Complex instruction-following with elaborate constraints is imperative for Large Language Models (LLMs). While existing methods have constructed data for complex instruction alignment, they all rely on a more advanced model, especially GPT-4, limiting their application. In this paper, we propose a Multi-granularity Self-Contrastive Training (MuSC) framework, to improve the complex instruction alignment without relying on a stronger model. Our method is conducted on both coarse and fine granularity. On coarse-granularity, we construct constraint-aware preference data based on instruction decomposition and recombination. On fine-granularity, we perform token-aware preference optimization with dynamic token-level supervision. Our method is evaluated on open-sourced models, and experiment results show our method achieves significant improvement on both complex and general instruction-following benchmarks, surpassing previous self-alignment methods.

New LLM benchmarks are important to align with the rapid development of Large Language Models (LLMs). In this work, we present Chinese SimpleQA, the first comprehensive Chinese benchmark to evaluate the factuality ability of LLMs to answer short questions, and Chinese SimpleQA mainly has five properties (i.e., Chinese, Diverse, High-quality, Static, Easy-to-evaluate). Specifically, first, we focus on the Chinese language over 6 major topics with 99 diverse subtopics. Second, we conduct a comprehensive quality control process to achieve high-quality questions and answers, where the reference answers are static and cannot be changed over time. Third, following SimpleQA, the questions and answers are very short, and the grading process is easy-to-evaluate. Based on Chinese SimpleQA, we perform a comprehensive evaluation of the factuality abilities of existing LLMs. Finally, we hope that Chinese SimpleQA could guide the developers to better understand the Chinese factuality abilities of their models and facilitate the growth of LLMs.

Recent advancements in Direct Preference Optimization (DPO) have significantly enhanced the alignment of Large Language Models (LLMs) with human preferences, owing to its simplicity and effectiveness. However, existing methods typically optimize a scalar score or ranking reward, thereby overlooking the multi-dimensional nature of human preferences. In this work, we propose to extend the preference of DPO to two dimensions: segments and aspects. We first introduce a 2D supervision dataset called HelpSteer-2D. For the segment dimension, we divide the response into sentences and assign scores to each segment. For the aspect dimension, we meticulously design several criteria covering the response quality rubrics. With the 2-dimensional signals as feedback, we develop a 2D-DPO framework, decomposing the overall objective into multi-segment and multi-aspect objectives. Extensive experiments on popular benchmarks demonstrate that 2D-DPO performs better than methods that optimize for scalar or 1-dimensional preferences.

Recently, there has been a growing trend of utilizing Large Language Model (LLM) to evaluate the quality of other LLMs. Many studies have fine-tuned judge models based on open-source LLMs for evaluation. While the fine-tuned judge models are claimed to achieve comparable evaluation capability with GPT-4, in this work, we conduct an empirical study of LLM-as-a-Judge. Our findings indicate that although the fine-tuned judge models achieve high performance on in-domain test sets, even surpassing GPT-4, they underperform GPT-4 across several dimensions, including generalizability, fairness and adaptability. We also reveal that the fine-tuned judge model inherently operates as a task-specific classifier, consequently imposing the limitations.

Semantic text representation is a fundamental task in the field of natural language processing. Existing text embedding (e.g., SimCSE and LLM2Vec) have demonstrated excellent performance, but the values of each dimension are difficult to trace and interpret. Bag-of-words, as classic sparse interpretable embeddings, suffers from poor performance. Recently, Benara et al. (2024) propose interpretable text embeddings using large language models, which forms ”0/1” embeddings based on responses to a series of questions. These interpretable text embeddings are typically high-dimensional (larger than 10,000). In this work, we propose Low-dimensional (lower than 500) Dense and Interpretable text embeddings with Relative representations (LDIR). The numerical values of its dimensions indicate semantic relatedness to different anchor texts through farthest point sampling, offering both semantic representation as well as a certain level of traceability and interpretability. We validate LDIR on multiple semantic textual similarity, retrieval, and clustering tasks. Extensive experimental results show that LDIR performs close to the black-box baseline models and outperforms the interpretable embeddings baselines with much fewer dimensions.

Majority voting is considered an effective method to enhance chain-of-thought reasoning, as it selects the answer with the highest ”self-consistency” among different reasoning paths (Wang et al., 2023). However, previous chain-of-thought reasoning methods typically generate only a single answer in each trial, thereby ignoring the possibility of other potential answers. As a result, these alternative answers are often overlooked in subsequent voting processes. In this work, we propose to generate ranked answers in each reasoning process and conduct ranked voting among multiple ranked answers from different responses, thereby making the overall self-consistency more reliable. Specifically, we use three ranked voting methods: Instant-runoff voting, Borda count voting, and mean reciprocal rank voting. We validate our methods on six datasets, including three multiple-choice and three open-ended question-answering tasks, using both advanced open-source and closed-source large language models. Extensive experimental results indicate that our proposed method outperforms the baselines, showcasing the potential of leveraging the information of ranked answers and using ranked voting to improve reasoning performance. Code and logs will be released.

Multimodal Large Language Models (MLLMs) pose unique safety challenges due to their integration of visual and textual data, thereby introducing new dimensions of potential attacks and complex risk combinations. In this paper, we begin with a detailed analysis aimed at disentangling risks through step-by-step reasoning within multimodal inputs. We find that systematic multimodal risk disentanglement substantially enhances the risk awareness of MLLMs. Via leveraging the strong discriminative abilities of multimodal risk disentanglement, we further introduce DREAM ( Disentangling Risks to Enhance Safety Alignment in MLLMs), a novel approach that enhances safety alignment in MLLMs through supervised fine-tuning and iterative Reinforcement Learning from AI Feedback (RLAIF). Experimental results show that DREAM significantly boosts safety during both inference and training phases without compromising performance on normal tasks (namely oversafety), achieving a 16.17% improvement in the SIUO safe&effective score compared to GPT-4V.

Recently, there has been a trend of evaluating the Large Language Model (LLM) quality in the flavor of LLM-as-a-Judge, namely leveraging another LLM to evaluate the current output quality. However, existing judges are proven to be biased, namely they would favor answers which present better superficial quality (such as verbosity, fluency) while ignoring the instruction following ability. In this work, we propose systematic research about the bias of LLM-as-a-Judge. Specifically, for closed-source judge models, we apply calibration to mitigate the significance of superficial quality, both on probability level and prompt level. For open-source judge models, we propose to mitigate the bias by contrastive training, with curated negative samples that deviate from instruction but present better superficial quality. We apply our methods on the bias evaluation benchmark, and experiment results show our methods mitigate the bias by a large margin while maintaining a satisfactory evaluation accuracy.

The proliferation of open-source Large Language Models (LLMs) underscores the pressing need for evaluation methods. Existing works primarily rely on external evaluators, focusing on training and prompting strategies. However, a crucial aspect – model-aware glass-box features – is overlooked. In this study, we explore the utility of glass-box features under the scenario of self-evaluation, namely applying an LLM to evaluate its own output. We investigate various glass-box feature groups and discovered that the softmax distribution serves as a reliable quality indicator for self-evaluation. Experimental results on public benchmarks validate the feasibility of self-evaluation of LLMs using glass-box features.

This paper describes our submission for SemEval-2024 Task 6: SHROOM, a Shared-task on Hallucinations and Related Observable Overgeneration Mistakes. We propose four groups of methods for hallucination detection: 1) Entailment Recognition; 2) Similarity Search; 3) Factuality Verification; 4) Confidence Estimation. The four methods rely on either the semantic relationship between the hypothesis and its source (target) or on the model-aware features during decoding. We participated in both the model-agnostic and model-aware tracks. Our method’s effectiveness is validated by our high rankings 3rd in the model-agnostic track and 5th in the model-aware track. We have released our code on GitHub.

State-of-the-art translation Quality Estimation (QE) models are proven to be biased. More specifically, they over-rely on monolingual features while ignoring the bilingual semantic alignment. In this work, we propose a novel method to mitigate the bias of the QE model and improve estimation performance. Our method is based on the contrastive learning between clean and noisy sentence pairs. We first introduce noise to the target side of the parallel sentence pair, forming the negative samples. With the original parallel pairs as the positive sample, the QE model is contrastively trained to distinguish the positive samples from the negative ones. This objective is jointly trained with the regression-style quality estimation, so as to prevent the QE model from overfitting to monolingual features. Experiments on WMT QE evaluation datasets demonstrate that our method improves the estimation performance by a large margin while mitigating the bias.

Recently, Large Language Models (LLMs) have boosted the research in natural language processing and shown impressive capabilities across numerous domains, including machine translation evaluation. This paper presents our methods developed for the machine translation evaluation sub-task of the Eval4NLP 2023 Shared Task. Based on the provided LLMs, we propose a generation-based method as well as a probability-based method to perform evaluation, explore different strategies when selecting the demonstrations for in-context learning, and try different ensemble methods to further improve the evaluation accuracy. The experiment results on the development set and test set demonstrate the effectiveness of our proposed method.

Unsupervised domain adaptation of machine translation, which adapts a pre-trained translation model to a specific domain without in-domain parallel data, has drawn extensive attention in recent years. However, most existing methods focus on the fine-tuning based techniques, which is non-extensible. In this paper, we propose a new method to perform unsupervised domain adaptation in a non-parametric manner. Our method only resorts to in-domain monolingual data, and we jointly perform nearest neighbour inference on both forward and backward translation directions. The forward translation model creates nearest neighbour datastore for the backward direction, and vice versa, strengthening each other in an iterative style. Experiments on multi-domain datasets demonstrate that our method significantly improves the in-domain translation performance and achieves state-of-the-art results among non-parametric methods.

Back-translation has been proven to be effective in unsupervised domain adaptation of neural machine translation (NMT). However, the existing back-translation methods mainly improve domain adaptability by generating in-domain pseudo-parallel data that contains sentence-structural knowledge, paying less attention to the in-domain lexical knowledge, which may lead to poor translation of unseen in-domain words. In this paper, we propose an Iterative Constrained Back-Translation (ICBT) method to incorporate in-domain lexical knowledge on the basis of BT for unsupervised domain adaptation of NMT. Specifically, we apply lexical constraints into back-translation to generate pseudo-parallel data with in-domain lexical knowledge, and then perform round-trip iterations to incorporate more lexical knowledge. Based on this, we further explore sampling strategies of constrained words in ICBT to introduce more targeted lexical knowledge, via domain specificity and confidence estimation. Experimental results on four domains show that our approach achieves state-of-the-art results, improving the BLEU score by up to 3.08 compared to the strongest baseline, which demonstrates the effectiveness of our approach.

This paper presents the BJTU-Toshiba joint submission for WMT 2022 quality estimation shared task. We only participate in Task 1 (quality prediction) of the shared task, focusing on the sentence-level MQM prediction. The techniques we experimented with include the integration of monolingual language models and the pre-finetuning of pre-trained representations. We tried two styles of pre-finetuning, namely Translation Language Modeling and Replaced Token Detection. We demonstrate the competitiveness of our system compared to the widely adopted XLM-RoBERTa baseline. Our system is also the top-ranking system on the Sentence-level MQM Prediction for the English-German language pairs.

Translation suggestion (TS) models are used to automatically provide alternative suggestions for incorrect spans in sentences generated by machine translation. This paper introduces the system used in our submission to the WMT’22 Translation Suggestion shared task. Our system is based on the ensemble of different translation architectures, including Transformer, SA-Transformer, and DynamicConv. We use three strategies to construct synthetic data from parallel corpora to compensate for the lack of supervised data. In addition, we introduce a multi-phase pre-training strategy, adding an additional pre-training phase with in-domain data. We rank second and third on the English-German and English-Chinese bidirectional tasks, respectively.

Recent multilingual pre-trained models, like XLM-RoBERTa (XLM-R), have been demonstrated effective in many cross-lingual tasks. However, there are still gaps between the contextualized representations of similar words in different languages. To solve this problem, we propose a novel framework named Multi-View Mixed Language Training (MVMLT), which leverages code-switched data with multi-view learning to fine-tune XLM-R. MVMLT uses gradient-based saliency to extract keywords which are the most relevant to downstream tasks and replaces them with the corresponding words in the target language dynamically. Furthermore, MVMLT utilizes multi-view learning to encourage contextualized embeddings to align into a more refined language-invariant space. Extensive experiments with four languages show that our model achieves state-of-the-art results on zero-shot cross-lingual sentiment classification and dialogue state tracking tasks, demonstrating the effectiveness of our proposed model.