Publications

2025

1. 

   Feedback Friction: LLMs Struggle to Fully Incorporate External Feedback, NeurIPS

   Dongwei Jiang , Alvin Zhang , Andrew Wang , Nicholas Andrews , and Daniel Khashabi

   CoRR, 2025

   Abstract PDF

   Recent studies have shown LLMs possess some ability to improve their responses when given external feedback. However, it remains unclear how effectively and thoroughly these models can incorporate extrinsic feedback. In an ideal scenario, if LLMs receive near-perfect and complete feedback, we would expect them to fully integrate the feedback and reach correct solutions. In this paper, we systematically investigate LLMs’ ability to incorporate feedback by designing a controlled experimental environment. For each problem, a solver model attempts a solution, then a feedback generator with access to near-complete ground-truth answers produces targeted feedback, after which the solver tries again. We evaluate this pipeline across a diverse range of tasks, including math reasoning, knowledge reasoning, scientific reasoning, and general multi-domain evaluations with state-of-the-art language models including Claude 3.7 with extended thinking. Surprisingly, even under these near-ideal conditions, solver models consistently show resistance to feedback, a limitation that we term FEEDBACK FRICTION. To mitigate this limita- tion, we experiment with sampling-based strategies like progressive temperature increases and explicit rejection of previously attempted incorrect answers, which yield improvements but still fail to help models achieve target performance. We analyze FEEDBACK FRICTION and find that models’ confidence on specific questions, measured by semantic entropy, predicts feedback resistance: high-confidence predictions remain resistant to external correction. We hope that highlighting this issue in LLMs will help future research in self-improvement.

2. 

   RATIONALYST: Pre-training Process-Supervision for Improving Reasoning, ACL

   Dongwei Jiang , Guoxuan Wang , Yining Lu , Andrew Wang , Jingyu Zhang , Chuyu Liu , Benjamin Van Durme , and Daniel Khashabi

   In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Jul 2025

   Abstract PDF

   The reasoning steps generated by LLMs might be incomplete, as they mimic logical leaps common in everyday communication found in their pre-training data: underlying rationales are frequently left implicit (unstated). To address this challenge, we introduce RATIONALYST, a model for process-supervision of reasoning based on pre-training on a vast collection of rationale annotations extracted from unlabeled data. We extract 79k rationales from web-scale unlabelled dataset (the Pile) and a combination of reasoning datasets with minimal human intervention. This web-scale pre-training for reasoning allows RATIONALYST to consistently generalize across diverse reasoning tasks, including mathematical, commonsense, scientific, and logical reasoning. Fine-tuned from LLaMa-3-8B, RATIONALYST improves the accuracy of reasoning by an average of 3.9% on 7 representative reasoning benchmarks. It also demonstrates superior performance compared to significantly larger verifiers like GPT-4 and similarly sized models fine-tuned on matching training sets.

3. 

   To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning, ICLR

   Zayne Rea Sprague , Fangcong Yin , Juan Diego Rodriguez ,  Dongwei Jiang , Manya Wadhwa , Prasann Singhal , Xinyu Zhao , Xi Ye , Kyle Mahowald , and Greg Durrett

   In The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025, Jul 2025

   Abstract PDF

   Chain-of-thought (CoT) via prompting is the de facto method for eliciting reasoning capabilities from large language models (LLMs). But for what kinds of tasks is this extra “thinking” really helpful? To analyze this, we conducted a quantitative meta-analysis covering over 100 papers using CoT and ran our own evaluations of 20 datasets across 14 models. Our results show that CoT gives strong performance benefits primarily on tasks involving math or logic, with much smaller gains on other types of tasks. On MMLU, directly generating the answer without CoT leads to almost identical accuracy as CoT unless the question or model’s response contains an equals sign, indicating symbolic operations and reasoning. Following this finding, we analyze the behavior of CoT on these problems by separating planning and execution and comparing against tool-augmented LLMs. Much of CoT’s gain comes from improving symbolic execution, but it underperforms relative to using a symbolic solver. Our results indicate that CoT can be applied selectively, maintaining performance while saving inference costs. Furthermore, they suggest a need to move beyond prompt-based CoT to new paradigms that better leverage intermediate computation across the whole range of LLM applications.

4. SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responses, AAAI

   Dongwei Jiang , Jingyu Zhang , Orion Weller , Nathaniel Weir , Benjamin Van Durme , and Daniel Khashabi

   In Proceedings of the AAAI Conference on Artificial Intelligence, Jul 2025

   Abstract PDF

   Can LLMs consistently improve their previous outputs for better results? For this to be true, LLMs would need to be better at discriminating among previously-generated alternatives, than generating initial responses. We explore the validity of this hypothesis in practice. We first formulate a unified framework that allows us to compare the generative and discriminative capability of any model on any task. In our resulting experimental analysis of several open-source and industrial LLMs, we observe that models are not reliably better at discriminating among previously-generated alternatives than generating initial responses. This finding challenges the notion that LLMs may be able to enhance their performance only through their own judgment.

2024

1. 

   LeanReasoner: Boosting Complex Logical Reasoning with Lean, NAACL

   Dongwei Jiang , Marcio Fonseca , and Shay B. Cohen

   In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), NAACL 2024, Mexico City, Mexico, June 16-21, 2024, Jul 2024

   Abstract PDF

   Large language models (LLMs) often struggle with complex logical reasoning due to logical inconsistencies and the inherent difficulty of such reasoning. We use Lean, a theorem proving framework, to address these challenges. By formalizing logical reasoning problems into theorems within Lean, we can solve them by proving or disproving the corresponding theorems. This method reduces the risk of logical inconsistencies with the help of Lean’s symbolic solver. It also enhances our ability to treat complex reasoning tasks by using Lean’s extensive library of theorem proofs. Our method achieves state-of-the-art performance on the FOLIO dataset and achieves performance near this level on ProofWriter. Notably, these results were accomplished by fine-tuning on fewer than 100 in-domain samples for each dataset.

2. Enhancing Systematic Decompositional Natural Language Inference Using Informal Logic, EMNLP

   Nathaniel Weir , Kate Sanders , Orion Weller , Shreya Sharma ,  Dongwei Jiang , Zhengping Jiang , Bhavana Dalvi Mishra , Oyvind Tafjord , Peter Jansen , Peter Clark , and Benjamin Van Durme

   In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, Nov 2024

   Abstract PDF

   Recent language models enable new opportunities for structured reasoning with text, such as the construction of intuitive, proof-like textual entailment trees without relying on brittle formal logic. However, progress in this direction has been hampered by a long-standing lack of a clear protocol for determining what _valid decompositional entailment_ is. This absence causes noisy datasets and limited performance gains by modern neuro-symbolic entailment engines. To address these problems, we formulate a consistent and theoretically grounded approach to annotating decompositional entailment and evaluate its impact on LLM-based textual inference. We find that our new dataset, RDTE (Recognizing Decompositional Textual Entailment), has a substantially higher internal consistency than prior decompositional entailment datasets, suggesting that RDTE is a significant step forward in the long-standing problem of forming a clear protocol for discerning entailment. We also find that training an RDTE-oriented entailment classifier via knowledge distillation and employing it in an entailment tree reasoning engine significantly improves both accuracy and proof quality, illustrating the practical benefit of this advance for textual inference.

3. Benchmarking Language Model Creativity: A Case Study on Code Generation, NAACL

   Yining Lu , Dixuan Wang , Tianjian Li ,  Dongwei Jiang , and Daniel Khashabi

   CoRR, Nov 2024

   Abstract PDF

   As LLMs become increasingly prevalent, it is interesting to consider how “creative” these models can be. From cognitive science, creativity consists of at least two key characteristics: \emphconvergent thinking (purposefulness to achieve a given goal) and \emphdivergent thinking (adaptability to new environments or constraints) \citeprunco2003critical. In this work, we introduce a framework for quantifying LLM creativity that incorporates the two characteristics. This is achieved by (1) Denial Prompting pushes LLMs to come up with more creative solutions to a given problem by incrementally imposing new constraints on the previous solution, compelling LLMs to adopt new strategies, and (2) defining and computing the NeoGauge metric which examines both convergent and divergent thinking in the generated creative responses by LLMs. We apply the proposed framework on Codeforces problems, a natural data source for collecting human coding solutions. We quantify NeoGauge for various proprietary and open-source models and find that even the most creative model, GPT-4, still falls short of demonstrating human-like creativity. We also experiment with advanced reasoning strategies (MCTS, self-correction, etc.) and observe no significant improvement in creativity. As a by-product of our analysis, we release NeoCoder dataset for reproducing our results on future models.

2021

1. A Further Study of Unsupervised Pretraining for Transformer Based Speech Recognition, ICASSP

   Dongwei Jiang , Wubo Li , Ruixiong Zhang , Miao Cao , Ne Luo , Yang Han , Wei Zou , Kun Han , and Xiangang Li

   In IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2021, Toronto, ON, Canada, June 6-11, 2021, Nov 2021

   Abstract PDF

   Building a good speech recognition system usually requires large amounts of transcribed data, which is expensive to collect. To tackle this problem, many unsupervised pre-training methods have been proposed. Among these methods, Masked Predictive Coding achieved significant improvements on various speech recognition datasets with BERT-like Masked Reconstruction loss and Transformer backbone. However, many aspects of MPC have not been fully investigated. In this paper, we conduct a further study on MPC and focus on three important aspects: the effect of pre-training data speaking style, its extension on streaming model, and how to better transfer learned knowledge from pre-training stage to downstream tasks. Experiments reveled that pre-training data with matching speaking style is more useful on downstream recognition tasks. A unified training objective with APC and MPC provided 8.46% relative error reduction on streaming model trained on HKUST. Also, the combination of target data adaption and layer-wise discriminative training helped the knowledge transfer of MPC, which achieved 3.99% relative error reduction on AISHELL over a strong baseline.

2. Transformer Based Unsupervised Pre-Training for Acoustic Representation Learning, ICASSP

   Ruixiong Zhang , Haiwei Wu , Wubo Li ,  Dongwei Jiang , Wei Zou , and Xiangang Li

   In IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2021, Toronto, ON, Canada, June 6-11, 2021, Nov 2021

   Abstract PDF

   Recently, a variety of acoustic tasks and related applications arised. For many acoustic tasks, the labeled data size may be limited. To handle this problem, we propose an unsupervised pre-training method using Transformer based encoder to learn a general and robust high-level representation for all acoustic tasks. Experiments have been conducted on three kinds of acoustic tasks: speech emotion recognition, sound event detection and speech translation. All the experiments have shown that pre-training using its own training data can significantly improve the performance. With a larger pre-training data combining MuST-C, Librispeech and ESC-US datasets, for speech emotion recognition, the UAR can further improve absolutely 4.3% on IEMOCAP dataset. For sound event detection, the F1 score can further improve absolutely 1.5% on DCASE2018 task5 development set and 2.1% on evaluation set. For speech translation, the BLEU score can further improve relatively 12.2% on En-De dataset and 8.4% on En-Fr dataset.

3. Didispeech: A Large Scale Mandarin Speech Corpus, InterSpeech

   Tingwei Guo , Cheng Wen ,  Dongwei Jiang , Ne Luo , Ruixiong Zhang , Shuaijiang Zhao , Wubo Li , Cheng Gong , Wei Zou , Kun Han , and Xiangang Li

   In IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2021, Toronto, ON, Canada, June 6-11, 2021, Nov 2021

   Abstract PDF

   This paper introduces a new open-sourced Mandarin speech corpus, called DiDiSpeech. It consists of about 800 hours of speech data at 48kHz sampling rate from 6000 speakers and the corresponding texts. All speech data in the corpus is recorded in quiet environment and is suitable for various speech processing tasks, such as voice conversion, multi-speaker text-to-speech and automatic speech recognition. We conduct experiments with multiple speech tasks and evaluate the performance, showing that it is promising to use the corpus for both academic research and practical application. The corpus is available at this https URL.

4. Speech SimCLR: Combining Contrastive and Reconstruction Objective for Self-Supervised Speech Representation Learning, InterSpeech

   Dongwei Jiang , Wubo Li , Miao Cao , Wei Zou , and Xiangang Li

   In 22nd Annual Conference of the International Speech Communication Association, Interspeech 2021, Brno, Czechia, August 30 - September 3, 2021, Nov 2021

   Abstract PDF

   Self-supervised visual pretraining has shown significant progress recently. Among those methods, SimCLR greatly advanced the state of the art in self-supervised and semi-supervised learning on ImageNet. The input feature representations for speech and visual tasks are both continuous, so it is natural to consider applying similar objective on speech representation learning. In this paper, we propose Speech SimCLR, a new self-supervised objective for speech representation learning. During training, Speech SimCLR applies augmentation on raw speech and its spectrogram. Its objective is the combination of contrastive loss that maximizes agreement between differently augmented samples in the latent space and reconstruction loss of input representation. The proposed method achieved competitive results on speech emotion recognition and speech recognition.

2020

1. TMT: A Transformer-Based Modal Translator for Improving Multimodal Sequence Representations in Audio Visual Scene-Aware Dialog, InterSpeech

   Wubo Li ,  Dongwei Jiang , Wei Zou , and Xiangang Li

   In 21st Annual Conference of the International Speech Communication Association, Interspeech 2020, Virtual Event, Shanghai, China, October 25-29, 2020, Nov 2020

   Abstract PDF

   Audio Visual Scene-aware Dialog (AVSD) is a task to generate responses when discussing about a given video. The previous state-of-the-art model shows superior performance for this task using Transformer-based architecture. However, there remain some limitations in learning better representation of modalities. Inspired by Neural Machine Translation (NMT), we propose the Transformer-based Modal Translator (TMT) to learn the representations of the source modal sequence by translating the source modal sequence to the related target modal sequence in a supervised manner. Based on Multimodal Transformer Networks (MTN), we apply TMT to video and dialog, proposing MTN-TMT for the video-grounded dialog system. On the AVSD track of the Dialog System Technology Challenge 7, MTN-TMT outperforms the MTN and other submission models in both Video and Text task and Text Only task. Compared with MTN, MTN-TMT improves all metrics, especially, achieving relative improvement up to 14.1% on CIDEr. Index Terms: multimodal learning, audio-visual scene-aware dialog, neural machine translation, multi-task learning

2019

1. Improving Transformer-based Speech Recognition Using Unsupervised Pre-training, arxiv

   Dongwei Jiang , Xiaoning Lei , Wubo Li , Ne Luo , Yuxuan Hu , Wei Zou , and Xiangang Li

   CoRR, Nov 2019

   Abstract PDF

   Speech recognition technologies are gaining enormous popularity in various industrial applications. However, building a good speech recognition system usually requires large amounts of transcribed data, which is expensive to collect. To tackle this problem, an unsupervised pre-training method called Masked Predictive Coding is proposed, which can be applied for unsupervised pre-training with Transformer based model. Experiments on HKUST show that using the same training data, we can achieve CER 23.3%, exceeding the best end-to-end model by over 0.2% absolute CER. With more pre-training data, we can further reduce the CER to 21.0%, or a 11.8% relative CER reduction over baseline.

2018

1. Comparable Study Of Modeling Units For End-To-End Mandarin Speech Recognition, ISCSLP

   Wei Zou ,  Dongwei Jiang , Shuaijiang Zhao , Guilin Yang , and Xiangang Li

   In 11th International Symposium on Chinese Spoken Language Processing, ISCSLP 2018, Taipei City, Taiwan, November 26-29, 2018, Nov 2018

   Abstract PDF

   End-To-End speech recognition have become increasingly popular in mandarin speech recognition and achieved delightful performance. Mandarin is a tonal language which is different from English and requires special treatment for the acoustic modeling units. There have been several different kinds of modeling units for mandarin such as phoneme, syllable and Chinese character. In this work, we explore two major end-to-end models: connectionist temporal classification (CTC) model and attention based encoder-decoder model for mandarin speech recognition. We compare the performance of three different scaled modeling units: context dependent phoneme(CDP), syllable with tone and Chinese character. We find that all types of modeling units can achieve approximate character error rate (CER) in CTC model and the performance of Chinese character attention model is better than syllable attention model. Furthermore, we find that Chinese character is a reasonable unit for mandarin speech recognition. On DidiCallcenter task, Chinese character attention model achieves a CER of 5.68% and CTC model gets a CER of 7.29%, on the other DidiReading task, CER are 4.89% and 5.79%, respectively. Moreover, attention model achieves a better performance than CTC model on both datasets.

2. An Analysis of Decoding for Attention-Based End-to-End Mandarin Speech Recognition, ISCSLP

   Dongwei Jiang , Wei Zou , Shuaijiang Zhao , Guilin Yang , and Xiangang Li

   In 11th International Symposium on Chinese Spoken Language Processing, ISCSLP 2018, Taipei City, Taiwan, November 26-29, 2018, Nov 2018

   Abstract PDF

   Many of the current state-of-the-art Mandarin Large Vocabulary Continuous Speech Recognition (LVCSR) Systems are built either with a hybrid of Deep Neural Network (DNN) and Hidden Markov Models (HMM) or with Neural Network model trained with Connectionist Temporal Classification (CTC) criterion. In both of these models, decoding is conducted by Weighted Finite State Transducer (WFST) that searches the word sequence which matches best with the speech given the acoustic and language models. Recently, attention-based end-toend method is becoming more and more popular for Mandarin speech recognition. This new method advocates replacing complex data processing pipelines with a single neural network trained in an end-to-end fashion. In this paper, we investigate the decoding process for attention-based Mandarin models using syllable and character as acoustic modeling units and discuss how to combine word information into the decoding process. We also conduct a detailed analysis on various factors that affect the performance of decoding including beam size, label smoothing, softmax temperature, attention smoothing and coverage.

3. Towards End-to-End Code-Switching Speech Recognition, arxiv

   Ne Luo ,  Dongwei Jiang , Shuaijiang Zhao , Caixia Gong , Wei Zou , and Xiangang Li

   CoRR, Nov 2018

   Abstract PDF

   Code-switching speech recognition has attracted an increasing interest recently, but the need for expert linguistic knowledge has always been a big issue. End-to-end automatic speech recognition (ASR) simplifies the building of ASR systems considerably by predicting graphemes or characters directly from acoustic input. In the mean time, the need of expert linguistic knowledge is also eliminated, which makes it an attractive choice for code-switching ASR. This paper presents a hybrid CTC-Attention based end-to-end Mandarin-English code-switching (CS) speech recognition system and studies the effect of hybrid CTC-Attention based models, different modeling units, the inclusion of language identification and different decoding strategies on the task of code-switching ASR. On the SEAME corpus, our system achieves a mixed error rate (MER) of 34.24%.