Papers by Ben Peters
Sparse Sequence-to-Sequence Models (P19-1)
Copied to clipboard
| Challenge: | Sequence-to-sequence models are dense and assigning nonzero probability to implausible outputs. |
| Approach: | They propose a new family of -entmax transformations that includes softmax and sparsemax as particular cases and is sparser for any > 1 . they provide fast algorithms to evaluate these transformations and their gradients, which scale well for large vocabulary sizes. |
| Outcome: | The proposed models are able to produce sparse alignments and assign nonzero probability to short list of plausible outputs, sometimes rendering beam search exact. |
Did Translation Models Get More Robust Without Anyone Even Noticing? (2025.acl-long)
Copied to clipboard
| Challenge: | Neural machine translation models are highly sensitive to “noisy” inputs, such as spelling errors, abbreviations, and formatting issues. |
| Approach: | They revisit this insight in light of recent multilingual MT models and large language models applied to machine translation. |
| Outcome: | The proposed models perform better on clean data than previous models, but none of the open models use robustness techniques. |
From Tower to Spire: Adding the Speech Modality to a Translation-Specialist LLM (2025.findings-emnlp)
Copied to clipboard
Kshitij Ambilduke, Ben Peters, Sonal Sannigrahi, Anil Keshwani, Tsz Kin Lam, Bruno Martins, Andre Martins, Marcely Zanon Boito
| Challenge: | Spire is a speech-augmented language model capable of translating speech input into 10 languages and transcribing text input in both directions. |
| Approach: | They introduce a speech-augmented language model capable of translating speech input into 10 languages . they integrate the model into existing multilingual LMs via speech discretization . |
| Outcome: | Spire integrates speech-augmented language model into existing multilingual model using speech discretization and pre-training using only 42.5 K hours of speech. |
Smoothing and Shrinking the Sparse Seq2Seq Search Space (2021.naacl-main)
Copied to clipboard
| Challenge: | entmax-based sparse sequence-to-sequence models give high scores to short hypotheses . ent max models can shrink the search space by assigning zero probability to bad hypothese . |
| Approach: | They propose entmax-based sparse sequence-to-sequence models that minimize cross-entropy and use softmax to compute local normalized probabilities over target sequences. |
| Outcome: | The proposed models remove a major source of model error for word-level tasks . the proposed models improve cross-lingual morphological inflection and machine translation . |