Papers by John DeNero
Measuring Immediate Adaptation Performance for Neural Machine Translation (N19-1)
Copied to clipboard
| Challenge: | Incremental domain adaptation improves interactive machine translation performance . users of interactive systems are sensitive to the speed of adaptation . |
| Approach: | They propose to measure the speed of lexical acquisition for in-domain vocabulary . they propose to use this to choose the most suitable adaptation method for neural machine translation . |
| Outcome: | The proposed measures measure the speed of lexical acquisition for in-domain vocabulary . they show that the most suitable adaptation method is chosen from a range of different techniques . |
Automatic Correction of Human Translations (2022.naacl-main)
Copied to clipboard
| Challenge: | Despite recent advances in machine translation, a tremendous amount of translated content in the world is still written by humans. |
| Approach: | They propose a task of translation error correction (TEC) that corrects human-generated translations by correcting all errors in a source sentence and a human-created translation. |
| Outcome: | The proposed system improves translation accuracy by 5.1 points compared to MT systems with human errors . |
End-to-End Neural Word Alignment Outperforms GIZA++ (2020.acl-main)
Copied to clipboard
| Challenge: | Word alignment was once a core unsupervised learning task in natural language processing . but word alignment still plays an important role in interactive applications of neural machine translation, such as annotation transfer and lexicon injection. |
| Approach: | They propose to use a Transformer model to train an unsupervised word alignment model. |
| Outcome: | The proposed method outperforms GIZA++ on three data sets and is tightly integrated and does not affect translation quality. |
Putting the Con in Context: Identifying Deceptive Actors in the Game of Mafia (2022.naacl-main)
Copied to clipboard
| Challenge: | Using neural networks, we can model the impact of speaker role on language use through the game of Mafia. |
| Approach: | They analyze the effect of speaker role on language use through the game of Mafia, in which players are assigned either an honest or a deceptive role. |
| Outcome: | The proposed model outperforms a standard BERT-based text classification approach on two auxiliary tasks and identifies features that distinguish between player roles. |
A Streaming Approach For Efficient Batched Beam Search (2020.emnlp-main)
Copied to clipboard
| Challenge: | During decoding, candidates terminate or are pruned according to heuristics, a streaming method is used to "refill" the batch after it finishes translating some fraction of the current batch. |
| Approach: | They propose an efficient batching strategy for variable-length decoding on GPU architectures by streamlining the batching process. |
| Outcome: | The proposed method reduces runtime by 71% compared to a fixed-width beam search baseline and 17% compared with a variable-widness baseline while matching baselines’ BLEU. |
Automatic Bilingual Markup Transfer (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing work on markup transfer is performed with machine translation . a human translator generates the target translation without markup, and then the system infers the placement of markup tags. |
| Approach: | They propose two metrics and evaluate several approaches to bilingual markup transfer . best approach achieves an average accuracy of 94.7% across six language pairs . |
| Outcome: | The proposed approach achieves an average accuracy of 94.7% across six language pairs . it is a novel approach that can be applied to a structured document translation corpus . |
Compact Personalized Models for Neural Machine Translation (D18-1)
Copied to clipboard
| Challenge: | a large proportion of model parameters can be frozen during adaptation with minimal or no reduction in translation quality. |
| Approach: | They propose gradient-based domain adaptation methods for self-attentive machine translation models . they encourage structured sparsity in the set of offset tensors during learning . |
| Outcome: | The proposed method achieves high space and time efficiency using sparse models . the results compare the proposed method with incremental adaptation . |