Get our free extension to see links to code for papers anywhere online!

Chrome logo Add to Chrome

Firefox logo Add to Firefox

"Text": models, code, and papers

Transfer learning of language-independent end-to-end ASR with language model fusion

Nov 06, 2018
Hirofumi Inaguma, Jaejin Cho, Murali Karthick Baskar, Tatsuya Kawahara, Shinji Watanabe

This work explores better adaptation methods to low-resource languages using an external language model (LM) under the framework of transfer learning. We first build a language-independent ASR system in a unified sequence-to-sequence (S2S) architecture with a shared vocabulary among all languages. During adaptation, we perform LM fusion transfer, where an external LM is integrated into the decoder network of the attention-based S2S model in the whole adaptation stage, to effectively incorporate linguistic context of the target language. We also investigate various seed models for transfer learning. Experimental evaluations using the IARPA BABEL data set show that LM fusion transfer improves performances on all target five languages compared with simple transfer learning when the external text data is available. Our final system drastically reduces the performance gap from the hybrid systems.


  Access Paper or Ask Questions

Data Augmentation for Spoken Language Understanding via Joint Variational Generation

Nov 06, 2018
Kang Min Yoo, Youhyun Shin, Sang-goo Lee

Data scarcity is one of the main obstacles of domain adaptation in spoken language understanding (SLU) due to the high cost of creating manually tagged SLU datasets. Recent works in neural text generative models, particularly latent variable models such as variational autoencoder (VAE), have shown promising results in regards to generating plausible and natural sentences. In this paper, we propose a novel generative architecture which leverages the generative power of latent variable models to jointly synthesize fully annotated utterances. Our experiments show that existing SLU models trained on the additional synthetic examples achieve performance gains. Our approach not only helps alleviate the data scarcity issue in the SLU task for many datasets but also indiscriminately improves language understanding performances for various SLU models, supported by extensive experiments and rigorous statistical testing.

* 8 pages, 3 figures, 4 tables, Accepted in AAAI2019 

  Access Paper or Ask Questions

Categorization of Comparative Sentences for Argument Mining

Sep 17, 2018
Mirco Franzek, Alexander Panchenko, Chris Biemann

We present the first work on domain-independent comparative argument mining (CAM), which is the automatic extraction of direct comparisons from text. After motivating the need and identifying the widespread use of this so far under-researched topic, we present the first publicly available open-domain dataset for CAM. The dataset was collected using crowdsourcing and contains 7199 unique sentences for 217 distinct comparison target pairs selected over several domains, of which 27% contain a directed (better vs. worse) comparison. In classification experiments, we examine the impact of representations, features, and classifiers, and reach an F1-score of 88% with a gradient boosting model based on pre-trained sentence embeddings, especially reliably identifying non-comparative sentences. This paves the way for domain-independent comparison extraction from web-scale corpora for the use in result ranking and presentation for comparative queries.


  Access Paper or Ask Questions

Identifying Relationships Among Sentences in Court Case Transcripts Using Discourse Relations

Sep 15, 2018
Gathika Ratnayaka, Thejan Rupasinghe, Nisansa de Silva, Menuka Warushavithana, Viraj Gamage, Amal Shehan Perera

Case Law has a significant impact on the proceedings of legal cases. Therefore, the information that can be obtained from previous court cases is valuable to lawyers and other legal officials when performing their duties. This paper describes a methodology of applying discourse relations between sentences when processing text documents related to the legal domain. In this study, we developed a mechanism to classify the relationships that can be observed among sentences in transcripts of United States court cases. First, we defined relationship types that can be observed between sentences in court case transcripts. Then we classified pairs of sentences according to the relationship type by combining a machine learning model and a rule-based approach. The results obtained through our system were evaluated using human judges. To the best of our knowledge, this is the first study where discourse relationships between sentences have been used to determine relationships among sentences in legal court case transcripts.

* Conference: 2018 International Conference on Advances in ICT for Emerging Regions (ICTer) 

  Access Paper or Ask Questions

Automatic Event Salience Identification

Sep 03, 2018
Zhengzhong Liu, Chenyan Xiong, Teruko Mitamura, Eduard Hovy

Identifying the salience (i.e. importance) of discourse units is an important task in language understanding. While events play important roles in text documents, little research exists on analyzing their saliency status. This paper empirically studies the Event Salience task and proposes two salience detection models based on content similarities and discourse relations. The first is a feature based salience model that incorporates similarities among discourse units. The second is a neural model that captures more complex relations between discourse units. Tested on our new large-scale event salience corpus, both methods significantly outperform the strong frequency baseline, while our neural model further improves the feature based one by a large margin. Our analyses demonstrate that our neural model captures interesting connections between salience and discourse unit relations (e.g., scripts and frame structures).

* EMNLP 2018, 11 pages. Datasets, models and codes: https://github.com/hunterhector/EventSalience 

  Access Paper or Ask Questions

Latent Dirichlet Allocation (LDA) for Topic Modeling of the CFPB Consumer Complaints

Jul 18, 2018
Kaveh Bastani, Hamed Namavari, Jeffry Shaffer

A text mining approach is proposed based on latent Dirichlet allocation (LDA) to analyze the Consumer Financial Protection Bureau (CFPB) consumer complaints. The proposed approach aims to extract latent topics in the CFPB complaint narratives, and explores their associated trends over time. The time trends will then be used to evaluate the effectiveness of the CFPB regulations and expectations on financial institutions in creating a consumer oriented culture that treats consumers fairly and prioritizes consumer protection in their decision making processes. The proposed approach can be easily operationalized as a decision support system to automate detection of emerging topics in consumer complaints. Hence, the technology-human partnership between the proposed approach and the CFPB team could certainly improve consumer protections from unfair, deceptive or abusive practices in the financial markets by providing more efficient and effective investigations of consumer complaint narratives.


  Access Paper or Ask Questions

Generating Counterfactual Explanations with Natural Language

Jun 26, 2018
Lisa Anne Hendricks, Ronghang Hu, Trevor Darrell, Zeynep Akata

Natural language explanations of deep neural network decisions provide an intuitive way for a AI agent to articulate a reasoning process. Current textual explanations learn to discuss class discriminative features in an image. However, it is also helpful to understand which attributes might change a classification decision if present in an image (e.g., "This is not a Scarlet Tanager because it does not have black wings.") We call such textual explanations counterfactual explanations, and propose an intuitive method to generate counterfactual explanations by inspecting which evidence in an input is missing, but might contribute to a different classification decision if present in the image. To demonstrate our method we consider a fine-grained image classification task in which we take as input an image and a counterfactual class and output text which explains why the image does not belong to a counterfactual class. We then analyze our generated counterfactual explanations both qualitatively and quantitatively using proposed automatic metrics.

* presented at 2018 ICML Workshop on Human Interpretability in Machine Learning (WHI 2018), Stockholm, Sweden 

  Access Paper or Ask Questions

Contextual Explanation Networks

Jan 30, 2018
Maruan Al-Shedivat, Avinava Dubey, Eric P. Xing

We introduce contextual explanation networks (CENs)---a class of models that learn to predict by generating and leveraging intermediate explanations. CENs are deep networks that generate parameters for context-specific probabilistic graphical models which are further used for prediction and play the role of explanations. Contrary to the existing post-hoc model-explanation tools, CENs learn to predict and to explain jointly. Our approach offers two major advantages: (i) for each prediction, valid instance-specific explanations are generated with no computational overhead and (ii) prediction via explanation acts as a regularization and boosts performance in low-resource settings. We prove that local approximations to the decision boundary of our networks are consistent with the generated explanations. Our results on image and text classification and survival analysis tasks demonstrate that CENs are competitive with the state-of-the-art while offering additional insights behind each prediction, valuable for decision support.


  Access Paper or Ask Questions

Vietnamese Open Information Extraction

Jan 23, 2018
Diem Truong, Duc-Thuan Vo, U. T Nguyen

Open information extraction (OIE) is the process to extract relations and their arguments automatically from textual documents without the need to restrict the search to predefined relations. In recent years, several OIE systems for the English language have been created but there is not any system for the Vietnamese language. In this paper, we propose a method of OIE for Vietnamese using a clause-based approach. Accordingly, we exploit Vietnamese dependency parsing using grammar clauses that strives to consider all possible relations in a sentence. The corresponding clause types are identified by their propositions as extractable relations based on their grammatical functions of constituents. As a result, our system is the first OIE system named vnOIE for the Vietnamese language that can generate open relations and their arguments from Vietnamese text with highly scalable extraction while being domain independent. Experimental results show that our OIE system achieves promising results with a precision of 83.71%.


  Access Paper or Ask Questions

Word Embeddings Quantify 100 Years of Gender and Ethnic Stereotypes

Nov 22, 2017
Nikhil Garg, Londa Schiebinger, Dan Jurafsky, James Zou

Word embeddings use vectors to represent words such that the geometry between vectors captures semantic relationship between the words. In this paper, we develop a framework to demonstrate how the temporal dynamics of the embedding can be leveraged to quantify changes in stereotypes and attitudes toward women and ethnic minorities in the 20th and 21st centuries in the United States. We integrate word embeddings trained on 100 years of text data with the U.S. Census to show that changes in the embedding track closely with demographic and occupation shifts over time. The embedding captures global social shifts -- e.g., the women's movement in the 1960s and Asian immigration into the U.S -- and also illuminates how specific adjectives and occupations became more closely associated with certain populations over time. Our framework for temporal analysis of word embedding opens up a powerful new intersection between machine learning and quantitative social science.


  Access Paper or Ask Questions

<<
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
>>