07298 | Explaining Predictions of Non-Linear Classifiers in NLP | Layer-wise relevance propagation (LRP) is a recently proposed technique for explaining predictions of complex non-linear classifiers in terms of input variables. In this paper, we apply LRP for the first time to natural language processing (NLP). More precisely, we use it to explain the predictions of a convolutional n... | {
"section_name": [
"Introduction",
"Explaining Predictions of Classifiers",
"Layer-Wise Relevance Propagation",
"Sensitivity Analysis",
"Experiments",
"CNN Model",
"Experimental Setup",
"Evaluating Word-Level Relevances",
"Document Highlighting",
"Document Visualization",
... | {
"question": [
"Do the experiments explore how various architectures and layers contribute towards certain decisions?"
],
"question_id": [
"0b8d64d6cdcfc2ba66efa41a52e09241729a697c"
],
"nlp_background": [
"two"
],
"topic_background": [
"research"
],
"paper_read": [
"no"
],
"se... | {
"caption": [
"Figure 1: Word deletion on initially correct (left) and false (right) classified test documents, using either LRP or SA. The target class is the true document class, words are deleted in decreasing (left) and increasing (right) order of their relevance. Random deletion is averaged over 10 runs (st... | Introduction
Following seminal work by Bengio and Collobert, the use of deep learning models for natural language processing (NLP) applications received an increasing attention in recent years. In parallel, initiated by the computer vision domain, there is also a trend toward understanding deep learning models through ... |
1911.03648 | Hate Speech Detection on Vietnamese Social Media Text using the Bidirectional-LSTM Model | In this paper, we describe our system which participates in the shared task of Hate Speech Detection on Social Networks of VLSP 2019 evaluation campaign. We are provided with the pre-labeled dataset and an unlabeled dataset for social media comments or posts. Our mission is to pre-process and build machine learning mod... | {
"section_name": [
"Introduction",
"Related Work",
"Bi-LSTM model for Vietnamese Hate Speech Detection",
"Bi-LSTM model for Vietnamese Hate Speech Detection ::: Long Short-Term Memory",
"Bi-LSTM model for Vietnamese Hate Speech Detection ::: Bidirectional Long Short-Term Memory",
"Pre-Process... | {
"question": [
"What social media platform does the data come from?"
],
"question_id": [
"891c4af5bb77d6b8635ec4109572de3401b60631"
],
"nlp_background": [
""
],
"topic_background": [
""
],
"paper_read": [
""
],
"search_query": [
"social media"
],
"question_writer": [
... | {
"caption": [
"Fig. 1. Bi-LSTM architecture [12]",
"TABLE I THE STATISTIC OF VLSP 2019 HSDOSN TRAINING DATASET",
"TABLE III THE RESULTS TABLE OF THE TOP 5 ON PUBLIC-TEST SET",
"TABLE II THE RESULTS TABLE OF MODELS."
],
"file": [
"2-Figure1-1.png",
"2-TableI-1.png",
"3-TableIII-1.png",... | Introduction
In recent years, social networking has grown and become prevalent with every people, it makes easy for people to interact and share with each other. However, every problem has two sides. It also has some negative issues, hate speech is a hot topic in the domain of social media. With the freedom of speech o... |
1712.03556 | Stochastic Answer Networks for Machine Reading Comprehension | We propose a simple yet robust stochastic answer network (SAN) that simulates multi-step reasoning in machine reading comprehension. Compared to previous work such as ReasoNet which used reinforcement learning to determine the number of steps, the unique feature is the use of a kind of stochastic prediction dropout on ... | {
"section_name": [
"Introduction",
"Proposed model: SAN",
"Experiment Setup",
"Results",
"How robust are the results?",
"Is it possible to use different numbers of steps in test vs. train?",
"How does the training time compare?",
"How does SAN perform by question type?",
"Experime... | {
"question": [
"How much performance improvements they achieve on SQuAD?"
],
"question_id": [
"39a450ac15688199575798e72a2cc016ef4316b5"
],
"nlp_background": [
"five"
],
"topic_background": [
"research"
],
"paper_read": [
"yes"
],
"search_query": [
"Machine Reading"
],
... | {
"caption": [
"Figure 1: Illustration of “stochastic prediction dropout” in the answer module during training. At each reasoning step t, the model combines memory (bottom row) with hidden states st−1 to generate a prediction (multinomial distribution). Here, there are three steps and three predictions, but one p... | Introduction
Machine reading comprehension (MRC) is a challenging task: the goal is to have machines read a text passage and then answer any question about the passage. This task is an useful benchmark to demonstrate natural language understanding, and also has important applications in e.g. conversational agents and c... |
1908.11279 | Grounded Agreement Games: Emphasizing Conversational Grounding in Visual Dialogue Settings | Where early work on dialogue in Computational Linguistics put much emphasis on dialogue structure and its relation to the mental states of the dialogue participants (e.g., Allen 1979, Grosz & Sidner 1986), current work mostly reduces dialogue to the task of producing at any one time a next utterance; e.g. in neural cha... | {
"section_name": [
"Introduction",
"Visual Dialogue as Example of the Scaling Up Approach",
"Agreement Games",
"Agreement Games ::: More formally",
"Some Examples ::: The MeetUp Game",
"Some Examples ::: The MatchIt Game",
"Some Examples ::: The Concept Learning Game",
"Conclusions"
... | {
"question": [
"Do the authors perform experiments using their proposed method?"
],
"question_id": [
"de015276dcde4e7d1d648c6e31100ec80f61960f"
],
"nlp_background": [
"two"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"questi... | {
"caption": [
"Figure 2: Interacting with a visual chat bot, see http: //demo.visualdialog.org/",
"Figure 1: Original example of Visual Dialogue System, from official demo video (https://vimeo. com/193092429)"
],
"file": [
"2-Figure2-1.png",
"2-Figure1-1.png"
]
} | Introduction
If you're good at replying to a single request, are you also likely to be good at doing dialogue? Much current work seems to assume that the answer to this question is yes, in that it attempts a scaling up from single pairs of utterance plus response to longer dialogues: See, e.g., the work on neural chatb... |
1708.00214 | Natural Language Processing with Small Feed-Forward Networks | We show that small and shallow feed-forward neural networks can achieve near state-of-the-art results on a range of unstructured and structured language processing tasks while being considerably cheaper in memory and computational requirements than deep recurrent models. Motivated by resource-constrained environments l... | {
"section_name": [
"Introduction",
"Small Feed-Forward Network Models",
"Experiments",
"Language Identification",
"POS Tagging",
"Segmentation",
"Preordering",
"Conclusions",
"Acknowledgments",
"Supplementary Material",
"Quantization Details",
"FLOPs Calculation",
... | {
"question": [
"What NLP tasks do the authors evaluate feed-forward networks on?"
],
"question_id": [
"56836afc57cae60210fa1e5294c88e40bb10cc0e"
],
"nlp_background": [
""
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_w... | {
"caption": [
"Figure 1: An example network structure for a model using bigrams of the previous, current and next word, and trigrams of the current word. Does not illustrate hashing.",
"Table 1: Language Identification. Quantization allows trading numerical precision for larger embeddings. The two models fro... | Introduction
Deep and recurrent neural networks with large network capacity have become increasingly accurate for challenging language processing tasks. For example, machine translation models have been able to attain impressive accuracies, with models that use hundreds of millions BIBREF0 , BIBREF1 or billions BIBREF2... |
1909.03464 | Back to the Future -- Sequential Alignment of Text Representations | Language evolves over time in many ways relevant to natural language processing tasks. For example, recent occurrences of tokens 'BERT' and 'ELMO' in publications refer to neural network architectures rather than persons. This type of temporal signal is typically overlooked, but is important if one aims to deploy a mac... | {
"section_name": [
"Introduction",
"Subspace Alignment",
"Subspace Alignment ::: Unsupervised subspace alignment",
"Subspace Alignment ::: Semi-supervised subspace alignment",
"Subspace Alignment ::: Extending SSA to Unbounded Time",
"Subspace Alignment ::: Considering Sample Similarities bet... | {
"question": [
"What are three challenging tasks authors evaluated their sequentially aligned representations?"
],
"question_id": [
"6147846520a3dc05b230241f2ad6d411d614e24c"
],
"nlp_background": [
"zero"
],
"topic_background": [
"unfamiliar"
],
"paper_read": [
"no"
],
"search... | {
"caption": [
"Figure 1: Example of word embeddings between two timesteps, blue = PERSON vs red = ARTEFACT (black = unknown class). (Top) Source data, (bottom) target data.",
"Figure 2: Example of subspace alignment procedure. (Left) Source data, (left middle) unsupervised alignment of the source data, (righ... | Introduction
As time passes, language usage changes. For example, the names `Bert' and `Elmo' would only rarely make an appearance prior to 2018 in the context of scientific writing. After the publication of BERT BIBREF0 and ELMo BIBREF1, however, usage has increased in frequency. In the context of named entities on Tw... |
1801.07537 | Analyzing Language Learned by an Active Question Answering Agent | We analyze the language learned by an agent trained with reinforcement learning as a component of the ActiveQA system [Buck et al., 2017]. In ActiveQA, question answering is framed as a reinforcement learning task in which an agent sits between the user and a black box question-answering system. The agent learns to ref... | {
"section_name": [
"Introduction",
"Analysis of the Agent's Language",
"The Language of SearchQA Questions",
"The Language of the Base NMT Model",
"The Language of the AQA Agent",
"Conclusions: Rediscovering IR?"
],
"paragraphs": [
[
" BIBREF0 propose a reinforcement learning fr... | {
"question": [
"What is the difference in findings of Buck et al? It looks like the same conclusion was mentioned in Buck et al.."
],
"question_id": [
"99cf494714c67723692ad1279132212db29295f3"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"n... | {
"caption": [
"Figure 1: An agent-environment framework for Active Question Answering.",
"Figure 2: Boxplot summaries of the statistics collected for all types of questions. Two-sample t-tests performed on all pairs in each box confirm that the differences in means are significant p < 10−3."
],
"file": [... | Introduction
BIBREF0 propose a reinforcement learning framework for question answering, called active question answering (ActiveQA), that aims to improve answering by systematically perturbing input questions (cf. BIBREF1 ). Figure 1 depicts the generic agent-environment framework. The agent (AQA) interacts with the e... |
1612.09113 | Deep Semi-Supervised Learning with Linguistically Motivated Sequence Labeling Task Hierarchies | In this paper we present a novel Neural Network algorithm for conducting semisupervised learning for sequence labeling tasks arranged in a linguistically motivated hierarchy. This relationship is exploited to regularise the representations of supervised tasks by backpropagating the error of the unsupervised task throug... | {
"section_name": [
"Introduction",
"Linguistically Motivated Task Hierarchies",
"Motivating our Choice of Tasks",
"Our Model",
"Supervision of Multiple Tasks",
"Bi-Directional RNNs",
"Implementation Details",
"Data Sets",
"Baseline Results",
"Semi-Supervised Experiments",
... | {
"question": [
"What is the baseline?",
"What is the unsupervised task in the final layer?",
"How many supervised tasks are used?",
"What is the network architecture?"
],
"question_id": [
"85e45b37408bb353c6068ba62c18e516d4f67fe9",
"f4e1d2276d3fc781b686d2bb44eead73e06fbf3f",
"bf2ebc9b... | {
"caption": [
"Figure 1: Our Hierarchical Network. In this network, junior tasks are supervised in lower layers, with an unsupervised task (Language Modeling) at the most senior layer.",
"Table 2: Hierarchy POS Results",
"Table 1: Hierarchy Chunking Results",
"Figure 2: T-SNE for Chunk labels. The or... | Introduction
It is natural to think of NLP tasks existing in a hierarchy, with each task building upon the previous tasks. For example, Part of Speech (POS) is known to be an extremely strong feature for Noun Phrase Chunking, and downstream tasks such as greedy Language Modeling (LM) can make use of information about t... |
1903.04715 | Context-Aware Learning for Neural Machine Translation | Interest in larger-context neural machine translation, including document-level and multi-modal translation, has been growing. Multiple works have proposed new network architectures or evaluation schemes, but potentially helpful context is still sometimes ignored by larger-context translation models. In this paper, we ... | {
"section_name": [
"Introduction",
"Background: Larger-Context Neural Machine Translation",
"Existing approaches to larger-context neural translation",
"Learning to use the context",
"Neutral, useful and harmful context",
"Context regularization",
"Estimating context-less sc... | {
"question": [
"Is the proposed model more sensitive than previous context-aware models too?",
"In what ways the larger context is ignored for the models that do consider larger context?"
],
"question_id": [
"6adde6bc3e27a32eac5daa57d30ab373f77690be",
"90ad8d7ee27192b89ffcfa4a68302f370e6333a8"
... | {
"caption": [
"Figure 1: Cumulative BLEU scores on the validation set sorted by the sentence-level score difference according to the larger-context model.",
"Table 1: We report the BLEU scores with the correctly paired context as well as with the incorrectly paired context (context-marginalized). Context-mar... | Introduction
Despite its rapid adoption by academia and industry and its recent success BIBREF0 , neural machine translation has been found largely incapable of exploiting additional context other than the current source sentence. This incapability stems from the fact that larger-context machine translation systems ten... |
1612.04675 | Recurrent Deep Stacking Networks for Speech Recognition | This paper presented our work on applying Recurrent Deep Stacking Networks (RDSNs) to Robust Automatic Speech Recognition (ASR) tasks. In the paper, we also proposed a more efficient yet comparable substitute to RDSN, Bi- Pass Stacking Network (BPSN). The main idea of these two models is to add phoneme-level informatio... | {
"section_name": [
"Introduction",
"Recurrent Deep Stacking Network",
"Compressing the Outputs",
"BiPass Stacking Network",
"Experiments",
"Conclusion"
],
"paragraphs": [
[
"Ever since the introduction of Deep Neural Networks (DNNs) to Automatic Speech Recognition (ASR) tasks BI... | {
"question": [
"What does recurrent deep stacking network do?"
],
"question_id": [
"ba1da61db264599963e340010b777a1723ffeb4c"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"research"
],
"paper_read": [
"no"
],
"search_query": [
""
],
"question_writer": [
... | {
"caption": [
"Fig. 1: Recurrent Deep Stacking Network Framework",
"Fig. 2: Cross Entropy Values of Baseline and RDSN"
],
"file": [
"1-Figure1-1.png",
"2-Figure2-1.png"
]
} | Introduction
Ever since the introduction of Deep Neural Networks (DNNs) to Automatic Speech Recognition (ASR) tasks BIBREF0 , researchers had been trying to use additional inputs to the raw input features. We extracted features that are more representative using the first and second order differentiates of the raw inpu... |
1702.03274 | Hybrid Code Networks: practical and efficient end-to-end dialog control with supervised and reinforcement learning | End-to-end learning of recurrent neural networks (RNNs) is an attractive solution for dialog systems; however, current techniques are data-intensive and require thousands of dialogs to learn simple behaviors. We introduce Hybrid Code Networks (HCNs), which combine an RNN with domain-specific knowledge encoded as softwa... | {
"section_name": [
"Introduction",
"Model description",
"Related work",
"Supervised learning evaluation I",
"Supervised learning evaluation II",
"Reinforcement learning illustration",
"Conclusion",
"Model implementation details",
"bAbI Task5 example dialog",
"bAbI Task6 exampl... | {
"question": [
"Does the latent dialogue state heklp their model?",
"Do the authors test on datasets other than bAbl?",
"What is the reward model for the reinforcement learning appraoch?"
],
"question_id": [
"ff814793387c8f3b61f09b88c73c00360a22a60e",
"059acc270062921ad27ee40a77fd50de6f02840a... | {
"caption": [
"Figure 1: Operational loop. Trapezoids refer to programmatic code provided by the software developer, and shaded boxes are trainable components. Vertical bars under “6” represent concatenated vectors which form the input to the RNN.",
"Table 1: Results on bAbI dialog Task5-OOV and Task6 (Borde... | Introduction
Task-oriented dialog systems help a user to accomplish some goal using natural language, such as making a restaurant reservation, getting technical support, or placing a phonecall. Historically, these dialog systems have been built as a pipeline, with modules for language understanding, state tracking, act... |
1610.03112 | Leveraging Recurrent Neural Networks for Multimodal Recognition of Social Norm Violation in Dialog | Social norms are shared rules that govern and facilitate social interaction. Violating such social norms via teasing and insults may serve to upend power imbalances or, on the contrary reinforce solidarity and rapport in conversation, rapport which is highly situated and context-dependent. In this work, we investigate ... | {
"section_name": [
"Introduction and Related Work",
"Data and Annotation",
"Model and Experiment",
"Models",
"Experiment Result",
"Conclusion and Future Work"
],
"paragraphs": [
[
"Social norms are informal understandings that govern human behavior. They serve as the basis for o... | {
"question": [
"Does this paper propose a new task that others can try to improve performance on?"
],
"question_id": [
"cacb83e15e160d700db93c3f67c79a11281d20c5"
],
"nlp_background": [
"infinity"
],
"topic_background": [
"familiar"
],
"paper_read": [
"no"
],
"search_query": [
... | {
"caption": [
"Table 1: Statistics of the corpus",
"Figure 1: Three proposed computational models.",
"Table 2: Performance comparsion for the 3 evaluated models"
],
"file": [
"2-Table1-1.png",
"3-Figure1-1.png",
"4-Table2-1.png"
]
} | Introduction and Related Work
Social norms are informal understandings that govern human behavior. They serve as the basis for our beliefs and expectations about others, and are instantiated in human-human conversation through verbal and nonverbal behaviors BIBREF0 , BIBREF1 . There is considerable body of work on mode... |
1607.03542 | Open-Vocabulary Semantic Parsing with both Distributional Statistics and Formal Knowledge | Traditional semantic parsers map language onto compositional, executable queries in a fixed schema. This mapping allows them to effectively leverage the information contained in large, formal knowledge bases (KBs, e.g., Freebase) to answer questions, but it is also fundamentally limiting---these semantic parsers can on... | {
"section_name": [
"Introduction",
"Open vocabulary semantic parsing",
"Converting Freebase queries to features",
"Subgraph feature extraction",
"Feature selection",
"Combined predicate models",
"Making full use of KB information",
"Logical form generation",
"Candidate entity gene... | {
"question": [
"What knowledge base do they use?",
"How big is their dataset?",
"What task do they evaluate on?"
],
"question_id": [
"33957fde72f9082a5c11844e7c47c58f8029c4ae",
"1c4cd22d6eaefffd47b93c2124f6779a06d2d9e1",
"2122bd05c03dde098aa17e36773e1ac7b6011969"
],
"nlp_background": ... | {
"caption": [
"Figure 1: Overview of the components of our model. Given an input text, we use a CCG parser and an entity linker to produce a logical form with predicates derived from the text (shown in italics). For each predicate, we learn a distributional vector θ, as well as weights ω associated with a set of... | Introduction
Semantic parsing is the task of mapping a phrase in natural language onto a formal query in some fixed schema, which can then be executed against a knowledge base (KB) BIBREF0 , BIBREF1 . For example, the phrase “Who is the president of the United States?” might be mapped onto the query $\lambda (x).$ $\te... |
1812.10860 | Can You Tell Me How to Get Past Sesame Street? Sentence-Level Pretraining Beyond Language Modeling | Natural language understanding has recently seen a surge of progress with the use of sentence encoders like ELMo (Peters et al., 2018a) and BERT (Devlin et al., 2019) which are pretrained on variants of language modeling. We conduct the first large-scale systematic study of candidate pretraining tasks, comparing 19 dif... | {
"section_name": [
"Introduction",
"Related Work",
"Pretraining Tasks",
"Models and Training Procedures",
"Results",
"Conclusions",
"Hyperparameters and Optimization Details",
"Multitask Learning Methods",
"Diagnostic Set Results"
],
"paragraphs": [
[
"State-of-the-a... | {
"question": [
"Do some pretraining objectives perform better than others for sentence level understanding tasks?"
],
"question_id": [
"1d6c42e3f545d55daa86bea6fabf0b1c52a93bbb"
],
"nlp_background": [
"five"
],
"topic_background": [
"familiar"
],
"paper_read": [
"somewhat"
],
... | {
"caption": [
"Figure 1: Our common model design: During pretraining, we train the shared encoder and the task-specific model for each pretraining task. We then freeze the shared encoder and train the task-specific model anew for each target evaluation task. Tasks may involve more than one sentence.",
"Table... | Introduction
State-of-the-art models for natural language processing (NLP) tasks like translation, question answering, and parsing include components intended to extract representations for the meaning and contents of each input sentence. These sentence encoder components are typically trained directly for the target t... |
1712.02121 | A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network | In this paper, we propose a novel embedding model, named ConvKB, for knowledge base completion. Our model ConvKB advances state-of-the-art models by employing a convolutional neural network, so that it can capture global relationships and transitional characteristics between entities and relations in knowledge bases. I... | {
"section_name": [
"Introduction",
"Proposed ConvKB model",
"Datasets",
"Evaluation protocol",
"Training protocol",
"Main experimental results",
"Conclusion",
"Acknowledgments"
],
"paragraphs": [
[
"Large-scale knowledge bases (KBs), such as YAGO BIBREF0 , Freebase BIBRE... | {
"question": [
"Did the authors try stacking multiple convolutional layers?",
"How many feature maps are generated for a given triple?",
"How does the number of parameters compare to other knowledge base completion models?"
],
"question_id": [
"480e10e5a1b9c0ae9f7763b7611eeae9e925096b",
"056f... | {
"caption": [
"Table 1: The score functions in previous SOTA models and in our ConvKB model. ‖v‖p denotes the p-norm of v. 〈vh,vr,vt〉 = ∑ i vhivrivti denotes a tri-linear dot product. g denotes a non-linear function. ∗ denotes a convolution operator. · denotes a dot product. concat denotes a concatenation operat... | Introduction
Large-scale knowledge bases (KBs), such as YAGO BIBREF0 , Freebase BIBREF1 and DBpedia BIBREF2 , are usually databases of triples representing the relationships between entities in the form of fact (head entity, relation, tail entity) denoted as (h, r, t), e.g., (Melbourne, cityOf, Australia). These KBs ar... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.