Saliency-driven Word Alignment Interpretation for Neural Machine Translation

Ding, Shuoyang; Xu, Hainan; Koehn, Philipp

doi:10.18653/v1/w19-5201

Cited by 56 publications

(73 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, there has been a debate on whether attention can be used to explain model decisions (Serrano and Smith, 2019;Jain and Wallace, 2019;Wiegreffe and Pinter, 2019), we thus present additional analysis of our proposed method based on saliency maps (Ding et al, 2019). Saliency maps have been shown to better capture word alignment than attention probabilities in neural machine translation.…”

Section: Discussionmentioning

confidence: 99%

Topics to Avoid: Demoting Latent Confounds in Text Classification

Kumar¹,

Wintner²,

Smith³

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

View full text Add to dashboard Cite

Despite impressive performance on many text classification tasks, deep neural networks tend to learn frequent superficial patterns that are specific to the training data and do not always generalize well. In this work, we observe this limitation with respect to the task of native language identification. We find that standard text classifiers which perform well on the test set end up learning topical features which are confounds of the prediction task (e.g., if the input text mentions Sweden, the classifier predicts that the author's native language is Swedish). We propose a method that represents the latent topical confounds and a model which "unlearns" confounding features by predicting both the label of the input text and the confound; but we train the two predictors adversarially in an alternating fashion to learn a text representation that predicts the correct label but is less prone to using information about the confound. We show that this model generalizes better and learns features that are indicative of the writing style rather than the content. 1

show abstract

Section: Discussionmentioning

confidence: 99%

Topics to Avoid: Demoting Latent Confounds in Text Classification

Kumar¹,

Wintner²,

Smith³

et al. 2019

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferen

View full text Add to dashboard Cite

show abstract

“…Garg et al (2019) show that attention weights from the penultimate layer, i.e., l = L − 1, can induce the best alignments. Although simple to implement, this method fails to obtain satisfactory word alignments (Ding et al, 2019;Garg et al, 2019). First of all, instead of the relevance between y i and x j , W l i,j measures the relevance between decoder hidden state z l i and encoder output h j .…”

Section: Alignment By Attentionmentioning

confidence: 99%

“…Recently, there is a resurgence of interest in the community to study word alignments for the Transformer (Ding et al, 2019;Li et al, 2019). One simple solution is NAIVE-ATT, which induces word alignments from the attention weights between the encoder and decoder.…”

Section: Introductionmentioning

confidence: 99%

“…However, such schedule only captures noisy word alignments (Ding et al, 2019;Garg et al, 2019). One of the major problems is that it induces alignment before observing the to-be-aligned target token (Peter et al, 2017;Ding et al, 2019). Suppose for the same source sentence, there are two alternative translations that diverge at decoding step i, generating y i and y i which respectively correspond to different source words.…”

Section: Introductionmentioning

confidence: 99%

“…To alleviate this problem, some researchers modify the transformer architecture by adding alignment modules that predict the to-be-aligned target token (Zenkel et al, 2019(Zenkel et al, , 2020 or modify the training loss by designing an alignment loss computed with full target sentence (Garg et al, 2019;Zenkel et al, 2020). Others argue that using only attention weights is insufficient for generating clean word alignment and propose to induce alignments with feature importance measures, such as leaveone-out measures (Li et al, 2019) and gradientbased measures (Ding et al, 2019). However, all previous work induces alignment for target word y i at step i, when y i is the decoder output.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Accurate Word Alignment Induction from Neural Machine Translation

Chen¹,

Liu²,

Chen³

et al. 2020

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

View full text Add to dashboard Cite

Despite its original goal to jointly learn to align and translate, prior researches suggest that Transformer captures poor word alignments through its attention mechanism. In this paper, we show that attention weights DO capture accurate word alignments and propose two novel word alignment induction methods SHIFT-ATT and SHIFT-AET. The main idea is to induce alignments at the step when the to-be-aligned target token is the decoder input rather than the decoder output as in previous work. SHIFT-ATT is an interpretation method that induces alignments from the attention weights of Transformer and does not require parameter update or architecture change. SHIFT-AET extracts alignments from an additional alignment module which is tightly integrated into Transformer and trained in isolation with supervision from symmetrized SHIFT-ATT alignments. Experiments on three publicly available datasets demonstrate that both methods perform better than their corresponding neural baselines and SHIFT-AET significantly outperforms GIZA++ by 1.4-4.8 AER points. 1 * Corresponding author. Part of the work was done when Yun was in Huawei Noah's Ark Lab.1 Code can be found at https://github.com/ sufe-nlp/transformer-alignment.Source: das weiß ich . Dec. input: i understand this . Dec. output: i understand this .

show abstract