Filters








257 Hits in 3.9 sec

Analyzing and Interpreting Neural Networks for NLP: A Report on the First BlackboxNLP Workshop [article]

Afra Alishahi and Grzegorz Chrupała and Tal Linzen
2019 arXiv   pre-print
The EMNLP 2018 workshop BlackboxNLP was dedicated to resources and techniques specifically developed for analyzing and understanding the inner-workings and representations acquired by neural models of  ...  Approaches included: systematic manipulation of input to neural networks and investigating the impact on their performance, testing whether interpretable knowledge can be decoded from intermediate representations  ...  Future trends and outlook This first edition of the BlackboxNLP workshop brought together a large amount of recent work on issues related to the analysis and interpretability of neural models, and thus  ... 
arXiv:1904.04063v1 fatcat:iwuio4l62jcpjindqpyzs6rmdy

Explicitly modeling case improves neural dependency parsing

Clara Vania, Adam Lopez
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
The tagger has the same structure as the parser's encoder, with an additional feedforward neural network with one hidden layer followed by a softmax layer.  ...  Dependency Parsing Model We use a neural graph-based dependency parser similar to that of Kiperwasser and Goldberg (2016) and Zhang et al. (2017) for all our experiments.  ... 
doi:10.18653/v1/w18-5447 dblp:conf/emnlp/VaniaL18 fatcat:bcs6xuw7erhunjeyaznkxdlxfm

Limitations in learning an interpreted language with recurrent models

Denis Paperno
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
In this submission I report work in progress on learning simplified interpreted languages by means of recurrent models.  ...  Preliminary results suggest that LSTM networks do generalise to compositional interpretation, albeit only in the most favorable learning setting.  ...  Acknowledgments The research has been supported by CNRS PEPS ReSeRVe grant. I also thank Germán Kruszewski for useful input on the topic.  ... 
doi:10.18653/v1/w18-5456 dblp:conf/emnlp/Paperno18 fatcat:hyp5h6fllre67fjwf5yl5bqgri

Grammar Induction with Neural Language Models: An Unusual Replication

Phu Mon Htut, Kyunghyun Cho, Samuel Bowman
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
We find that this model represents the first empirical success for neural network latent tree learning, and that neural language modeling warrants further study as a setting for grammar induction.  ...  In a recent paper published at ICLR, Shen et al. (2018) introduce such a model and report near state-ofthe-art results on the target task of language modeling, and the first strong latent tree learning  ... 
doi:10.18653/v1/w18-5452 dblp:conf/emnlp/HtutCB18a fatcat:2sqtlc7xsrgevcv3rtsnsywa6e

Linguistic representations in multi-task neural networks for ellipsis resolution

Ola Rønning, Daniel Hardt, Anders Søgaard
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
Both perspectives provide evidence for substantial linguistic knowledge being deployed by the neural networks.  ...  A recent model instead used partial parsing as an auxiliary task in sequential neural network architectures to inject syntactic information.  ...  On a slightly different version of the news dataset, Rønning et al. report a tokenlevel F1 score of 0.70, compared to 0.67 for Anand and Hardt's system.  ... 
doi:10.18653/v1/w18-5409 dblp:conf/emnlp/RonningHS18 fatcat:nijnqb6nufa5tetzu7rzfowrwq

Language Models Learn POS First

Naomi Saphra, Adam Lopez
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
Thus if a neural network relies heavily on a small number of cells in an activation pattern, the activation is very concentrated.  ...  ., 2017) , which interprets an arbitrary selection of neurons in terms of how they relate to another selection of neurons from any network run on the same input data.  ... 
doi:10.18653/v1/w18-5438 dblp:conf/emnlp/SaphraL18 fatcat:eeustyyfv5ahnpy74ttbtsnhq4

Iterative Recursive Attention Model for Interpretable Sequence Classification

Martin Tutek, Jan Šnajder
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
However, standard attention models are of limited interpretability for tasks that involve a series of inference steps.  ...  We train our model on sentiment classification datasets and demonstrate its capacity to identify and combine different aspects of the input in an easily interpretable manner, while obtaining performance  ...  Acknowledgment This research has been supported by the European Regional Development Fund under the grant KK.01.1.1.01.0009 (DATACROSS).  ... 
doi:10.18653/v1/w18-5427 dblp:conf/emnlp/TutekS18 fatcat:o4n7qtulhnah3crkrgeg5hmrey

Probing sentence embeddings for structure-dependent tense

Geoff Bacon, Terry Regier
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
First, we train autoencoders for English, Spanish, French and Italian where both the encoder and decoder are either Simple Recurrent Networks (SRNs, Elman, 1990) or Long Short-Term Memory networks (LSTMs  ...  A prominent and successful approach is to train recurrent neural networks (RNNs) to encode sentences into fixed length vectors (Conneau et al., 2018; Nie et al., 2017) .  ... 
doi:10.18653/v1/w18-5440 dblp:conf/emnlp/BaconR18 fatcat:2zkdkzeixbaifeyvbidaid5yl4

Screening Gender Transfer in Neural Machine Translation [article]

Guillaume Wisniewski, Lichao Zhu, Nicolas Ballier, François Yvon
2022 arXiv   pre-print
Our results show that gender information can be found in all token representations built by the encoder and the decoder and lead us to conclude that there are multiple pathways for gender transfer.  ...  on the internal representations used in the MT system.  ...  Acknowledgements This work was partially funded by the NeuroViz project (Explorations and visualizations of a neural translation system), supported by the Ile-de-France Region within the DIM RFSI 2020  ... 
arXiv:2202.12568v1 fatcat:j7eca73b3fgs5ems6ca6xuzoke

An Operation Sequence Model for Explainable Neural Machine Translation

Felix Stahlberg, Danielle Saunders, Bill Byrne
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
Our technique can outperform a plain text system in terms of BLEU score under the recent Transformer architecture on Japanese-English and Portuguese-English, and is within 0.5 BLEU difference on Spanish-English  ...  Word reordering is modeled by operations which allow setting markers in the target sentence and move a target-side write head between those markers.  ...  We thank Joanna Stadnik who produced the recurrent translation and alignment models during her 4th year project.  ... 
doi:10.18653/v1/w18-5420 dblp:conf/emnlp/StahlbergSB18 fatcat:dwkt3parvfdljgdp7ewwv7aovm

Learning and Evaluating Sparse Interpretable Sentence Embeddings

Valentin Trifonov, Octavian-Eugen Ganea, Anna Potapenko, Thomas Hofmann
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
We observe an increase in interpretability compared to dense models, on a dataset of movie dialogs and on the scene descriptions from the MS COCO dataset.  ...  We further introduce a novel, quantitative and automated evaluation metric for sentence embedding interpretability, based on topic coherence methods.  ...  To be precise, the author focuses on encoder-decoder neural networks and uses sparse coding to recover interpretable features in the latent spaces of a variational autoencoder (Kingma and Welling, 2013  ... 
doi:10.18653/v1/w18-5422 dblp:conf/emnlp/TrifonovGPH18 fatcat:d5sws3e26jhzllgqb33hlgucau

Analyzing Learned Representations of a Deep ASR Performance Prediction Model

Zied Elloumi, Laurent Besacier, Olivier Galibert, Benjamin Lecouteux
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
This paper addresses a relatively new task: prediction of ASR performance on unseen broadcast programs.  ...  This work is dedicated to the analysis of speech signal embeddings and text embeddings learnt by the CNN while training our prediction model.  ...  This lack of interpretability of the representations learned by deep neural networks is a 1 https://github.com/hlt-mt/TranscRater general problem in AI.  ... 
doi:10.18653/v1/w18-5402 dblp:conf/emnlp/ElloumiBGL18 fatcat:fyzhgzywrrce7hvtggngyly2fa

Introspection for convolutional automatic speech recognition

Andreas Krug, Sebastian Stober
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
Artificial Neural Networks (ANNs) have experienced great success in the past few years. The increasing complexity of these models leads to less understanding about their decision processes.  ...  Our method integrates information from many data examples through local introspection techniques for Convolutional Neural Networks (CNNs).  ...  Acknowledgments This research has been funded by the Federal Ministry of Education and Research of Germany (BMBF) and supported by the donation of a GeForce GTX Titan X graphics card from the NVIDIA Corporation  ... 
doi:10.18653/v1/w18-5421 dblp:conf/emnlp/KrugS18 fatcat:jv5pl3qptrebbix2kriz733z3m

Rearranging the Familiar: Testing Compositional Generalization in Recurrent Networks

João Loula, Marco Baroni, Brenden Lake
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
Systematic compositionality is the ability to recombine meaningful units with regular and predictable outcomes, and it's seen as key to the human capacity for generalization in language.  ...  Recent work (Lake and Baroni, 2018) has studied systematic compositionality in modern seq2seq models using generalization to novel navigation instructions in a grounded environment as a probing tool.  ...  Acknowledgments We'd like to thank Joost Bastings, Kyunghyun Cho and Jason Weston for helpful comments and conversations.  ... 
doi:10.18653/v1/w18-5413 dblp:conf/emnlp/LoulaBL18 fatcat:bs2gt3nllrbzhfgyayko3u5qhq

On the Role of Text Preprocessing in Neural Network Architectures: An Evaluation Study on Text Categorization and Sentiment Analysis

Jose Camacho-Collados, Mohammad Taher Pilehvar
2018 Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP  
Text preprocessing is often the first step in the pipeline of a Natural Language Processing (NLP) system, with potential impact in its final performance.  ...  In this paper we investigate the impact of simple text preprocessing decisions (particularly tokenizing, lemmatizing, lowercasing and multiword grouping) on the performance of a standard neural text classifier  ...  Acknowledgments Jose Camacho-Collados is supported by the ERC Starting Grant 637277.  ... 
doi:10.18653/v1/w18-5406 dblp:conf/emnlp/Camacho-Collados18 fatcat:tzh57m2hdzb25d6fyzsu6kq7vi
« Previous Showing results 1 — 15 out of 257 results