EMNLP 2023 Papers

Given a textual passage and an answer, humans are able to ask questions with various expressions, but this ability is still challenging for most question generation (QG) systems. Existing solutions mainly focus on the internal knowledge within the given passage or the semantic word space for diverse content planning. These methods, however, have not considered the potential of external knowledge for expression diversity. To bridge this gap, we propose RAST, a framework for Retrieval-Augmented Style Transfer, where the objective is to utilize the style of diverse templates for question generation. For training RAST, we develop a novel Reinforcement Learning (RL) based approach that maximizes a weighted combination of diversity reward and consistency reward. Here, the consistency reward is computed by a Question-Answering (QA) model, whereas the diversity reward measures how much the final output mimics the retrieved template. Experimental results show that our method outperforms previous diversity-driven baselines on diversity while being comparable in terms of consistency scores. Our code is available at \url{https://github.com/gouqi666/RAST}.

Generative AI models exhibit remarkable potential; however, hallucinations across various tasks present a significant challenge, particularly for longer inputs that current approaches struggle to address effectively. We introduce SCALE (Source Chunking Approach for Large-scale inconsistency Evaluation), a task-agnostic model for detecting factual inconsistencies using a novel chunking strategy. Specifically, SCALE is a Natural Language Inference (NLI) based model that uses large text chunks to condition over long texts. This approach achieves state-of-the-art performance in factual inconsistency detection for diverse tasks and long inputs. Additionally, we leverage the chunking mechanism and employ a novel algorithm to explain SCALE's decisions through relevant source sentence retrieval. Our evaluations reveal that SCALE outperforms existing methods on both standard benchmarks and a new long-form dialogue dataset ScreenEval we constructed. Moreover, SCALE surpasses competitive systems in efficiency and model explanation evaluations. We have released our code and data publicly to GitHub.

In recent years, few-shot relation classification has evoked many research interests. Yet a more challenging problem, i.e. none-of-the-above (NOTA), is under-explored. Existing works mainly regard NOTA as an extra class and treat it the same as known relations. However, such a solution ignores the overall instance distribution, where NOTA instances are actually outliers and distributed unnaturally compared with known ones. In this paper, we propose a density-aware prototypical network (D-Proto) to treat various instances distinctly. Specifically, we design unique training objectives to separate known instances and isolate NOTA instances, respectively. This produces an ideal instance distribution, where known instances are dense yet NOTAs have a small density. Moreover, we propose a NOTA detection module to further enlarge the density of known samples, and discriminate NOTA and known samples accurately. Experimental results demonstrate that the proposed method outperforms strong baselines with robustness towards various NOTA rates. The code will be made public after the paper is accepted.

One of the main methods for computational interpretation of a text is mapping it into a vector in some embedding space. Such vectors can then be used for a variety of textual processing tasks. Recently, most embedding spaces are a product of training large language models (LLMs). One major drawback of this type of representation is their incomprehensibility to humans. Understanding the embedding space is crucial for several important needs, including the need to debug the embedding method and compare it to alternatives, and the need to detect biases hidden in the model. In this paper, we present a novel method of understanding embeddings by transforming a latent embedding space into a comprehensible conceptual space. We present an algorithm for deriving a conceptual space with dynamic on-demand granularity. We devise a new evaluation method, using either human rater or LLM-based raters, to show that the conceptualized vectors indeed represent the semantics of the original latent ones. We show the use of our method for various tasks, including comparing the semantics of alternative models and tracing the layers of the LLM. The code is available online https://github.com/adiSimhi/Interpreting-Embedding-Spaces-by-Conceptualization.

The classical deep clustering optimization methods basically leverage information such as clustering centers, mutual information, and distance metrics to construct implicit generalized labels to establish information feedback (weak supervision) and thus optimize the deep model. However, the resulting generalized labels have different degrees of errors in the whole clustering process due to the limitation of clustering accuracy, which greatly interferes with the clustering process. To this end, this paper proposes a general deep clustering optimization method from the perspective of empirical risk minimization, using the correlation relationship between the samples. Experiments on two classical deep clustering methods demonstrate the necessity and effectiveness of the method. Code is available at https://github.com/yangzonghao1024/DCGLU.

Recent Language Models (LMs) have shown impressive capabilities in generating texts with the knowledge internalized in parameters. Yet, LMs often generate the factually incorrect responses to the given queries, since their knowledge may be inaccurate, incomplete, and outdated. To address this problem, previous works propose to augment LMs with the knowledge retrieved from an external knowledge source. However, such approaches often show suboptimal text generation performance due to two reasons: 1) the model may fail to retrieve the knowledge relevant to the given query, or 2) the model may not faithfully reflect the retrieved knowledge in the generated text. To overcome these, we propose to verify the output and the knowledge of the knowledge-augmented LMs with a separate verifier, which is a small LM that is trained to detect those two types of errors through instruction-finetuning. Then, when the verifier recognizes an error, we can rectify it by either retrieving new knowledge or generating new text. Further, we use an ensemble of the outputs from different instructions with a single verifier to enhance the reliability of the verification processes. We validate the effectiveness of the proposed verification steps on multiple question answering benchmarks, whose results show that the proposed verifier effectively identifies retrieval and generation errors, allowing LMs to provide more factually correct outputs. Our code is available at https://github.com/JinheonBaek/KALMV.

Math word problems (MWP) involving advanced operators such as linear equation solver cannot be easily tackled by earlier MWP methods, because the existing generation methods suffer from repeated sub-expression generation and deductive methods are restricted to dealing with binary operations. This paper propose a new multivariate directed acyclic graph (mDAG) as an alternative to the generation methods' binary expression tree or the deductive methods' binary directed acyclic graph. Then to produce the topological ordering of mDAG, we propose a generation-based deductive (GeDe) model, which equips a generation model with a re-encoder to keep the deductive property but avoid the expensive enumeration of the deductive methods. GeDe performs well on math problems with many operators on the widely used benchmarks as well as solving multivariate operators on our own CMWPA benchmark. Our code is available at https://github.com/hyx1999/GeDe

Large Language Models (LLMs) have showcased impressive performance. However, due to their inability to capture relationships among samples, these frozen LLMs inevitably keep repeating similar mistakes. In this work, we propose our Tuning-free Rule Accumulation (TRAN) framework, which guides LLMs in improving their performance by learning from previous mistakes. Considering data arrives sequentially, LLMs gradually accumulate rules from incorrect cases, forming a rule collection. These rules are then utilized by the LLMs to avoid making similar mistakes when processing subsequent inputs. Moreover, the rules remain independent of the primary prompts, seamlessly complementing prompt design strategies. Experimentally, we show that TRAN improves over recent baselines by a large margin.

Maintaining a consistent persona is a key quality for any open domain dialogue system. Current state-of-the-art systems do this by training agents with supervised learning or online reinforcement learning (RL). However, systems trained with supervised learning often lack consistency as they are never punished for uttering contradictions. Additional training with RL can alleviate some of these issues, however the training process is expensive. Instead, we propose an offline RL framework to improve the persona consistency of dialogue systems. Our framework allows us to combine the advantages of previous methods as we can inexpensively train our model on existing data as in supervised learning, while punishing and rewarding specific utterances as in RL. We also introduce a simple importance sampling method to reduce the variance of importance weights in offline RL training which we call Variance-Reducing MLE-Initialized (VaRMI) importance sampling. Our automatic and human evaluations show that our framework improves both the persona consistency and dialogue quality of a state-of-the-art social chatbot.

Retrieval-augmented language models show promise in addressing issues like outdated information and hallucinations in language models (LMs). However, current research faces two main problems: 1) determining what information to retrieve, and 2) effectively combining retrieved information during generation. We argue that valuable retrieved information should not only be related to the current source text but also consider the future target text, given the nature of LMs that model future tokens. Moreover, we propose that aggregation using latent variables derived from a compact latent space is more efficient than utilizing explicit raw text, which is limited by context length and susceptible to noise. Therefore, we introduce RegaVAE, a retrieval-augmented language model built upon the variational auto-encoder (VAE). It encodes the text corpus into a latent space, capturing current and future information from both source and target text. Additionally, we leverage the VAE to initialize the latent space and adopt the probabilistic form of the retrieval generation paradigm by expanding the Gaussian prior distribution into a Gaussian mixture distribution. Theoretical analysis provides an optimizable upper bound for RegaVAE. Experimental results on various datasets demonstrate significant improvements in text generation quality and hallucination removal.

In this paper we improve the zero-shot generalization ability of language models via Mixture-Of-Memory Augmentation (MoMA), a mechanism that retrieves augmentation documents from multiple information corpora (external memories), with the option to ''plug in'' unseen memory at inference time. We develop a joint learning mechanism that trains the augmentation component with latent labels derived from the end retrieval task, paired with hard negatives from the memory mixture. We instantiate the model in a zero-shot dense retrieval setting by augmenting strong T5-based retrievers with MoMA. With only T5-base, our model obtains strong zero-shot retrieval accuracy on the eighteen tasks included in the standard BEIR benchmark, outperforming some systems with larger model sizes. As a plug-in-play model, our model can efficiently generalize to any unseen corpus, meanwhile achieving comparable or even better performance than methods relying on target-specific pretraining. Our analysis further illustrates the necessity of augmenting with mixture-of-memory for robust generalization, the benefits of augmentation learning, and how MoMA utilizes the plug-in memory at inference time without changing its parameters. Our code can be found at https://github.com/gesy17/MoMA.

Instruction tuning (IT) achieves impressive zero-shot generalization results by training large language models (LLMs) on a massive amount of diverse tasks with instructions. However, how to select new tasks to improve the performance and generalizability of IT models remains an open question. Training on all existing tasks is impractical due to prohibiting computation requirements, and randomly selecting tasks can lead to suboptimal performance. In this work, we propose active instruction tuning based on prompt uncertainty, a novel framework to identify informative tasks, and then actively tune the models on the selected tasks. We represent the informativeness of new tasks with the disagreement of the current model outputs over perturbed prompts. Our experiments on NIV2 and Self-Instruct datasets demonstrate that our method consistently outperforms other baseline strategies for task selection, achieving better out-of-distribution generalization with fewer training tasks. Additionally, we introduce a task map that categorizes and diagnoses tasks based on prompt uncertainty and prediction probability. We discover that training on ambiguous (prompt-uncertain) tasks improves generalization while training on difficult (prompt-certain and low-probability) tasks offers no benefit, underscoring the importance of task selection for instruction tuning.

Retrieval-Augmented Machine Translation (RAMT) is attracting growing attention. This is because RAMT not only improves translation metrics, but is also assumed to implement some form of domain adaptation. In this contribution, we study another salient trait of RAMT, its ability to make translation decisions more transparent by allowing users to go back to examples that contributed to these decisions. For this, we propose a novel architecture aiming to increase this transparency. This model adapts a retrieval-augmented version of the Levenshtein Transformer and makes it amenable to simultaneously edit multiple fuzzy matches found in memory. We discuss how to perform training and inference in this model, based on multi-way alignment algorithms and imitation learning. Our experiments show that editing several examples positively impacts translation scores, notably increasing the number of target spans that are copied from existing instances.

Large Language Models (LLMs) have attained the impressive capability to resolve a wide range of NLP tasks by fine-tuning high-quality instruction data. However, collecting human-written data of high quality, especially multi-turn dialogues, is expensive and unattainable for most people. Though previous studies have used powerful LLMs to generate the dialogues automatically, they all suffer from generating untruthful dialogues because of the model hallucination. Therefore, we propose a method called RefGPT to generate enormous truthful and customized dialogues without worrying about factual errors caused by the model hallucination. RefGPT solves the model hallucination in dialogue generation by restricting the LLMs to leverage the given reference instead of reciting their own knowledge to generate dialogues. Additionally, RefGPT adds detailed controls on every utterance to enable high customization capability, which previous studies have ignored. On the basis of RefGPT, we also propose two high-quality dialogue datasets generated by GPT-4, namely **RefGPT-Fact** and **RefGPT-Code**. RefGPT-Fact is a dataset with 100k multi-turn dialogues based on factual knowledge and RefGPT-Code has 76k multi-turn dialogues covering a wide range of coding scenarios. Our code and datasets are released in https://github.com/mutonix/RefGPT.

Even the most advanced language models remain susceptible to errors necessitating to modify these models without initiating a comprehensive retraining process. Model editing refers to the modification of a model's knowledge or representations in a manner that produces the desired outcomes. Prior research primarily centered around editing factual data e.g. "Messi plays for Inter Miami" confining the definition of an edit to a knowledge triplet i.e. (subject, object, relation). However, as the applications of language models expand, so do the diverse ways in which we wish to edit and refine their outputs. In this study, we broaden the scope of the editing problem to include an array of editing cases such as debiasing and rectifying reasoning errors and define an edit as any natural language expression that solicits a change in the model's outputs. We are introducing DUnE, an editing benchmark where edits are natural language sentences and propose that DUnE presents a challenging yet relevant task. To substantiate this claim, we conduct an extensive series of experiments testing various editing approaches to address DUnE, demonstrating their respective strengths and weaknesses. We argue that retrieval-augmented language modeling can outperform specialized editing techniques and neither set of approaches has fully solved the generalized editing problem covered by our benchmark.

In this paper, we propose a novel negotiation agent designed for the online marketplace. Our dialogue agent is integrative in nature i.e, it possesses the capability to negotiate on price as well as other factors, such as the addition or removal of items from a deal bundle, thereby offering a more flexible and comprehensive negotiation experience. To enable this functionality, we create a new dataset called Integrative Negotiation Dataset (IND). For this dataset creation, we introduce a new semi-automated data creation method, which combines defining negotiation intents, actions, and intent-action simulation between users and the agent to generate potential dialogue flows. Finally, the prompting of GPT-J, a state-of-the-art language model, is done to generate dialogues for a given intent, with a human-in-the-loop process for post-editing and refining minor errors to ensure high data quality. We first train a maximum likelihood loss based model on IND, and then employ a set of novel rewards specifically tailored for the negotiation task to train our Integrative Negotiation Agent (INA). These rewards incentivize the agent to learn effective negotiation strategies that can adapt to various contextual requirements and price proposals. We train our model and conduct experiments to evaluate the effectiveness of our reward-based dialogue agent for negotiation. Our results demonstrate that the proposed approach and reward functions significantly enhance the negotiation capabilities of the dialogue agent. The INA successfully engages in integrative negotiations, displaying the ability to dynamically adjust prices and negotiate the inclusion or exclusion of items in a deal bundle.

Language serves as a powerful tool for the manifestation of societal belief systems. In doing so, it also perpetuates the prevalent biases in our society. Gender bias is one of the most pervasive biases in our society and is seen in online and offline discourses. With LLMs increasingly gaining human-like fluency in text generation, gaining a nuanced understanding of the biases these systems can generate is imperative. Prior work often treats gender bias as a binary classification task. However, acknowledging that bias must be perceived at a relative scale; we investigate the generation and consequent receptivity of manual annotators to bias of varying degrees. Specifically, we create the first dataset of GPT-generated English text with normative ratings of gender bias. Ratings were obtained using Best--Worst Scaling -- an efficient comparative annotation framework. Next, we systematically analyze the variation of themes of gender biases in the observed ranking and show that identity-attack is most closely related to gender bias. Finally, we show the performance of existing automated models trained on related concepts on our dataset.

Inverted file structure is a common technique for accelerating dense retrieval. It clusters documents based on their embeddings; during searching, it probes nearby clusters w.r.t. an input query and only evaluates documents within them by subsequent codecs, thus avoiding the expensive cost from exhaustive traversal. However, the clustering is always lossy, which results in the miss of relevant documents in the probed clusters and hence degrades retrieval quality. In contrast, lexical matching, such as overlaps of salient terms, tend to be strong features for identifying relevant documents. In this work, we present the Hybrid Inverted Index (HI$^2$), where the embedding clusters and salient terms work collaboratively to accelerate dense retrieval. To make best of both effectiveness and efficiency, we devise a cluster selector and a term selector, to construct compact inverted lists and efficiently searching through them. Moreover, we leverage simple unsupervised algorithms as well as end-to-end knowledge distillation to learn these two modules, with the latter further boosting the effectiveness. Based on comprehensive experiments on popular retrieval benchmarks, we verify that clusters and terms indeed complement each other, enabling HI$^2$ to achieve lossless retrieval quality with competitive efficiency across a variety of index settings.

Recently, with the chain of thought (CoT) prompting, large language models (LLMs), e.g., GPT-3, have shown strong reasoning ability in several natural language processing tasks such as arithmetic, commonsense, and logical reasoning. However, LLMs with CoT require multi-step prompting and multi-token prediction, which is highly sensitive to individual mistakes and vulnerable to error accumulation. The above issues make the LLMs need the ability to verify the answers. In fact, after inferring conclusions in some thinking decision tasks, people often check them by re-verifying steps to avoid some mistakes. In this paper, we propose and prove that LLMs also have similar self-verification abilities. We take the conclusion obtained by CoT as one of the conditions for solving the original problem. By performing a backward verification of the answers that LLM deduced for itself, we can obtain interpretable answer validation scores to select the candidate answer with the highest score. Experimental results demonstrate that the proposed method can improve the reasoning performance on various arithmetic, commonsense, and logical reasoning datasets. Our code is publicly available at: https://github.com/WENGSYX/Self-Verification.