Papers by Boris Ginsburg

9 papers
VoiceTextBlender: Augmenting Large Language Models with Speech Capabilities via Single-Stage Joint Speech-Text Supervised Fine-Tuning (2025.naacl-long)

Copied to clipboard

Challenge: Recent studies have augmented large language models (LLMs) with speech capabilities, leading to the development of speech language models.
Approach: They propose a single-stage joint speech-text SFT approach for training SpeechLMs . their model combines text-only SFT data with three types of speech-related data .
Outcome: The proposed model outperforms previous SpeechLMs on speech-based QA tasks while maintaining original speech-only capabilities.
Scaling Test-Time Compute to Achieve IOI Gold Medal with Open-Weight Models (2026.acl-long)

Copied to clipboard

Challenge: Competitive programming has become a rigorous benchmark for evaluating the reasoning and problem-solving capabilities of large language models (LLMs).
Approach: They propose a scalable and reproducible test-time compute framework that achieves IOI gold-level performance using open-weight models.
Outcome: The proposed framework achieves IOI gold-level performance using open-weight models . it scales consistently with available compute, narrowing the gap between open and closed systems.
SWAN: An Efficient and Scalable Approach for Long-Context Language Modeling (2025.emnlp-main)

Copied to clipboard

Challenge: Existing decoder-only models struggle with context lengths beyond their training distribution.
Approach: They propose a causal Transformer architecture that generalizes robustly to sequence lengths longer than seen during training.
Outcome: The proposed decoder-only architecture can generalize robustly to longer contexts . it is more computationally efficient than the standard Transformer architecture, the authors say .
Genetic Instruct: Scaling up Synthetic Generation of Coding Instructions for Large Language Models (2025.acl-industry)

Copied to clipboard

Challenge: Large Language Models (LLMs) require high quality instruction data for effective alignment, especially in code generation tasks where expert curated datasets are expensive to produce.
Approach: They propose a scalable algorithm for synthesizing large-scale, high quality coding instructions using evolutionary principles.
Outcome: The proposed approach generates 7.5 million coding instructions with a small seed population and is highly parallelizable and effective even with weaker generator models.
Hierarchical Policy Optimization for Simultaneous Translation of Unbounded Speech (2026.acl-long)

Copied to clipboard

Challenge: Existing synthesis methods cannot guarantee data quality.
Approach: They propose a hierarchical reward that balances translation quality and latency objectives by combining supervised fine-tuning data with supervised inputs.
Outcome: The proposed model can reuse key-value caches across both modalities and eliminate redundant feature recomputation.
Speech-Hands: A Self-Reflection Voice Agentic Approach to Speech Recognition and Audio Reasoning with Omni Perception (2026.acl-long)

Copied to clipboard

Challenge: naively fine-tuning an omni-model on speech recognition and external sound understanding tasks often degrades performance . Xie and Wu's framework, Speech-Hands, recasts the problem as an explicit self-reflection decision.
Approach: They propose a voice-agentic framework that learns one critical omni-understanding skill: trusting itself versus external audio perception.
Outcome: The proposed framework outperforms baseline models on the OpenASR leaderboard by 12.1% WER and high F1 on audio QA decisions.
Anticipating Future with Large Language Model for Simultaneous Machine Translation (2025.naacl-long)

Copied to clipboard

Challenge: Existing methods only use the partial utterance that has already arrived at the input and the generated hypothesis.
Approach: They propose to use a large language model to predict future source words and opportunistically translate without introducing too much risk.
Outcome: The proposed method outperforms baselines on four language directions and achieves the best translation quality-latency trade-off by up to 5 BLEU points at the same latency.
Extending Automatic Machine Translation Evaluation to Book-Length Documents (2025.emnlp-main)

Copied to clipboard

Challenge: Large Language Models (LLMs) have superior translation performance and long-context capabilities, but evaluation methodologies remain constrained to sentence-level assessment due to dataset limitations and token number restrictions in metrics.
Approach: They propose an evaluation scheme that extends existing automatic metrics to long-document translation by treating documents as continuous text and applying sentence segmentation and alignment methods.
Outcome: The proposed evaluation scheme outperforms existing long-form document evaluation schemes while accounting for under-/over-translations and varied sentence boundaries.
NeKo: Cross-Modality Post-Recognition Error Correction with Tasks-Guided Mixture-of-Experts Language Model (2025.acl-industry)

Copied to clipboard

Challenge: Existing methods to train a model on a mixture of domain datasets require separate correction language models.
Approach: They propose a multi-task correction MoE that trains experts to become an "expert" of speech-to-text, language-totext and vision-to text datasets by learning to route each dataset’s tokens to its mapped expert.
Outcome: The proposed model outperforms GPT-3.5 and Claude-3.5-Sonnet on the Open ASR Leaderboard and reaches an average relative 5.0% WER reduction and substantial improvements in BLEU scores.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations