GPT-4 vs. Human Translators: A Comprehensive Evaluation of Translation Quality Across Languages, Domains, and Expertise Levels

This study comprehensively evaluates the translation quality of Large Language Models (LLMs), specifically GPT-4, against human translators of varying expertise levels across multiple language pairs and domains. Through carefully designed annotation rounds, we find that GPT-4 performs comparably to junior translators in terms of total errors made but lags behind medium and senior translators. We also observe the imbalanced performance across different languages and domains, with GPT-4's translation capability gradually weakening from resource-rich to resource-poor directions. In addition, we qualitatively study the translation given by GPT-4 and human translators, and find that GPT-4 translator suffers from literal translations, but human translators sometimes overthink the background information. To our knowledge, this study is the first to evaluate LLMs against human translators and analyze the systematic differences between their outputs, providing valuable insights into the current state of LLM-based translation and its potential limitations.

What Have We Achieved on Non-autoregressive Translation?

Recent advances have made non-autoregressive (NAT) translation comparable to autoregressive methods (AT). However, their evaluation using BLEU has been shown to weakly correlate with human annotations. Limited research compares non-autoregressive translation and autoregressive translation comprehensively, leaving uncertainty about the true proximity of NAT to AT. To address this gap, we systematically evaluate four representative NAT methods across various dimensions, including human evaluation. Our empirical results demonstrate that despite narrowing the performance gap, state-of-the-art NAT still underperforms AT under more reliable evaluation metrics. Furthermore, we discover that explicitly modeling dependencies is crucial for generating natural language and generalizing to out-of-distribution sequences.

RefuteBench: Evaluating Refuting Instruction-Following for Large Language Models

The application scope of large language models (LLMs) is increasingly expanding. In practical use, users might provide feedback based on the model's output, hoping for a responsive model that can complete responses according to their feedback. Whether the model can appropriately respond to users' refuting feedback and consistently follow through with execution has not been thoroughly analyzed. In light of this, this paper proposes a comprehensive benchmark, RefuteBench, covering tasks such as question answering, machine translation, and email writing. The evaluation aims to assess whether models can positively accept feedback in form of refuting instructions and whether they can consistently adhere to user demands throughout the conversation. We conduct evaluations on numerous LLMs and find that LLMs are stubborn, i.e. exhibit inclination to their internal knowledge, often failing to comply with user feedback. Additionally, as the length of the conversation increases, models gradually forget the user's stated feedback and roll back to their own responses. We further propose a recall-and-repeat prompts as a simple and effective way to enhance the model's responsiveness to feedback.

Potential and Challenges of Model Editing for Social Debiasing

Large language models (LLMs) trained on vast corpora suffer from inevitable stereotype biases. Mitigating these biases with fine-tuning could be both costly and data-hungry. Model editing methods, which focus on modifying LLMs in a post-hoc manner, are of great potential to address debiasing. However, it lacks a comprehensive study that facilitates both internal and external model editing methods, supports various bias types, as well as understands the pros and cons of applying editing methods to stereotypical debiasing. To mitigate this gap, we carefully formulate social debiasing into an editing problem and benchmark seven existing model editing algorithms on stereotypical debiasing, i.e., debias editing. Our findings in three scenarios reveal both the potential and challenges of debias editing: (1) Existing model editing methods can effectively preserve knowledge and mitigate biases, while the generalization of debias effect from edited sentences to semantically equivalent sentences is limited.(2) Sequential editing highlights the robustness of SERAC (Mitchell et al. 2022b), while internal editing methods degenerate with the number of edits. (3) Model editing algorithms achieve generalization towards unseen biases both within the same type and from different types. In light of these findings, we further propose two simple but effective methods to improve debias editing, and experimentally show the effectiveness of the proposed methods.

Dynamics of Instruction Tuning: Each Ability of Large Language Models Has Its Own Growth Pace

Instruction tuning is a burgeoning method to elicit the general intelligence of Large Language Models (LLMs). However, the creation of instruction data is still largely heuristic, leading to significant variation in quality and distribution across existing datasets. Experimental conclusions drawn from these datasets are also inconsistent, with some studies emphasizing the importance of scaling instruction numbers, while others argue that a limited number of samples suffice. To better understand data construction guidelines, we deepen our focus from the overall model performance to the growth of each underlying ability, such as creative writing, code generation, and logical reasoning. We systematically investigate the effects of data volume, parameter size, and data construction methods on the development of various abilities, using hundreds of model checkpoints (7b to 33b) fully instruction-tuned on a new collection of over 40k human-curated instruction data. This proposed dataset is stringently quality-controlled and categorized into ten distinct LLM abilities. Our study reveals three primary findings: (i) Despite data volume and parameter scale directly impacting models' overall performance, some abilities are more responsive to their increases and can be effectively trained using limited data, while some are highly resistant to these changes. (ii) Human-curated data strongly outperforms synthetic data from GPT-4 in efficiency and can constantly enhance model performance with volume increases, but is unachievable with synthetic data. (iii) Instruction data brings powerful cross-ability generalization, with evaluation results on out-of-domain data mirroring the first two observations. Furthermore, we demonstrate how these findings can guide more efficient data constructions, leading to practical performance improvements on public benchmarks.

Understanding In-Context Learning from Repetitions

This paper explores the elusive mechanism underpinning in-context learning in Large Language Models (LLMs). Our work provides a novel perspective by examining in-context learning via the lens of surface repetitions. We quantitatively investigate the role of surface features in text generation, and empirically establish the existence of \emph{token co-occurrence reinforcement}, a principle that strengthens the relationship between two tokens based on their contextual co-occurrences. By investigating the dual impacts of these features, our research illuminates the internal workings of in-context learning and expounds on the reasons for its failures. This paper provides an essential contribution to the understanding of in-context learning and its potential limitations, providing a fresh perspective on this exciting capability.

Revisiting Cross-Lingual Summarization: A Corpus-based Study and A New Benchmark with Improved Annotation

Most existing cross-lingual summarization (CLS) work constructs CLS corpora by simply and directly translating pre-annotated summaries from one language to another, which can contain errors from both summarization and translation processes. To address this issue, we propose ConvSumX, a cross-lingual conversation summarization benchmark, through a new annotation schema that explicitly considers source input context. ConvSumX consists of 2 sub-tasks under different real-world scenarios, with each covering 3 language directions. We conduct thorough analysis on ConvSumX and 3 widely-used manually annotated CLS corpora and empirically find that ConvSumX is more faithful towards input text. Additionally, based on the same intuition, we propose a 2-Step method, which takes both conversation and summary as input to simulate human annotation process. Experimental results show that 2-Step method surpasses strong baselines on ConvSumX under both automatic and human evaluation. Analysis shows that both source input text and summary are crucial for modeling cross-lingual summaries.

Explicit Syntactic Guidance for Neural Text Generation

Most existing text generation models follow the sequence-to-sequence paradigm. Generative Grammar suggests that humans generate natural language texts by learning language grammar. We propose a syntax-guided generation schema, which generates the sequence guided by a constituency parse tree in a top-down direction. The decoding process can be decomposed into two parts: (1) predicting the infilling texts for each constituent in the lexicalized syntax context given the source sentence; (2) mapping and expanding each constituent to construct the next-level syntax context. Accordingly, we propose a structural beam search method to find possible syntax structures hierarchically. Experiments on paraphrase generation and machine translation show that the proposed method outperforms autoregressive baselines, while also demonstrating effectiveness in terms of interpretability, controllability, and diversity.

DC-MBR: Distributional Cooling for Minimum Bayesian Risk Decoding

Minimum Bayesian Risk Decoding (MBR) emerges as a promising decoding algorithm in Neural Machine Translation. However, MBR performs poorly with label smoothing, which is surprising as label smoothing provides decent improvement with beam search and improves generality in various tasks. In this work, we show that the issue arises from the un-consistency of label smoothing on the token-level and sequence-level distributions. We demonstrate that even though label smoothing only causes a slight change in the token-level, the sequence-level distribution is highly skewed. We coin the issue \emph{distributional over-smoothness}. To address this issue, we propose a simple and effective method, Distributional Cooling MBR (DC-MBR), which manipulates the entropy of output distributions by tuning down the Softmax temperature. We theoretically prove the equivalence between pre-tuning label smoothing factor and distributional cooling. Experiments on NMT benchmarks validate that distributional cooling improves MBR's efficiency and effectiveness in various settings.

Probing Causes of Hallucinations in Neural Machine Translations

Hallucination, one kind of pathological translations that bothers Neural Machine Translation, has recently drawn much attention. In simple terms, hallucinated translations are fluent sentences but barely related to source inputs. Arguably, it remains an open problem how hallucination occurs. In this paper, we propose to use probing methods to investigate the causes of hallucinations from the perspective of model architecture, aiming to avoid such problems in future architecture designs. By conducting experiments over various NMT datasets, we find that hallucination is often accompanied by the deficient encoder, especially embeddings, and vulnerable cross-attentions, while, interestingly, cross-attention mitigates some errors caused by the encoder.