Filters








113 Hits in 5.1 sec

A Constrained Viterbi Relaxation for Bidirectional Word Alignment

Yin-Wen Chang, Alexander M. Rush, John DeNero, Michael Collins
2014 Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)  
(y * ) for some x, y that is valid set of the e→f alignment Y : set of the e→f alignment without the forward constraints Relaxed max-marginal valuesThe highest dual value of all alignments using the  ...  ω (i , i, j) = ω(i , i, j) − λ(i, j) + λ(i , j − 1) Standard Viterbi update for computing x(i, j), adding in the score of y (i, j) The Lagrangian Relaxation Algorithm Lagrangian dual is the upper bound  ... 
doi:10.3115/v1/p14-1139 dblp:conf/acl/ChangRDC14 fatcat:k33xvjr2ofdslasbwvlwqrhutu

Model-Based Aligner Combination Using Dual Decomposition

John DeNero, Klaus Macherey
2011 Annual Meeting of the Association for Computational Linguistics  
Our bidirectional model enforces a one-to-one phrase constraint while accounting for the uncertainty in the underlying directional models.  ...  Unsupervised word alignment is most often modeled as a Markov process that generates a sentence f conditioned on its translation e.  ...  Model Definition Our bidirectional model G = (V, D) is a globally normalized, undirected graphical model of the word alignment for a fixed sentence pair (e, f ).  ... 
dblp:conf/acl/DeNeroM11 fatcat:uxomlrt2t5bcrjqcq6r3wewi3a

Active Learning-Based Elicitation for Semi-Supervised Word Alignment

Vamshi Ambati, Stephan Vogel, Jaime G. Carbonell
2010 Annual Meeting of the Association for Computational Linguistics  
Our experiments show that by active selection of uncertain and informative links, we reduce the overall manual effort involved in elicitation of alignment link data for training a semisupervised word aligner  ...  Semi-supervised word alignment aims to improve the accuracy of automatic word alignment by incorporating full or partial manual alignments.  ...  The first author would like to thank Qin Gao for the semi-supervised word alignment software and help with running experiments.  ... 
dblp:conf/acl/AmbatiVC10 fatcat:acgcdlqpprecbo5jw47mf7axcm

On The Alignment Problem In Multi-Head Attention-Based Neural Machine Translation

Tamer Alkhouli, Gabriel Bretschner, Hermann Ney
2018 Proceedings of the Third Conference on Machine Translation: Research Papers  
We also propose alignment pruning to speed up decoding in alignment-based neural machine translation (ANMT), which speeds up translation by a factor of 1.8 without loss in translation performance.  ...  To study the effect of adding the alignment head, we simulate a dictionaryguided translation task, where the user wants to guide translation using pre-defined dictionary entries.  ...  Tamer Alkhouli was partly funded by the 2016 Google PhD fellowship for North America, Europe and the Middle East.  ... 
doi:10.18653/v1/w18-6318 dblp:conf/wmt/AlkhouliBN18 fatcat:wgombtvahre35ejxyeaxp6o3xe

On The Alignment Problem In Multi-Head Attention-Based Neural Machine Translation [article]

Tamer Alkhouli, Gabriel Bretschner, Hermann Ney
2018 arXiv   pre-print
We also propose alignment pruning to speed up decoding in alignment-based neural machine translation (ANMT), which speeds up translation by a factor of 1.8 without loss in translation performance.  ...  To study the effect of adding the alignment head, we simulate a dictionary-guided translation task, where the user wants to guide translation using pre-defined dictionary entries.  ...  Tamer Alkhouli was partly funded by the 2016 Google PhD fellowship for North America, Europe and the Middle East.  ... 
arXiv:1809.03985v1 fatcat:i6n6zubwwnd7djhdbtia5cu7ge

Unsupervised Word Alignment Using Frequency Constraint in Posterior Regularized EM

Hidetaka Kamigaito, Taro Watanabe, Hiroya Takamura, Manabu Okumura, Eiichiro Sumita
2016 Journal of Natural Language Processing  
In particular, function words are not trivial to align for grammatically different language pairs, such as Japanese and English.  ...  We discriminate a function word and a content word using word frequency in the same way as done by Setiawan, Kan, and Li (2007) .  ...  Acknowledgement We would like to thank the reviewers for their detailed comments and suggestions, which are helpful to improve our manuscripts.  ... 
doi:10.5715/jnlp.23.327 fatcat:myjlendguvgfte6ovr2c4h5qvi

Comparison of Lattice-Free and Lattice-Based Sequence Discriminative Training Criteria for LVCSR

Wilfried Michel, Ralf Schlüter, Hermann Ney
2019 Interspeech 2019  
In this work we present a memory efficient implementation of the forward-backward computation that allows us to use unigram word-level language models in the denominator calculation while still doing a  ...  This allows for a direct comparison of lattice-based and lattice-free sequence discriminative training criteria such as MMI and sMBR, both using the same language model during training.  ...  under the European Union's Horizon 2020 research and innovation programme (grant agreement No 694537, project "SEQCLAS" and Marie Skłodowska-Curie grant agreement No 644283, project "LISTEN") and from a  ... 
doi:10.21437/interspeech.2019-2254 dblp:conf/interspeech/MichelSN19 fatcat:j7woaudy45binfbrdrhwempga4

Towards Decoding as Continuous Optimization in Neural Machine Translation [article]

Cong Duy Vu Hoang, Gholamreza Haffari
2017 arXiv   pre-print
We propose a novel decoding approach for neural machine translation (NMT) based on continuous optimisation.  ...  The resulting constrained continuous optimisation problem is then tackled using gradient-based methods.  ...  Acknowledgments We thank the reviewers for valuable feedbacks and discussions.  ... 
arXiv:1701.02854v4 fatcat:rorgtdyaazetxettp7vg3orp4a

Acoustic modelling with CD-CTC-SMBR LSTM RNNS

Andrew, Hasim Sak, Felix de Chaumont Quitry, Tara Sainath, Kanishka Rao
2015 2015 IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU)  
We also investigate the latency of CTC models and show that constraining forward-backward alignment in training can reduce the delay for a real-time streaming speech recognition system.  ...  Finally we investigate transferring knowledge from one network to another through alignments.  ...  Forced-alignment is the process of finding the maximum-likelihood label sequence for the acoustic frames and gives labels for every frame either in {0, 1} for Viterbi alignment or in [0, 1] for Baum-Welch  ... 
doi:10.1109/asru.2015.7404851 dblp:conf/asru/SeniorSQSR15 fatcat:cdx2r37ggzddxi5albmcst2iie

Improving Human Text Comprehension through Semi-

Sebastian Gehrmann, Steven Layne, Franck Dernoncourt
2019 Proceedings of the 2019 Conference of the North  
Our compression approach is based on a Semi-Markov Conditional Random Field that leverages unsupervised word-representations such as ELMo or BERT, eliminating the need for a complex encoder-decoder architecture  ...  In particular, we present an extractive pipeline for section title generation by first selecting the most salient sentence and then applying deletion-based compression.  ...  Acknowledgments We are grateful for the helpful feedback from the three anonymous reviewers.  ... 
doi:10.18653/v1/n19-1168 dblp:conf/naacl/GehrmannLD19 fatcat:32vczkr5frgfvbmdpzxuh354nq

Improving Human Text Comprehension through Semi-Markov CRF-based Neural Section Title Generation [article]

Sebastian Gehrmann and Steven Layne and Franck Dernoncourt
2019 arXiv   pre-print
Our compression approach is based on a Semi-Markov Conditional Random Field that leverages unsupervised word-representations such as ELMo or BERT, eliminating the need for a complex encoder-decoder architecture  ...  In particular, we present an extractive pipeline for section title generation by first selecting the most salient sentence and then applying deletion-based compression.  ...  Acknowledgments We are grateful for the helpful feedback from the three anonymous reviewers.  ... 
arXiv:1904.07142v1 fatcat:zgnbk4kconcejoetmuzpvo5dya

Spotting handwritten words and REGEX using a two stage BLSTM-HMM architecture

Gautier Bideault, Luc Mioulet, Clément Chatelain, Thierry Paquet, Eric K. Ringger, Bart Lamiroy
2015 Document Recognition and Retrieval XXII  
In this article, we propose a hybrid model for spotting words and regular expressions (REGEX) in handwritten documents.  ...  The model is made of the state-of-the-art BLSTM (Bidirectional Long Short Time Memory) neural network for recognizing and segmenting characters, coupled with a HMM to build line models able to spot the  ...  As the transitions in the HMM meta models are ergodic, the Viterbi alignment will only be driven by the local classification of BLSTM-CTC.  ... 
doi:10.1117/12.2075796 dblp:conf/drr/BideaultMCP15 fatcat:tppirlqiore5xcxaokidyow4nm

Automatic Line Segmentation and Ground-Truth Alignment of Handwritten Documents

Theodore Bluche, Bastien Moysset, Christopher Kermorvant
2014 2014 14th International Conference on Frontiers in Handwriting Recognition  
In this paper, we present a method for the automatic segmentation and transcript alignment of documents, for which we only have the transcript at the document level.  ...  With the automatically segmented and annotated lines, we record a relative improvement in Word Error Rate of 35.6%.  ...  They perform Viterbi alignment with HMMs corresponding to the transcript, including spelling variants, and model replacement for unknown characters.  ... 
doi:10.1109/icfhr.2014.117 dblp:conf/icfhr/BlucheMK14 fatcat:j6gtaamrcrhexibqvzdoiqae4y

Towards Decoding as Continuous Optimisation in Neural Machine Translation

Cong Duy Vu Hoang, Gholamreza Haffari, Trevor Cohn
2017 Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing  
We propose a novel decoding approach for neural machine translation (NMT) based on continuous optimisation.  ...  We reformulate decoding, a discrete optimization problem, into a continuous problem, such that optimization can make use of efficient gradient-based techniques.  ...  Acknowledgments We thank the reviewers for valuable feedbacks and discussions.  ... 
doi:10.18653/v1/d17-1014 dblp:conf/emnlp/HoangHC17 fatcat:vr4gnxg4cfayfoepwfj4ihdily

High Order Recurrent Neural Networks for Acoustic Modelling [article]

Chao Zhang, Philip Woodland
2018 arXiv   pre-print
Speech recognition experiments using British English multi-genre broadcast (MGB3) data showed that the proposed HORNN architectures for rectified linear unit and sigmoid activation functions reduced word  ...  This paper addresses the vanishing gradient problem using a high order RNN (HORNN) which has additional connections from multiple previous time steps.  ...  The maximum parameter changes were constrained by update value clipping with a threshold of 0.32 for a minibatch with 800 samples.  ... 
arXiv:1802.08314v1 fatcat:rkwfnft5m5atddjs2qzs7krudy
« Previous Showing results 1 — 15 out of 113 results