Computation and Language (cs.CL)

Abstract: Speech fluency/disfluency can be evaluated by analyzing a range of phonetic and prosodic features. Deep neural networks are commonly trained to map fluency-related features into the human scores. However, the effectiveness of deep learning-based models is constrained by the limited amount of labeled training samples. To address this, we introduce a self-supervised learning (SSL) approach that takes into account phonetic and prosody awareness for fluency scoring. Specifically, we first pre-train the model using a reconstruction loss function, by masking phones and their durations jointly on a large amount of unlabeled speech and text prompts. We then fine-tune the pre-trained model using human-annotated scoring data. Our experimental results, conducted on datasets such as Speechocean762 and our non-native datasets, show that our proposed method outperforms the baseline systems in terms of Pearson correlation coefficients (PCC). Moreover, we also conduct an ablation study to better understand the contribution of phonetic and prosody factors during the pre-training stage.

Abstract: Existing solutions to zero-shot text classification either conduct prompting with pre-trained language models, which is sensitive to the choices of templates, or rely on large-scale annotated data of relevant tasks for meta-tuning. In this work, we propose a new paradigm based on self-supervised learning to solve zero-shot text classification tasks by tuning the language models with unlabeled data, called self-supervised tuning. By exploring the inherent structure of free texts, we propose a new learning objective called first sentence prediction to bridge the gap between unlabeled data and text classification tasks. After tuning the model to learn to predict the first sentence in a paragraph based on the rest, the model is able to conduct zero-shot inference on unseen tasks such as topic classification and sentiment analysis. Experimental results show that our model outperforms the state-of-the-art baselines on 7 out of 10 tasks. Moreover, the analysis reveals that our model is less sensitive to the prompt design. Our code and pre-trained models are publicly available at https://github.com/DAMO-NLP-SG/SSTuning .

Abstract: We have constructed Arukikata Travelogue Dataset and released it free of charge for academic research. This dataset is a Japanese text dataset with a total of over 31 million words, comprising 4,672 Japanese domestic travelogues and 9,607 overseas travelogues. Before providing our dataset, there was a scarcity of widely available travelogue data for research purposes, and each researcher had to prepare their own data. This hinders the replication of existing studies and fair comparative analysis of experimental results. Our dataset enables any researchers to conduct investigation on the same data and to ensure transparency and reproducibility in research. In this paper, we describe the academic significance, characteristics, and prospects of our dataset.

Abstract: Existing research has shown that a multilingual pre-trained language model fine-tuned with one (source) language also performs well on downstream tasks for non-source languages, even though no fine-tuning is done on these languages. However, there is a clear gap between the performance of the source language and that of the non-source languages. This paper analyzes the fine-tuning process, discovers when the performance gap changes and identifies which network weights affect the overall performance most. Additionally, the paper seeks to answer to what extent the gap can be reduced by reducing forgetting. Based on the analysis results, a method named Fine-tuning slow and fast with four training policies is proposed to address these issues. Experimental results show the proposed method outperforms baselines by a clear margin.

Abstract: A centerpiece of the ever-popular reinforcement learning from human feedback (RLHF) approach to fine-tuning autoregressive language models is the explicit training of a reward model to emulate human feedback, distinct from the language model itself. This reward model is then coupled with policy-gradient methods to dramatically improve the alignment between language model outputs and desired responses. In this work, we adopt a novel perspective wherein a pre-trained language model is itself simultaneously a policy, reward function, and transition function. An immediate consequence of this is that reward learning and language model fine-tuning can be performed jointly and directly, without requiring any further downstream policy optimization. While this perspective does indeed break the traditional agent-environment interface, we nevertheless maintain that there can be enormous statistical benefits afforded by bringing to bear traditional algorithmic concepts from reinforcement learning. Our experiments demonstrate one concrete instance of this through efficient exploration based on the representation and resolution of epistemic uncertainty. In order to illustrate these ideas in a transparent manner, we restrict attention to a simple didactic data generating process and leave for future work extension to systems of practical scale.

Abstract: Finding an agreement among diverse opinions is a challenging topic in multiagent systems. Recently, large language models (LLMs) have shown great potential in addressing this challenge due to their remarkable capabilities in comprehending human opinions and generating human-like text. However, they typically rely on extensive human-annotated data. In this paper, we propose Self-Agreement, a novel framework for fine-tuning LLMs to autonomously find agreement using data generated by LLM itself. Specifically, our approach employs the generative pre-trained transformer-3 (GPT-3) to generate multiple opinions for each question in a question dataset and create several agreement candidates among these opinions. Then, a bidirectional encoder representations from transformers (BERT)-based model evaluates the agreement score of each agreement candidate and selects the one with the highest agreement score. This process yields a dataset of question-opinion-agreements, which we use to fine-tune a pre-trained LLM for discovering agreements among diverse opinions. Remarkably, a pre-trained LLM fine-tuned by our Self-Agreement framework achieves comparable performance to GPT-3 with only 1/25 of its parameters, showcasing its ability to identify agreement among various opinions without the need for human-annotated data.

Abstract: Breakthroughs in deep learning and memory networks have made major advances in natural language understanding. Language is sequential and information carried through the sequence can be captured through memory networks. Learning the sequence is one of the key aspects in learning the language. However, memory networks are not capable of holding infinitely long sequences in their memories and are limited by various constraints such as the vanishing or exploding gradient problem. Therefore, natural language understanding models are affected when presented with long sequential text. We introduce Long Term Memory network (LTM) to learn from infinitely long sequences. LTM gives priority to the current inputs to allow it to have a high impact. Language modeling is an important factor in natural language understanding. LTM was tested in language modeling, which requires long term memory. LTM is tested on Penn Tree bank dataset, Google Billion Word dataset and WikiText-2 dataset. We compare LTM with other language models which require long term memory.

Abstract: We propose and study Complementary Concept Generation (CCGen): given a concept of interest, e.g., "Digital Cameras", generating a list of complementary concepts, e.g., 1) Camera Lenses 2) Batteries 3) Camera Cases 4) Memory Cards 5) Battery Chargers. CCGen is beneficial for various applications like query suggestion and item recommendation, especially in the e-commerce domain. To solve CCGen, we propose to train language models to generate ranked lists of concepts with a two-step training strategy. We also teach the models to generate explanations by incorporating explanations distilled from large teacher models. Extensive experiments and analysis demonstrate that our model can generate high-quality concepts complementary to the input concept while producing explanations to justify the predictions.

Abstract: The personalized dialogue explores the consistent relationship between dialogue generation and personality. Existing personalized dialogue agents model persona profiles from three resources: sparse or dense persona descriptions and dialogue histories. However, sparse structured persona attributes are explicit but uninformative, dense persona texts contain rich persona descriptions with much noise, and dialogue history query is both noisy and uninformative for persona modeling. In this work, we combine the advantages of the three resources to obtain a richer and more accurate persona. We design a Contrastive Latent Variable-based model (CLV) that clusters the dense persona descriptions into sparse categories, which are combined with the history query to generate personalized responses. Experimental results on Chinese and English datasets demonstrate our model's superiority in personalization.

Abstract: Event Extraction (EE), aiming to identify and classify event triggers and arguments from event mentions, has benefited from pre-trained language models (PLMs). However, existing PLM-based methods ignore the information of trigger/argument fields, which is crucial for understanding event schemas. To this end, we propose a Probabilistic reCoupling model enhanced Event extraction framework (ProCE). Specifically, we first model the syntactic-related event fields as probabilistic biases, to clarify the event fields from ambiguous entanglement. Furthermore, considering multiple occurrences of the same triggers/arguments in EE, we explore probabilistic interaction strategies among multiple fields of the same triggers/arguments, to recouple the corresponding clarified distributions and capture more latent information fields. Experiments on EE datasets demonstrate the effectiveness and generalization of our proposed approach.

Abstract: Large language Models (LLMs) have achieved promising performance on arithmetic reasoning tasks by incorporating step-by-step chain-of-thought (CoT) prompting. However, LLMs face challenges in maintaining factual consistency during reasoning, exhibiting tendencies to condition overlooking, question misinterpretation, and condition hallucination over given problems. Existing methods use coarse-grained feedback (e.g., whether the answer is correct) to improve factual consistency. In this work, we propose RCoT (Reversing Chain-of-Thought), a novel method to improve LLMs' reasoning abilities by automatically detecting and rectifying factual inconsistency in LLMs' generated solutions. To detect factual inconsistency, RCoT first asks LLMs to reconstruct the problem based on generated solutions. Then fine-grained comparisons between the original problem and the reconstructed problem expose the factual inconsistency in the original solutions. To rectify the solution, RCoT formulates detected factual inconsistency into fine-grained feedback to guide LLMs in revising solutions. Experimental results demonstrate consistent improvements of RCoT over standard CoT across seven arithmetic datasets. Moreover, we find that manually written fine-grained feedback can dramatically improve LLMs' reasoning abilities (e.g., ChatGPT reaches 94.6% accuracy on GSM8K), encouraging the community to further explore the fine-grained feedback generation methods.

Abstract: Entity Alignment (EA) aims to find the equivalent entities between two Knowledge Graphs (KGs). Existing methods usually encode the triples of entities as embeddings and learn to align the embeddings, which prevents the direct interaction between the original information of the cross-KG entities. Moreover, they encode the relational triples and attribute triples of an entity in heterogeneous embedding spaces, which prevents them from helping each other. In this paper, we transform both triples into unified textual sequences, and model the EA task as a bi-directional textual entailment task between the sequences of cross-KG entities. Specifically, we feed the sequences of two entities simultaneously into a pre-trained language model (PLM) and propose two kinds of PLM-based entity aligners that model the entailment probability between sequences as the similarity between entities. Our approach captures the unified correlation pattern of two kinds of information between entities, and explicitly models the fine-grained interaction between original entity information. The experiments on five cross-lingual EA datasets show that our approach outperforms the state-of-the-art EA methods and enables the mutual enhancement of the heterogeneous information. Codes are available at https://github.com/OreOZhao/TEA.

Abstract: Automatic summarization plays an important role in the exponential document growth on the Web. On content websites such as CNN.com and WikiHow.com, there often exist various kinds of side information along with the main document for attention attraction and easier understanding, such as videos, images, and queries. Such information can be used for better summarization, as they often explicitly or implicitly mention the essence of the article. However, most of the existing side-aware summarization methods are designed to incorporate either single-modal or multi-modal side information, and cannot effectively adapt to each other. In this paper, we propose a general summarization framework, which can flexibly incorporate various modalities of side information. The main challenges in designing a flexible summarization model with side information include: (1) the side information can be in textual or visual format, and the model needs to align and unify it with the document into the same semantic space, (2) the side inputs can contain information from various aspects, and the model should recognize the aspects useful for summarization. To address these two challenges, we first propose a unified topic encoder, which jointly discovers latent topics from the document and various kinds of side information. The learned topics flexibly bridge and guide the information flow between multiple inputs in a graph encoder through a topic-aware interaction. We secondly propose a triplet contrastive learning mechanism to align the single-modal or multi-modal information into a unified semantic space, where the summary quality is enhanced by better understanding the document and side information. Results show that our model significantly surpasses strong baselines on three public single-modal or multi-modal benchmark summarization datasets.

Abstract: Medical dialogue systems aim to provide accurate answers to patients, necessitating specific domain knowledge. Recent advancements in Large Language Models (LLMs) have demonstrated their exceptional capabilities in the medical Q&A domain, indicating a rich understanding of common sense. However, LLMs are insufficient for direct diagnosis due to the absence of diagnostic strategies. The conventional approach to address this challenge involves expensive fine-tuning of LLMs. Alternatively, a more appealing solution is the development of a plugin that empowers LLMs to perform medical conversation tasks. Drawing inspiration from in-context learning, we propose PlugMed, a Plug-and-Play Medical Dialogue System that facilitates appropriate dialogue actions by LLMs through two modules: the prompt generation (PG) module and the response ranking (RR) module. The PG module is designed to capture dialogue information from both global and local perspectives. It selects suitable prompts by assessing their similarity to the entire dialogue history and recent utterances grouped by patient symptoms, respectively. Additionally, the RR module incorporates fine-tuned SLMs as response filters and selects appropriate responses generated by LLMs. Moreover, we devise a novel evaluation method based on intent and medical entities matching to assess the efficacy of dialogue strategies in medical conversations more effectively. Experimental evaluations conducted on three unlabeled medical dialogue datasets, including both automatic and manual assessments, demonstrate that our model surpasses the strong fine-tuning baselines.

Abstract: In this paper, we propose methods for discovering semantic differences in words appearing in two corpora based on the norms of contextualized word vectors. The key idea is that the coverage of meanings is reflected in the norm of its mean word vector. The proposed methods do not require the assumptions concerning words and corpora for comparison that the previous methods do. All they require are to compute the mean vector of contextualized word vectors and its norm for each word type. Nevertheless, they are (i) robust for the skew in corpus size; (ii) capable of detecting semantic differences in infrequent words; and (iii) effective in pinpointing word instances that have a meaning missing in one of the two corpora for comparison. We show these advantages for native and non-native English corpora and also for historical corpora.

Abstract: Diffusion models have emerged as the new state-of-the-art family of deep generative models, and their promising potentials for text generation have recently attracted increasing attention. Existing studies mostly adopt a single encoder architecture with partially noising processes for conditional text generation, but its degree of flexibility for conditional modeling is limited. In fact, the encoder-decoder architecture is naturally more flexible for its detachable encoder and decoder modules, which is extensible to multilingual and multimodal generation tasks for conditions and target texts. However, the encoding process of conditional texts lacks the understanding of target texts. To this end, a spiral interaction architecture for encoder-decoder text diffusion (DiffuSIA) is proposed. Concretely, the conditional information from encoder is designed to be captured by the diffusion decoder, while the target information from decoder is designed to be captured by the conditional encoder. These two types of information flow run through multilayer interaction spirally for deep fusion and understanding. DiffuSIA is evaluated on four text generation tasks, including paraphrase, text simplification, question generation, and open-domain dialogue generation. Experimental results show that DiffuSIA achieves competitive performance among previous methods on all four tasks, demonstrating the effectiveness and generalization ability of the proposed method.

Abstract: We introduce a new Information Extraction (IE) task dubbed Instruction-based IE, which aims to ask the system to follow specific instructions or guidelines to extract information. To facilitate research in this area, we construct a dataset called InstructIE, consisting of 270,000 weakly supervised data from Chinese Wikipedia and 1,000 high-quality crowdsourced annotated instances. We further evaluate the performance of various baseline models on the InstructIE dataset. The results reveal that although current models exhibit promising performance, there is still room for improvement. Furthermore, we conduct a comprehensive case study analysis, underlining the challenges inherent in the Instruction-based IE task. Code and dataset are available at https://github.com/zjunlp/DeepKE/tree/main/example/llm.

Abstract: This paper proposes a sequence-to-sequence learning approach for Arabic pronoun resolution, which explores the effectiveness of using advanced natural language processing (NLP) techniques, specifically Bi-LSTM and the BERT pre-trained Language Model, in solving the pronoun resolution problem in Arabic. The proposed approach is evaluated on the AnATAr dataset, and its performance is compared to several baseline models, including traditional machine learning models and handcrafted feature-based models. Our results demonstrate that the proposed model outperforms the baseline models, which include KNN, logistic regression, and SVM, across all metrics. In addition, we explore the effectiveness of various modifications to the model, including concatenating the anaphor text beside the paragraph text as input, adding a mask to focus on candidate scores, and filtering candidates based on gender and number agreement with the anaphor. Our results show that these modifications significantly improve the model's performance, achieving up to 81% on MRR and 71% for F1 score while also demonstrating higher precision, recall, and accuracy. These findings suggest that the proposed model is an effective approach to Arabic pronoun resolution and highlights the potential benefits of leveraging advanced NLP neural models.

Abstract: Disaster summarization approaches provide an overview of the important information posted during disaster events on social media platforms, such as, Twitter. However, the type of information posted significantly varies across disasters depending on several factors like the location, type, severity, etc. Verification of the effectiveness of disaster summarization approaches still suffer due to the lack of availability of good spectrum of datasets along with the ground-truth summary. Existing approaches for ground-truth summary generation (ground-truth for extractive summarization) relies on the wisdom and intuition of the annotators. Annotators are provided with a complete set of input tweets from which a subset of tweets is selected by the annotators for the summary. This process requires immense human effort and significant time. Additionally, this intuition-based selection of the tweets might lead to a high variance in summaries generated across annotators. Therefore, to handle these challenges, we propose a hybrid (semi-automated) approach (PORTRAIT) where we partly automate the ground-truth summary generation procedure. This approach reduces the effort and time of the annotators while ensuring the quality of the created ground-truth summary. We validate the effectiveness of PORTRAIT on 5 disaster events through quantitative and qualitative comparisons of ground-truth summaries generated by existing intuitive approaches, a semi-automated approach, and PORTRAIT. We prepare and release the ground-truth summaries for 5 disaster events which consist of both natural and man-made disaster events belonging to 4 different countries. Finally, we provide a study about the performance of various state-of-the-art summarization approaches on the ground-truth summaries generated by PORTRAIT using ROUGE-N F1-scores.

Abstract: Large Language Model (LLM) has gained popularity and achieved remarkable results in open-domain tasks, but its performance in real industrial domain-specific scenarios is average since there is no specific knowledge in it. This issue has attracted widespread attention, but there are few relevant benchmarks available. In this paper, we provide a benchmark Question Answering (QA) dataset named MSQA, which is about Microsoft products and IT technical problems encountered by customers. This dataset contains industry cloud-specific QA knowledge, which is not available for general LLM, so it is well suited for evaluating methods aimed at improving domain-specific capabilities of LLM. In addition, we propose a new model interaction paradigm that can empower LLM to achieve better performance on domain-specific tasks where it is not proficient. Extensive experiments demonstrate that the approach following our model fusion framework outperforms the commonly used LLM with retrieval methods.

Abstract: As the foundation of current natural language processing methods, pre-trained language model has achieved excellent performance. However, the black-box structure of the deep neural network in pre-trained language models seriously limits the interpretability of the language modeling process. After revisiting the coupled requirement of deep neural representation and semantics logic of language modeling, a Word-Context-Coupled Space (W2CSpace) is proposed by introducing the alignment processing between uninterpretable neural representation and interpretable statistical logic. Moreover, a clustering process is also designed to connect the word- and context-level semantics. Specifically, an associative knowledge network (AKN), considered interpretable statistical logic, is introduced in the alignment process for word-level semantics. Furthermore, the context-relative distance is employed as the semantic feature for the downstream classifier, which is greatly different from the current uninterpretable semantic representations of pre-trained models. Our experiments for performance evaluation and interpretable analysis are executed on several types of datasets, including SIGHAN, Weibo, and ChnSenti. Wherein a novel evaluation strategy for the interpretability of machine learning models is first proposed. According to the experimental results, our language model can achieve better performance and highly credible interpretable ability compared to related state-of-the-art methods.

Abstract: We argue that translation quality alone is not a sufficient metric for measuring knowledge transfer in multilingual neural machine translation. To support this claim, we introduce Representational Transfer Potential (RTP), which measures representational similarities between languages. We show that RTP can measure both positive and negative transfer (interference), and find that RTP is strongly correlated with changes in translation quality, indicating that transfer does occur. Furthermore, we investigate data and language characteristics that are relevant for transfer, and find that multi-parallel overlap is an important yet under-explored feature. Based on this, we develop a novel training scheme, which uses an auxiliary similarity loss that encourages representations to be more invariant across languages by taking advantage of multi-parallel data. We show that our method yields increased translation quality for low- and mid-resource languages across multiple data and model setups.

Abstract: The abstracts of scientific papers consist of premises and conclusions. Structured abstracts explicitly highlight the conclusion sentences, whereas non-structured abstracts may have conclusion sentences at uncertain positions. This implicit nature of conclusion positions makes the automatic segmentation of scientific abstracts into premises and conclusions a challenging task. In this work, we empirically explore using Normalized Mutual Information (NMI) for abstract segmentation. We consider each abstract as a recurrent cycle of sentences and place segmentation boundaries by greedily optimizing the NMI score between premises and conclusions. On non-structured abstracts, our proposed unsupervised approach GreedyCAS achieves the best performance across all evaluation metrics; on structured abstracts, GreedyCAS outperforms all baseline methods measured by $P_k$. The strong correlation of NMI to our evaluation metrics reveals the effectiveness of NMI for abstract segmentation.

Abstract: Augmenting large language models (LLMs) with external tools has emerged as a promising approach to solving complex problems. However, traditional methods, which finetune LLMs with tool demonstration data, can be both costly and restricted to a predefined set of tools. Recent in-context learning paradigm alleviates these issues, but the limited context length only allows for a few shots of demonstrations, leading to suboptimal understandings of the tools. Moreover, when there are numerous tools to choose from, in-context learning could completely fail to work. In this paper, we propose an alternative approach, $\textbf{ToolkenGPT}$, which combines the benefits of both sides. Our approach represents each $\underline{tool}$ as a to$\underline{ken}$ ($\textit{toolken}$) and learns an embedding for it, enabling tool calls in the same way as generating a regular word token. Once a toolken is triggered, the LLM is prompted to complete arguments for the tool to execute. ToolkenGPT offers the flexibility to plug in an arbitrary number of tools by expanding the set of toolkens on the fly. In addition, it improves tool use by allowing extensive demonstration data for learning the toolken embeddings. In diverse domains, including numerical reasoning, knowledge-based question answering, and embodied plan generation, our approach effectively augments LLMs with tools and substantially outperforms various latest baselines. ToolkenGPT demonstrates the promising ability to use relevant tools from a large tool set in complex scenarios.

Abstract: Pre-trained language models(PLM) have made impressive results in various NLP tasks. It has been revealed that one of the key factors to their success is the parameters of these models implicitly learn all kinds of knowledge during pre-training. However, encoding knowledge implicitly in the model parameters has two fundamental drawbacks. First, the knowledge is neither editable nor scalable once the model is trained, which is especially problematic in that knowledge is consistently evolving. Second, it lacks interpretability and prevents humans from understanding which knowledge PLM requires for a certain problem. In this paper, we introduce PlugLM, a pre-training model with differentiable plug-in memory(DPM). The key intuition is to decouple the knowledge storage from model parameters with an editable and scalable key-value memory and leverage knowledge in an explainable manner by knowledge retrieval in the DPM. To justify this design choice, we conduct evaluations in three settings including: (1) domain adaptation. PlugLM obtains 3.95 F1 improvements across four domains on average without any in-domain pre-training. (2) knowledge update. PlugLM could absorb new knowledge in a training-free way after pre-training is done. (3) in-task knowledge learning. PlugLM could be further improved by incorporating training samples into DPM with knowledge prompting.

Abstract: Recently, speech-text pre-training methods have shown remarkable success in many speech and natural language processing tasks. However, most previous pre-trained models are usually tailored for one or two specific tasks, but fail to conquer a wide range of speech-text tasks. In addition, existing speech-text pre-training methods fail to explore the contextual information within a dialogue to enrich utterance representations. In this paper, we propose Speech-text dialog Pre-training for spoken dialog understanding with ExpliCiT cRoss-Modal Alignment (SPECTRA), which is the first-ever speech-text dialog pre-training model. Concretely, to consider the temporality of speech modality, we design a novel temporal position prediction task to capture the speech-text alignment. This pre-training task aims to predict the start and end time of each textual word in the corresponding speech waveform. In addition, to learn the characteristics of spoken dialogs, we generalize a response selection task from textual dialog pre-training to speech-text dialog pre-training scenarios. Experimental results on four different downstream speech-text tasks demonstrate the superiority of SPECTRA in learning speech-text alignment and multi-turn dialog context.

Abstract: Online social media platforms, such as Twitter, are one of the most valuable sources of information during disaster events. Therefore, humanitarian organizations, government agencies, and volunteers rely on a summary of this information, i.e., tweets, for effective disaster management. Although there are several existing supervised and unsupervised approaches for automated tweet summary approaches, these approaches either require extensive labeled information or do not incorporate specific domain knowledge of disasters. Additionally, the most recent approaches to disaster summarization have proposed BERT-based models to enhance the summary quality. However, for further improved performance, we introduce the utilization of domain-specific knowledge without any human efforts to understand the importance (salience) of a tweet which further aids in summary creation and improves summary quality. In this paper, we propose a disaster-specific tweet summarization framework, IKDSumm, which initially identifies the crucial and important information from each tweet related to a disaster through key-phrases of that tweet. We identify these key-phrases by utilizing the domain knowledge (using existing ontology) of disasters without any human intervention. Further, we utilize these key-phrases to automatically generate a summary of the tweets. Therefore, given tweets related to a disaster, IKDSumm ensures fulfillment of the summarization key objectives, such as information coverage, relevance, and diversity in summary without any human intervention. We evaluate the performance of IKDSumm with 8 state-of-the-art techniques on 12 disaster datasets. The evaluation results show that IKDSumm outperforms existing techniques by approximately 2-79% in terms of ROUGE-N F1-score.

Abstract: Large language models (LLMs) have demonstrated impressive zero-shot or few-shot commonsense reasoning performance on various natural language processing (NLP) tasks. However, despite their strong commonsense reasoning abilities, LLMs still exhibit various kinds of inconsistency problems. While previous researches mainly focused on the self-consistency within a single LLM, we propose to explore the inter-consistency issue between two or more LLMs, which is critical for diverse and precise decision-making processes. Since the LLMs possess human-like intelligence after instruction tuning and reinforcement learning with human feedback (RLHF), we design a formal debate framework to delve into the inter-consistency problem among LLMs with three-stage debate: fair debate, mismatched debate, and roundtable debate. Through extensive experiments on 7 commonsense reasoning datasets, LLMs not only become more inter-consistent by compromising and refuting but also achieve higher performance and stronger interpretability. Furthermore, we find a much stronger LLM would be dominant in mismatched debates, while it will be easily misled by relatively weaker LLMs in a more complex debate scenario such as roundtable debate.

Abstract: Modern NLP models are often trained over large untrusted datasets, raising the potential for a malicious adversary to compromise model behaviour. For instance, backdoors can be implanted through crafting training instances with a specific textual trigger and a target label. This paper posits that backdoor poisoning attacks exhibit spurious correlation between simple text features and classification labels, and accordingly, proposes methods for mitigating spurious correlation as means of defence. Our empirical study reveals that the malicious triggers are highly correlated to their target labels; therefore such correlations are extremely distinguishable compared to those scores of benign features, and can be used to filter out potentially problematic instances. Compared with several existing defences, our defence method significantly reduces attack success rates across backdoor attacks, and in the case of insertion based attacks, our method provides a near-perfect defence.

Abstract: We propose a method for unsupervised opinion summarization that encodes sentences from customer reviews into a hierarchical discrete latent space, then identifies common opinions based on the frequency of their encodings. We are able to generate both abstractive summaries by decoding these frequent encodings, and extractive summaries by selecting the sentences assigned to the same frequent encodings. Our method is attributable, because the model identifies sentences used to generate the summary as part of the summarization process. It scales easily to many hundreds of input reviews, because aggregation is performed in the latent space rather than over long sequences of tokens. We also demonstrate that our appraoch enables a degree of control, generating aspect-specific summaries by restricting the model to parts of the encoding space that correspond to desired aspects (e.g., location or food). Automatic and human evaluation on two datasets from different domains demonstrates that our method generates summaries that are more informative than prior work and better grounded in the input reviews.

Abstract: Code search is an important task that has seen many developments in recent years. However, previous attempts have mostly considered the problem of searching for code by a text query. We argue that using a code snippet (and possibly an associated traceback) as a query and looking for answers with bugfixing instructions and code samples is a natural use case that is not covered by existing approaches. Moreover, existing datasets use comments extracted from code rather than full-text descriptions as text, making them unsuitable for this use case. We present a new SearchBySnippet dataset implementing the search-by-code use case based on StackOverflow data; it turns out that in this setting, existing architectures fall short of the simplest BM25 baseline even after fine-tuning. We present a new single encoder model SnippeR that outperforms several strong baselines on the SearchBySnippet dataset with a result of 0.451 Recall@10; we propose the SearchBySnippet dataset and SnippeR as a new important benchmark for code search evaluation.

Abstract: We consider the clone detection and information retrieval problems for source code, well-known tasks important for any programming language. Although it is also an important and interesting problem to find code snippets that operate identically but are written in different programming languages, to the best of our knowledge multilingual clone detection has not been studied in literature. In this work, we formulate the multilingual clone detection problem and present XCD, a new benchmark dataset produced from the CodeForces submissions dataset. Moreover, we present a novel training procedure, called cross-consistency training (CCT), that we apply to train language models on source code in different programming languages. The resulting CCT-LM model, initialized with GraphCodeBERT and fine-tuned with CCT, achieves new state of the art, outperforming existing approaches on the POJ-104 clone detection benchmark with 95.67\% MAP and AdvTest code search benchmark with 47.18\% MRR; it also shows the best results on the newly created multilingual clone detection benchmark XCD across all programming languages.

Abstract: Large language models (LLMs) have shown remarkable capabilities in language understanding and generation. However, such impressive capability typically comes with a substantial model size, which presents significant challenges in both the deployment, inference, and training stages. With LLM being a general-purpose task solver, we explore its compression in a task-agnostic manner, which aims to preserve the multi-task solving and language generation ability of the original LLM. One challenge to achieving this is the enormous size of the training corpus of LLM, which makes both data transfer and model post-training over-burdensome. Thus, we tackle the compression of LLMs within the bound of two constraints: being task-agnostic and minimizing the reliance on the original training dataset. Our method, named LLM-Pruner, adopts structural pruning that selectively removes non-critical coupled structures based on gradient information, maximally preserving the majority of the LLM's functionality. To this end, the performance of pruned models can be efficiently recovered through tuning techniques, LoRA, in merely 3 hours, requiring only 50K data. We validate the LLM-Pruner on three LLMs, including LLaMA, Vicuna, and ChatGLM, and demonstrate that the compressed models still exhibit satisfactory capabilities in zero-shot classification and generation. The code is available at: https://github.com/horseee/LLM-Pruner

Abstract: At the staggering pace with which the capabilities of large language models (LLMs) are increasing, creating future-proof evaluation sets to assess their understanding becomes more and more challenging. In this paper, we propose a novel paradigm for evaluating LLMs which leverages the idea that correct world understanding should be consistent across different (Fregean) senses of the same meaning. Accordingly, we measure understanding not in terms of correctness but by evaluating consistency across multiple senses that are generated by the model itself. We showcase our approach by instantiating a test where the different senses are different languages, hence using multilingual self-consistency as a litmus test for the model's understanding and simultaneously addressing the important topic of multilingualism. Taking one of the latest versions of ChatGPT as our object of study, we evaluate multilingual consistency for two different tasks across three different languages. We show that its multilingual consistency is still lacking, and that its task and world understanding are thus not language-independent. As our approach does not require any static evaluation corpora in languages other than English, it can easily and cheaply be extended to different languages and tasks and could become an integral part of future benchmarking efforts.

Abstract: Sentiment analysis (SA) systems are used in many products and hundreds of languages. Gender and racial biases are well-studied in English SA systems, but understudied in other languages, with few resources for such studies. To remedy this, we build a counterfactual evaluation corpus for gender and racial/migrant bias in four languages. We demonstrate its usefulness by answering a simple but important question that an engineer might need to answer when deploying a system: What biases do systems import from pre-trained models when compared to a baseline with no pre-training? Our evaluation corpus, by virtue of being counterfactual, not only reveals which models have less bias, but also pinpoints changes in model bias behaviour, which enables more targeted mitigation strategies. We release our code and evaluation corpora to facilitate future research.

Abstract: Formulating selective information needs results in queries that implicitly specify set operations, such as intersection, union, and difference. For instance, one might search for "shorebirds that are not sandpipers" or "science-fiction films shot in England". To study the ability of retrieval systems to meet such information needs, we construct QUEST, a dataset of 3357 natural language queries with implicit set operations, that map to a set of entities corresponding to Wikipedia documents. The dataset challenges models to match multiple constraints mentioned in queries with corresponding evidence in documents and correctly perform various set operations. The dataset is constructed semi-automatically using Wikipedia category names. Queries are automatically composed from individual categories, then paraphrased and further validated for naturalness and fluency by crowdworkers. Crowdworkers also assess the relevance of entities based on their documents and highlight attribution of query constraints to spans of document text. We analyze several modern retrieval systems, finding that they often struggle on such queries. Queries involving negation and conjunction are particularly challenging and systems are further challenged with combinations of these operations.

Abstract: In Natural Language Generation (NLG) tasks, for any input, multiple communicative goals are plausible, and any goal can be put into words, or produced, in multiple ways. We characterise the extent to which human production varies lexically, syntactically, and semantically across four NLG tasks, connecting human production variability to aleatoric or data uncertainty. We then inspect the space of output strings shaped by a generation system's predicted probability distribution and decoding algorithm to probe its uncertainty. For each test input, we measure the generator's calibration to human production variability. Following this instance-level approach, we analyse NLG models and decoding strategies, demonstrating that probing a generator with multiple samples and, when possible, multiple references, provides the level of detail necessary to gain understanding of a model's representation of uncertainty.

Abstract: Answering multi-hop questions over hybrid factual knowledge from the given text and table (TextTableQA) is a challenging task. Existing models mainly adopt a retriever-reader framework, which have several deficiencies, such as noisy labeling in training retriever, insufficient utilization of heterogeneous information over text and table, and deficient ability for different reasoning operations. In this paper, we propose a three-stage TextTableQA framework S3HQA, which comprises of retriever, selector, and reasoner. We use a retriever with refinement training to solve the noisy labeling problem. Then, a hybrid selector considers the linked relationships between heterogeneous data to select the most relevant factual knowledge. For the final stage, instead of adapting a reading comprehension module like in previous methods, we employ a generation-based reasoner to obtain answers. This includes two approaches: a row-wise generator and an LLM prompting generator~(first time used in this task). The experimental results demonstrate that our method achieves competitive results in the few-shot setting. When trained on the full dataset, our approach outperforms all baseline methods, ranking first on the HybridQA leaderboard.

Abstract: Digital technologies have led to an influx of text created daily in a variety of languages, styles, and formats. A great deal of the popularity of spell-checking systems can be attributed to this phenomenon since they are crucial to polishing the digitally conceived text. In this study, we tackle Typographical Error Type Detection in Persian, which has been relatively understudied. In this paper, we present a public dataset named FarsTypo, containing 3.4 million chronologically ordered and part-of-speech tagged words of diverse topics and linguistic styles. An algorithm for applying Persian-specific errors is developed and applied to a scalable size of these words, forming a parallel dataset of correct and incorrect words. Using FarsTypo, we establish a firm baseline and compare different methodologies using various architectures. In addition, we present a novel Many-to-Many Deep Sequential Neural Network to perform token classification using both word and character embeddings in combination with bidirectional LSTM layers to detect typographical errors across 51 classes. We compare our approach with highly-advanced industrial systems that, unlike this study, have been developed utilizing a variety of resources. The results of our final method were competitive in that we achieved an accuracy of 97.62%, a precision of 98.83%, a recall of 98.61%, and outperformed the rest in terms of speed.

Abstract: Recent developments in large language models (LLMs) have been impressive. However, these models sometimes show inconsistencies and problematic behavior, such as hallucinating facts, generating flawed code, or creating offensive and toxic content. Unlike these models, humans typically utilize external tools to cross-check and refine their initial content, like using a search engine for fact-checking, or a code interpreter for debugging. Inspired by this observation, we introduce a framework called CRITIC that allows LLMs, which are essentially "black boxes" to validate and progressively amend their own outputs in a manner similar to human interaction with tools. More specifically, starting with an initial output, CRITIC interacts with appropriate tools to evaluate certain aspects of the text, and then revises the output based on the feedback obtained during this validation process. Comprehensive evaluations involving free-form question answering, mathematical program synthesis, and toxicity reduction demonstrate that CRITIC consistently enhances the performance of LLMs. Meanwhile, our research highlights the crucial importance of external feedback in promoting the ongoing self-improvement of LLMs.

Abstract: Hallucinations in machine translation are translations that contain information completely unrelated to the input. Omissions are translations that do not include some of the input information. While both cases tend to be catastrophic errors undermining user trust, annotated data with these types of pathologies is extremely scarce and is limited to a few high-resource languages. In this work, we release an annotated dataset for the hallucination and omission phenomena covering 18 translation directions with varying resource levels and scripts. Our annotation covers different levels of partial and full hallucinations as well as omissions both at the sentence and at the word level. Additionally, we revisit previous methods for hallucination and omission detection, show that conclusions made based on a single language pair largely do not hold for a large-scale evaluation, and establish new solid baselines.

Abstract: Large language models (LLMs), such as ChatGPT, are prone to generate hallucinations, \ie content that conflicts with the source or cannot be verified by the factual knowledge. To understand what types of content and to which extent LLMs are apt to hallucinate, we introduce the Hallucination Evaluation for Large Language Models (HELMA) benchmark, a large collection of generated and human-annotated hallucinated samples for evaluating the performance of LLMs in recognizing and alleviating hallucination. To generate these samples, we propose a ChatGPT-based two-step framework, \ie sampling-then-filtering. Specifically, we first adopt two different sampling methods to generate hallucinated samples based on instructions, and then use an example-enhanced filtering method to select the best one. Furthermore, we also hire some human labelers to annotate the hallucinations in ChatGPT responses. The empirical results suggest that ChatGPT has some probabilities to generate hallucinations and existing LLMs face great challenges in recognizing the hallucinations in text. In addition, the performance can be improved by providing external knowledge or adding reasoning steps. Our benchmark can be accessed at https://github.com/RUCAIBox/HELMA.

Abstract: Large Language Models (LLMs) are known to memorize significant portions of their training data. Parts of this memorized content have been shown to be extractable by simply querying the model, which poses a privacy risk. We present a novel approach which uses prompt-tuning to control the extraction rates of memorized content in LLMs. We present two prompt training strategies to increase and decrease extraction rates, which correspond to an attack and a defense, respectively. We demonstrate the effectiveness of our techniques by using models from the GPT-Neo family on a public benchmark. For the 1.3B parameter GPT-Neo model, our attack yields a 9.3 percentage point increase in extraction rate compared to our baseline. Our defense can be tuned to achieve different privacy-utility trade-offs by a user-specified hyperparameter. We achieve an extraction rate reduction of up to 97.7% relative to our baseline, with a perplexity increase of 16.9%.

Abstract: Our proposed method, ReSeTOX (REdo SEarch if TOXic), addresses the issue of Neural Machine Translation (NMT) generating translation outputs that contain toxic words not present in the input. The objective is to mitigate the introduction of toxic language without the need for re-training. In the case of identified added toxicity during the inference process, ReSeTOX dynamically adjusts the key-value self-attention weights and re-evaluates the beam search hypotheses. Experimental results demonstrate that ReSeTOX achieves a remarkable 57% reduction in added toxicity while maintaining an average translation quality of 99.5% across 164 languages.

Abstract: The recent rapid progress in pre-training Large Language Models has relied on using self-supervised language modeling objectives like next token prediction or span corruption. On the other hand, Machine Translation Systems are mostly trained using cross-lingual supervision that requires aligned data between source and target languages. We demonstrate that pre-training Large Language Models on a mixture of a self-supervised Language Modeling objective and the supervised Machine Translation objective, therefore including cross-lingual parallel data during pre-training, yields models with better in-context learning abilities. As pre-training is a very resource-intensive process and a grid search on the best mixing ratio between the two objectives is prohibitively expensive, we propose a simple yet effective strategy to learn it during pre-training.

Abstract: The recognition of dataset names is a critical task for automatic information extraction in scientific literature, enabling researchers to understand and identify research opportunities. However, existing corpora for dataset mention detection are limited in size and naming diversity. In this paper, we introduce the Dataset Mentions Detection Dataset (DMDD), the largest publicly available corpus for this task. DMDD consists of the DMDD main corpus, comprising 31,219 scientific articles with over 449,000 dataset mentions weakly annotated in the format of in-text spans, and an evaluation set, which comprises of 450 scientific articles manually annotated for evaluation purposes. We use DMDD to establish baseline performance for dataset mention detection and linking. By analyzing the performance of various models on DMDD, we are able to identify open problems in dataset mention detection. We invite the community to use our dataset as a challenge to develop novel dataset mention detection models.

Abstract: Humans work together to solve common problems by having discussions, explaining, and agreeing or disagreeing with each other. Similarly, if a system can have discussions with humans when solving tasks, it can improve the system's performance and reliability. In previous research on explainability, it has only been possible for the system to make predictions and for humans to ask questions about them rather than having a mutual exchange of opinions. This research aims to create a dataset and computational framework for systems that discuss and refine their predictions through dialogue. Through experiments, we show that the proposed system can have beneficial discussions with humans improving the accuracy by up to 25 points in the natural language inference task.

Abstract: Prompting with natural language instructions has recently emerged as a popular method of harnessing the capabilities of large language models. Given the inherent ambiguity present in natural language, it is intuitive to consider the possible advantages of prompting with less ambiguous prompt styles, such as the use of pseudo-code. In this paper we explore if prompting via pseudo-code instructions helps improve the performance of pre-trained language models. We manually create a dataset of pseudo-code prompts for 132 different tasks spanning classification, QA and generative language tasks, sourced from the Super-NaturalInstructions dataset. Using these prompts along with their counterparts in natural language, we study their performance on two LLM families - BLOOM and CodeGen. Our experiments show that using pseudo-code instructions leads to better results, with an average increase (absolute) of 7-16 points in F1 scores for classification tasks and an improvement (relative) of 12-38% in aggregate ROUGE scores across all tasks. We include detailed ablation studies which indicate that code comments, docstrings, and the structural clues encoded in pseudo-code all contribute towards the improvement in performance. To the best of our knowledge our work is the first to demonstrate how pseudo-code prompts can be helpful in improving the performance of pre-trained LMs.

Abstract: Recently, data augmentation (DA) methods have been proven to be effective for pre-trained language models (PLMs) in low-resource settings, including few-shot named entity recognition (NER). However, conventional NER DA methods are mostly aimed at sequence labeling models, i.e., token-level classification, and few are compatible with unified autoregressive generation frameworks, which can handle a wider range of NER tasks, such as nested NER. Furthermore, these generation frameworks have a strong assumption that the entities will appear in the target sequence with the same left-to-right order as the source sequence. In this paper, we claim that there is no need to keep this strict order, and more diversified but reasonable target entity sequences can be provided during the training stage as a novel DA method. Nevertheless, a naive mixture of augmented data can confuse the model since one source sequence will then be paired with different target sequences. Therefore, we propose a simple but effective Prompt Ordering based Data Augmentation (PODA) method to improve the training of unified autoregressive generation frameworks under few-shot NER scenarios. Experimental results on three public NER datasets and further analyses demonstrate the effectiveness of our approach.

Abstract: The way and content in which users ask questions can provide insight into their current status, including their personality, emotions, and psychology. Instead of directly prompting the large language models (LLMs), we explore how chain-of-thought prompting helps in this scenario to perform reasoning and planning according to user status, aiming to provide a more personalized and engaging experience for the user query. To this end, we first construct a benchmark of 6 dialogue or question-answering datasets in both English and Chinese, covering 3 different aspects of user status (\textit{including} \textit{personality}, \textit{emotion}, and \textit{psychology}). Then we prompt the LLMs to generate the response regarding the user status as intermediate reasoning processing. We propose a novel demonstration selection strategy using the semantic similarity of intermediate reasoning instead of test queries. To evaluate the effectiveness and robustness of our approach, we conduct extensive experiments with 7 LLMs under zero-shot and one-shot settings. The experimental results show that our approach consistently outperforms standard prompting in terms of both \textit{helpfulness} and \textit{acceptness} across all datasets, regardless of the LLMs used. The code and dataset can be found at \url{https://github.com/ruleGreen/Dialogue\_CoT.git}.

Abstract: Neural metrics for machine translation evaluation, such as COMET, exhibit significant improvements in their correlation with human judgments, as compared to traditional metrics based on lexical overlap, such as BLEU. Yet, neural metrics are, to a great extent, "black boxes" returning a single sentence-level score without transparency about the decision-making process. In this work, we develop and compare several neural explainability methods and demonstrate their effectiveness for interpreting state-of-the-art fine-tuned neural metrics. Our study reveals that these metrics leverage token-level information that can be directly attributed to translation errors, as assessed through comparison of token-level neural saliency maps with Multidimensional Quality Metrics (MQM) annotations and with synthetically-generated critical translation errors. To ease future research, we release our code at: https://github.com/Unbabel/COMET/tree/explainable-metrics.

Abstract: Like many other machine learning applications, neural machine translation (NMT) benefits from over-parameterized deep neural models. However, these models have been observed to be brittle: NMT model predictions are sensitive to small input changes and can show significant variation across re-training or incremental model updates. This work studies a frequently used method in NMT, pseudo-label training (PLT), which is common to the related techniques of forward-translation (or self-training) and sequence-level knowledge distillation. While the effect of PLT on quality is well-documented, we highlight a lesser-known effect: PLT can enhance a model's stability to model updates and input perturbations, a set of properties we call model inertia. We study inertia effects under different training settings and we identify distribution simplification as a mechanism behind the observed results.

Abstract: Recent language models have made tremendous progress in the structured data to text generation task. However, these models still give sub-optimal performance where logical inference is required to generate the descriptions. In this work, we specifically focus on analytical text generation from structured data such as tables. Building on the taxonomy proposed in (Gupta et al., 2020) we focus on controllable table to text generation for the following reasoning categories: numerical reasoning, commonsense reasoning, temporal reasoning, table knowledge, and entity knowledge. We propose STOAT model, which is table and reasoning aware, with vector-quantization to infuse the given reasoning categories in the output. We observe that our model provides 10.19%, 1.13% improvement on the PARENT metric in iToTTo and Infotabs for the analytical sentence task. We also found that our model generates 15.3% more faithful and analytical descriptions as compared to the baseline models in human evaluation. We curate and release two reasoning category annotated table-to-interesting text generation datasets based on the ToTTo (Parikh et al., 2020) and InfoTabs datasets (Gupta et al.,2020).

Abstract: Medical systematic reviews are crucial for informing clinical decision making and healthcare policy. But producing such reviews is onerous and time-consuming. Thus, high-quality evidence synopses are not available for many questions and may be outdated even when they are available. Large language models (LLMs) are now capable of generating long-form texts, suggesting the tantalizing possibility of automatically generating literature reviews on demand. However, LLMs sometimes generate inaccurate (and potentially misleading) texts by hallucinating or omitting important information. In the healthcare context, this may render LLMs unusable at best and dangerous at worst. Most discussion surrounding the benefits and risks of LLMs have been divorced from specific applications. In this work, we seek to qualitatively characterize the potential utility and risks of LLMs for assisting in production of medical evidence reviews. We conducted 16 semi-structured interviews with international experts in systematic reviews, grounding discussion in the context of generating evidence reviews. Domain experts indicated that LLMs could aid writing reviews, as a tool for drafting or creating plain language summaries, generating templates or suggestions, distilling information, crosschecking, and synthesizing or interpreting text inputs. But they also identified issues with model outputs and expressed concerns about potential downstream harms of confidently composed but inaccurate LLM outputs which might mislead. Other anticipated potential downstream harms included lessened accountability and proliferation of automatically generated reviews that might be of low quality. Informed by this qualitative analysis, we identify criteria for rigorous evaluation of biomedical LLMs aligned with domain expert views.

Abstract: Stereotype benchmark datasets are crucial to detect and mitigate social stereotypes about groups of people in NLP models. However, existing datasets are limited in size and coverage, and are largely restricted to stereotypes prevalent in the Western society. This is especially problematic as language technologies gain hold across the globe. To address this gap, we present SeeGULL, a broad-coverage stereotype dataset, built by utilizing generative capabilities of large language models such as PaLM, and GPT-3, and leveraging a globally diverse rater pool to validate the prevalence of those stereotypes in society. SeeGULL is in English, and contains stereotypes about identity groups spanning 178 countries across 8 different geo-political regions across 6 continents, as well as state-level identities within the US and India. We also include fine-grained offensiveness scores for different stereotypes and demonstrate their global disparities. Furthermore, we include comparative annotations about the same groups by annotators living in the region vs. those that are based in North America, and demonstrate that within-region stereotypes about groups differ from those prevalent in North America. CONTENT WARNING: This paper contains stereotype examples that may be offensive.

Abstract: Reaction diagram parsing is the task of extracting reaction schemes from a diagram in the chemistry literature. The reaction diagrams can be arbitrarily complex, thus robustly parsing them into structured data is an open challenge. In this paper, we present RxnScribe, a machine learning model for parsing reaction diagrams of varying styles. We formulate this structured prediction task with a sequence generation approach, which condenses the traditional pipeline into an end-to-end model. We train RxnScribe on a dataset of 1,378 diagrams and evaluate it with cross validation, achieving an 80.0% soft match F1 score, with significant improvements over previous models. Our code and data are publicly available at https://github.com/thomas0809/RxnScribe.

Abstract: Large language models (LLMs) with in-context learning have demonstrated remarkable capability in the text-to-SQL task. Previous research has prompted LLMs with various demonstration-retrieval strategies and intermediate reasoning steps to enhance the performance of LLMs. However, those works often employ varied strategies when constructing the prompt text for text-to-SQL inputs, such as databases and demonstration examples. This leads to a lack of comparability in both the prompt constructions and their primary contributions. Furthermore, selecting an effective prompt construction has emerged as a persistent problem for future research. To address this limitation, we comprehensively investigate the impact of prompt constructions across various settings and provide insights for future work.

Abstract: Evidence retrieval is a core part of automatic fact-checking. Prior work makes simplifying assumptions in retrieval that depart from real-world use cases: either no access to evidence, access to evidence curated by a human fact-checker, or access to evidence available long after the claim has been made. In this work, we present the first fully automated pipeline to check real-world claims by retrieving raw evidence from the web. We restrict our retriever to only search documents available prior to the claim's making, modeling the realistic scenario where an emerging claim needs to be checked. Our pipeline includes five components: claim decomposition, raw document retrieval, fine-grained evidence retrieval, claim-focused summarization, and veracity judgment. We conduct experiments on complex political claims in the ClaimDecomp dataset and show that the aggregated evidence produced by our pipeline improves veracity judgments. Human evaluation finds the evidence summary produced by our system is reliable (it does not hallucinate information) and relevant to answering key questions about a claim, suggesting that it can assist fact-checkers even when it cannot surface a complete evidence set.

Abstract: A popular approach for improving the correctness of output from large language models (LLMs) is Self-Consistency - poll the LLM multiple times and output the most frequent solution. Existing Self-Consistency techniques always draw a constant number of samples per question, where a better approach will be to non-uniformly distribute the available budget based on the amount of agreement in the samples drawn so far. In response, we introduce Adaptive-Consistency, a cost-efficient, model-agnostic technique that dynamically adjusts the number of samples per question using a lightweight stopping criterion. Our experiments over 13 datasets and two LLMs demonstrate that Adaptive-Consistency reduces sample budget by up to 6.0 times with an average accuracy drop of less than 0.1%.

Abstract: Large Language Models (LLMs) have demonstrated remarkable performance in various tasks and gained significant attention. LLMs are also used for local sequence transduction tasks, including grammatical error correction (GEC) and formality style transfer, where most tokens in a source text are kept unchanged. However, it is inefficient to generate all target tokens because a prediction error of a target token may cause a catastrophe in predicting subsequent tokens and because the computational cost grows quadratically with the target sequence length. This paper proposes to predict a set of edit operations for the source text for local sequence transduction tasks. Representing an edit operation with a span of the source text and changed tokens, we can reduce the length of the target sequence and thus the computational cost for inference. We apply instruction tuning for LLMs on the supervision data of edit operations. Experiments show that the proposed method achieves comparable performance to the baseline in four tasks, paraphrasing, formality style transfer, GEC, and text simplification, despite reducing the length of the target text by as small as 21\%. Furthermore, we report that the instruction tuning with the proposed method achieved the state-of-the-art performance in the four tasks.

Abstract: Representations from transformer-based unidirectional language models are known to be effective at predicting brain responses to natural language. However, most studies comparing language models to brains have used GPT-2 or similarly sized language models. Here we tested whether larger open-source models such as those from the OPT and LLaMA families are better at predicting brain responses recorded using fMRI. Mirroring scaling results from other contexts, we found that brain prediction performance scales log-linearly with model size from 125M to 30B parameter models, with ~15% increased encoding performance as measured by correlation with a held-out test set across 3 subjects. Similar log-linear behavior was observed when scaling the size of the fMRI training set. We also characterized scaling for acoustic encoding models that use HuBERT, WavLM, and Whisper, and we found comparable improvements with model size. A noise ceiling analysis of these large, high-performance encoding models showed that performance is nearing the theoretical maximum for brain areas such as the precuneus and higher auditory cortex. These results suggest that increasing scale in both models and data will yield incredibly effective models of language processing in the brain, enabling better scientific understanding as well as applications such as decoding.