mlsanityData to 14 Oct 2026

Scaling Attention Head Analysis via Gradient-Based Attribution in Context-Aware Machine Translation

arXiv 2609.28117 ↗First posted 23 Sep 2026cs.CL cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

In this paper, we introduce a gradient-based head attribution strategy where the Token-level Max-Margin loss is backpropagated to the attention maps. This framework enables a large-scale causal analysis of attention heads, making it suitable for LLMs. We evaluate our method on the task of disambiguation in Context-aware Machine Translation, where we analyze 50 phenomena across 4 models and 4 language directions. We empirically show the alignment of our method with the effects of increasing the attention scores of token-to-token relations on three models and two language directions, ensuring the robustness of our method. Our analysis reveals the presence of the "general-purpose" attention heads that improve the model's performance when attending to different relations. We find that the average attention a head assigns to a relation does not necessarily relate to the model's performance, which suggests that the models developed redundancies during training in terms of the head functions.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 53 of its 74 references that are in the map. First posted on arXiv 23 Sep 2026.

Builds on

53 of 74 references.

References · 53

  1. 2017Attention Is All You Needcited by 60,361
  2. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  3. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  4. 2018A Call for Clarity in Reporting BLEU Scorescited by 1,799
  5. 2017Learning Important Features Through Propagating Activation Differencescited by 1,458
  6. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,302
  7. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,165
  8. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,059
  9. 2020COMET: A Neural Framework for MT Evaluationcited by 1,033
  10. 2019What Does BERT Look At? An Analysis of BERT's Attentioncited by 1,003
  11. 2022In-context Learning and Induction Headscited by 964
  12. 2019Attention is not Explanationcited by 929
38 more
  1. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 856
  2. 2020Quantifying Attention Flow in Transformerscited by 775
  3. 2019Attention is not not Explanationcited by 636
  4. 2020Transformer Interpretability Beyond Attention Visualizationcited by 627
  5. 2019Analyzing the Structure of Attention in a Transformer Language Modelcited by 243
  6. 2020A Survey of the State of Explainable AI for Natural Language Processingcited by 206
  7. 2024Tower: An Open Multilingual Large Language Model for Translation-Related Taskscited by 206
  8. 2021Post-hoc Interpretability for Neural NLP: A Surveycited by 156
  9. 2019Interpreting and improving natural-language processing (in machines) with natural language-processing (in the brain)cited by 128
  10. 2018Identifying and Controlling Important Neurons in Neural Machine Translationcited by 111
  11. 2024EuroLLM: Multilingual Language Models for Europecited by 106
  12. 2017Neural Machine Translation with Extended Contextcited by 95
  13. 2022Interpreting Language Models with Contrastive Explanationscited by 92
  14. 2017Evaluating Discourse Phenomena in Neural Machine Translationcited by 87
  15. 2018Document-Level Neural Machine Translation with Hierarchical Attention Networkscited by 85
  16. 2021Cross-Attention is All You Need: Adapting Pretrained Transformers for Machine Translationcited by 74
  17. 2019Selective Attention for Context-aware Neural Machine Translationcited by 60
  18. 2024Adapting Large Language Models for Document-Level Machine Translationcited by 59
  19. 2021Incorporating Residual and Normalization Layers into Analysis of Masked Language Modelscited by 51
  20. 2021G-Transformer for Document-level Machine Translationcited by 46
  21. 2022Towards Opening the Black Box of Neural Machine Translation: Source and Target Interpretations of the Transformercited by 45
  22. 2023Quantifying Context Mixing in Transformerscited by 43
  23. 2019Context-Aware Monolingual Repair for Neural Machine Translationcited by 41
  24. 2020Gradient-based Analysis of NLP Models is Manipulablecited by 41
  25. 2017Does Neural Machine Translation Benefit from Larger Context?cited by 38
  26. 2024The Same But Different: Structural Similarities and Differences in Multilingual Language Modelingcited by 28
  27. 2020Long-Short Term Masking Transformer: A Simple but Effective Baseline for Document-level Neural Machine Translationcited by 27
  28. 2023Escaping the sentence-level paradigm in machine translationcited by 27
  29. 2020Diving Deep into Context-Aware Neural Machine Translationcited by 25
  30. 2024Efficiently Exploring Large Language Models for Document-Level Machine Translation with In-context Learningcited by 23
  31. 2023DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformerscited by 22
  32. 2022Learn To Remember: Transformer with Recurrent Memory for Document-Level Machine Translationcited by 16
  33. 2023Identifying Context-Dependent Translations for Evaluation Set Productioncited by 14
  34. 2022A baseline revisited: Pushing the limits of multi-segment models for context-aware translationcited by 11
  35. 2025GMAR: Gradient-Driven Multi-Head Attention Rollout for Vision Transformer Interpretabilitycited by 11
  36. 2016Context Gates for Neural Machine Translationcited by 10
  37. 2024On Measuring Context Utilization in Document-Level MT Systemscited by 9
  38. 2024Better Explain Transformers by Illuminating Important Informationcited by 8

50 most cited of 53 All 53 (JSON)

Built on by

Not cited yet.