Authors: Bin Gao, Zhuomin He, Puru Sharma, Qingxuan Kang, Djordje Jevdjic, Junbo Deng, Xingkun Yang, Zhou Yu, Pengfei Zuo
Abstract: Interacting with humans through multi-turn conversations is a fundamental feature of large language models (LLMs). However, existing LLM serving engines for executing multi-turn conversations are inefficient due to the need to repeatedly compute the key-value (KV) caches of historical tokens, incurring high serving costs. To address the problem, this paper proposes AttentionStore, a new attention mechanism that enables the reuse of KV caches (i.e., attention reuse) across multi-turn conversations, significantly reducing the repetitive computation overheads. AttentionStore maintains a hierarchical KV caching system that leverages cost-effective memory/storage mediums to save KV caches for all requests. To reduce KV cache access overheads from slow mediums, AttentionStore employs layer-wise pre-loading and asynchronous saving schemes to overlap the KV cache access with the GPU computation. To ensure that the KV caches to be accessed are placed in the fastest hierarchy, AttentionStore employs scheduler-aware fetching and eviction schemes to consciously place the KV caches in different layers based on the hints from the inference job scheduler. To avoid the invalidation of the saved KV caches incurred by context window overflow, AttentionStore enables the saved KV caches to remain valid via decoupling the positional encoding and effectively truncating the KV caches. Extensive experimental results demonstrate that AttentionStore significantly decreases the time to the first token (TTFT) by up to 88%, improves the prompt prefilling throughput by 8.2$\times$ for multi-turn conversations, and reduces the end-to-end inference cost by up to 56%. For long sequence inference, AttentionStore reduces the TTFT by up to 95% and improves the prompt prefilling throughput by 22$\times$.
Authors: Beliz Gunel, James B. Wendt, Jing Xie, Yichao Zhou, Nguyen Vo, Zachary Fisher, Sandeep Tata
Abstract: Users often struggle with decision-making between two options (A vs B), as it usually requires time-consuming research across multiple web pages. We propose STRUM-LLM that addresses this challenge by generating attributed, structured, and helpful contrastive summaries that highlight key differences between the two options. STRUM-LLM identifies helpful contrast: the specific attributes along which the two options differ significantly and which are most likely to influence the user's decision. Our technique is domain-agnostic, and does not require any human-labeled data or fixed attribute list as supervision. STRUM-LLM attributes all extractions back to the input sources along with textual evidence, and it does not have a limit on the length of input sources that it can process. STRUM-LLM Distilled has 100x more throughput than the models with comparable performance while being 10x smaller. In this paper, we provide extensive evaluations for our method and lay out future directions for our currently deployed system.
Authors: Jingyuan Wang, Shengdong Xu, Yang Yang
Abstract: This report provide a detailed description of the method that we proposed in the TRAC-2024 Offline Harm Potential dentification which encloses two sub-tasks. The investigation utilized a rich dataset comprised of social media comments in several Indian languages, annotated with precision by expert judges to capture the nuanced implications for offline context harm. The objective assigned to the participants was to design algorithms capable of accurately assessing the likelihood of harm in given situations and identifying the most likely target(s) of offline harm. Our approach ranked second in two separate tracks, with F1 values of 0.73 and 0.96 respectively. Our method principally involved selecting pretrained models for finetuning, incorporating contrastive learning techniques, and culminating in an ensemble approach for the test set.
Authors: Jingtao Zhan, Qingyao Ai, Yiqun Liu, Jia Chen, Shaoping Ma
Abstract: Text-to-image generation systems have emerged as revolutionary tools in the realm of artistic creation, offering unprecedented ease in transforming textual prompts into visual art. However, the efficacy of these systems is intricately linked to the quality of user-provided prompts, which often poses a challenge to users unfamiliar with prompt crafting. This paper addresses this challenge by leveraging user reformulation data from interaction logs to develop an automatic prompt reformulation model. Our in-depth analysis of these logs reveals that user prompt reformulation is heavily dependent on the individual user's capability, resulting in significant variance in the quality of reformulation pairs. To effectively use this data for training, we introduce the Capability-aware Prompt Reformulation (CAPR) framework. CAPR innovatively integrates user capability into the reformulation process through two key components: the Conditional Reformulation Model (CRM) and Configurable Capability Features (CCF). CRM reformulates prompts according to a specified user capability, as represented by CCF. The CCF, in turn, offers the flexibility to tune and guide the CRM's behavior. This enables CAPR to effectively learn diverse reformulation strategies across various user capacities and to simulate high-capability user reformulation during inference. Extensive experiments on standard text-to-image generation benchmarks showcase CAPR's superior performance over existing baselines and its remarkable robustness on unseen systems. Furthermore, comprehensive analyses validate the effectiveness of different components. CAPR can facilitate user-friendly interaction with text-to-image systems and make advanced artistic creation more achievable for a broader range of users.
Authors: Rihui Jin, Yu Li, Guilin Qi, Nan Hu, Yuan-Fang Li, Jiaoyan Chen, Jianan Wang, Yongrui Chen, Dehai Min
Abstract: Table understanding (TU) has achieved promising advancements, but it faces the challenges of the scarcity of manually labeled tables and the presence of complex table structures.To address these challenges, we propose HGT, a framework with a heterogeneous graph (HG)-enhanced large language model (LLM) to tackle few-shot TU tasks.It leverages the LLM by aligning the table semantics with the LLM's parametric knowledge through soft prompts and instruction turning and deals with complex tables by a multi-task pre-training scheme involving three novel multi-granularity self-supervised HG pre-training objectives.We empirically demonstrate the effectiveness of HGT, showing that it outperforms the SOTA for few-shot complex TU on several benchmarks.
Authors: Yaqi Xie, Anjali Rawal, Yujing Cen, Dixuan Zhao, Sunil K Narang, Shanu Sushmita
Abstract: As advanced modern systems like deep neural networks (DNNs) and generative AI continue to enhance their capabilities in producing convincing and realistic content, the need to distinguish between user-generated and machine generated content is becoming increasingly evident. In this research, we undertake a comparative evaluation of eight traditional machine-learning algorithms to distinguish between machine-generated and human-generated data across three diverse datasets: Poems, Abstracts, and Essays. Our results indicate that traditional methods demonstrate a high level of accuracy in identifying machine-generated data, reflecting the documented effectiveness of popular pre-trained models like RoBERT. We note that machine-generated texts tend to be shorter and exhibit less word variety compared to human-generated content. While specific domain-related keywords commonly utilized by humans, albeit disregarded by current LLMs (Large Language Models), may contribute to this high detection accuracy, we show that deeper word representations like word2vec can capture subtle semantic variances. Furthermore, readability, bias, moral, and affect comparisons reveal a discernible contrast between machine-generated and human generated content. There are variations in expression styles and potentially underlying biases in the data sources (human and machine-generated). This study provides valuable insights into the advancing capacities and challenges associated with machine-generated content across various domains.
Authors: Nesrine Bannour (STL), Christophe Servan (STL), Aur\'elie N\'ev\'eol (STL), Xavier Tannier (LIMICS)
Abstract: Background: Transformer-based language models have shown strong performance on many Natural LanguageProcessing (NLP) tasks. Masked Language Models (MLMs) attract sustained interest because they can be adaptedto different languages and sub-domains through training or fine-tuning on specific corpora while remaining lighterthan modern Large Language Models (LLMs). Recently, several MLMs have been released for the biomedicaldomain in French, and experiments suggest that they outperform standard French counterparts. However, nosystematic evaluation comparing all models on the same corpora is available. Objective: This paper presentsan evaluation of masked language models for biomedical French on the task of clinical named entity recognition.Material and methods: We evaluate biomedical models CamemBERT-bio and DrBERT and compare them tostandard French models CamemBERT, FlauBERT and FrALBERT as well as multilingual mBERT using three publicallyavailable corpora for clinical named entity recognition in French. The evaluation set-up relies on gold-standardcorpora as released by the corpus developers. Results: Results suggest that CamemBERT-bio outperformsDrBERT consistently while FlauBERT offers competitive performance and FrAlBERT achieves the lowest carbonfootprint. Conclusion: This is the first benchmark evaluation of biomedical masked language models for Frenchclinical entity recognition that compares model performance consistently on nested entity recognition using metricscovering performance and environmental impact.
Authors: Nad\`ege Alavoine (STL), Ga\"elle Laperriere (LIA), Christophe Servan (STL), Sahar Ghannay (STL), Sophie Rosset (STL)
Abstract: Intent classification and slot-filling are essential tasks of Spoken Language Understanding (SLU). In most SLUsystems, those tasks are realized by independent modules. For about fifteen years, models achieving both of themjointly and exploiting their mutual enhancement have been proposed. A multilingual module using a joint modelwas envisioned to create a touristic dialogue system for a European project, HumanE-AI-Net. A combination ofmultiple datasets, including the MEDIA dataset, was suggested for training this joint model. The MEDIA SLU datasetis a French dataset distributed since 2005 by ELRA, mainly used by the French research community and free foracademic research since 2020. Unfortunately, it is annotated only in slots but not intents. An enhanced version ofMEDIA annotated with intents has been built to extend its use to more tasks and use cases. This paper presents thesemi-automatic methodology used to obtain this enhanced version. In addition, we present the first results of SLUexperiments on this enhanced dataset using joint models for intent classification and slot-filling.
Authors: Jayathi Hewapathirana, Deshan Sumanathilaka
Abstract: This work explores the utilization of Romanized Sinhala social media data to identify individuals at risk of depression. A machine learning-based framework is presented for the automatic screening of depression symptoms by analyzing language patterns, sentiment, and behavioural cues within a comprehensive dataset of social media posts. The research has been carried out to compare the suitability of Neural Networks over the classical machine learning techniques. The proposed Neural Network with an attention layer which is capable of handling long sequence data, attains a remarkable accuracy of 93.25% in detecting depression symptoms, surpassing current state-of-the-art methods. These findings underscore the efficacy of this approach in pinpointing individuals in need of proactive interventions and support. Mental health professionals, policymakers, and social media companies can gain valuable insights through the proposed model. Leveraging natural language processing techniques and machine learning algorithms, this work offers a promising pathway for mental health screening in the digital era. By harnessing the potential of social media data, the framework introduces a proactive method for recognizing and assisting individuals at risk of depression. In conclusion, this research contributes to the advancement of proactive interventions and support systems for mental health, thereby influencing both research and practical applications in the field.
Authors: Mohsen Gholami, Mohammad Akbari, Cindy Hu, Vaden Masrani, Z. Jane Wang, Yong Zhang
Abstract: Knowledge distillation from LLMs is essential for the efficient deployment of language models. Prior works have proposed data generation using LLMs for preparing distilled models. We argue that generating data with LLMs is prone to sampling mainly from the center of original content distribution. This limitation hinders the distilled model from learning the true underlying data distribution and to forget the tails of the distributions (samples with lower probability). To this end, we propose GOLD, a task-agnostic data generation and knowledge distillation framework, which employs an iterative out-of-distribution-guided feedback mechanism for the LLM. As a result, the generated data improves the generalizability of distilled models. An energy-based OOD evaluation approach is also introduced to deal with noisy generated data. Our extensive experiments on 10 different classification and sequence-to-sequence tasks in NLP show that GOLD respectively outperforms prior arts and the LLM with an average improvement of 5% and 14%. We will also show that the proposed method is applicable to less explored and novel tasks. The code is available.
Authors: Niall Taylor, Dan Schofield, Andrey Kormilitzin, Dan W Joyce, Alejo Nevado-Holgado
Abstract: Pre-trained Large Language Models (LLMs) often struggle on out-of-domain datasets like healthcare focused text. We explore specialized pre-training to adapt smaller LLMs to different healthcare datasets. Three methods are assessed: traditional masked language modeling, Deep Contrastive Learning for Unsupervised Textual Representations (DeCLUTR), and a novel pre-training objective utilizing metadata categories from the healthcare settings. These schemes are evaluated on downstream document classification tasks for each dataset, with additional analysis of the resultant embedding spaces. Contrastively trained models outperform other approaches on the classification tasks, delivering strong performance from limited labeled data and with fewer model parameter updates required. While metadata-based pre-training does not further improve classifications across the datasets, it yields interesting embedding cluster separability. All domain adapted LLMs outperform their publicly available general base LLM, validating the importance of domain-specialization. This research illustrates efficient approaches to instill healthcare competency in compact LLMs even under tight computational budgets, an essential capability for responsible and sustainable deployment in local healthcare settings. We provide pre-training guidelines for specialized healthcare LLMs, motivate continued inquiry into contrastive objectives, and demonstrates adaptation techniques to align small LLMs with privacy-sensitive medical tasks.
Authors: Yash Jain, David Chan, Pranav Dheram, Aparna Khare, Olabanji Shonibare, Venkatesh Ravichandran, Shalini Ghosh
Abstract: Recent advances in machine learning have demonstrated that multi-modal pre-training can improve automatic speech recognition (ASR) performance compared to randomly initialized models, even when models are fine-tuned on uni-modal tasks. Existing multi-modal pre-training methods for the ASR task have primarily focused on single-stage pre-training where a single unsupervised task is used for pre-training followed by fine-tuning on the downstream task. In this work, we introduce a novel method combining multi-modal and multi-task unsupervised pre-training with a translation-based supervised mid-training approach. We empirically demonstrate that such a multi-stage approach leads to relative word error rate (WER) improvements of up to 38.45% over baselines on both Librispeech and SUPERB. Additionally, we share several important findings for choosing pre-training methods and datasets.
Authors: Kanishka Misra, Kyle Mahowald
Abstract: Language models learn rare syntactic phenomena, but it has been argued that they rely on rote memorization, as opposed to grammatical generalization. Training on a corpus of human-scale in size (100M words), we iteratively trained transformer language models on systematically manipulated corpora and then evaluated their learning of a particular rare grammatical phenomenon: the English Article+Adjective+Numeral+Noun (AANN) construction (``a beautiful five days''). We first compared how well this construction was learned on the default corpus relative to a counterfactual corpus in which the AANN sentences were removed. AANNs were still learned better than systematically perturbed variants of the construction. Using additional counterfactual corpora, we suggest that this learning occurs through generalization from related constructions (e.g., ``a few days''). An additional experiment showed that this learning is enhanced when there is more variability in the input. Taken together, our results provide an existence proof that models learn rare grammatical phenomena by generalization from less rare phenomena. Code available at https://github.com/kanishkamisra/aannalysis
Authors: Nazanin Jafari, James Allan
Abstract: Identifying the targets of hate speech is a crucial step in grasping the nature of such speech and, ultimately, in improving the detection of offensive posts on online forums. Much harmful content on online platforms uses implicit language especially when targeting vulnerable and protected groups such as using stereotypical characteristics instead of explicit target names, making it harder to detect and mitigate the language. In this study, we focus on identifying implied targets of hate speech, essential for recognizing subtler hate speech and enhancing the detection of harmful content on digital platforms. We define a new task aimed at identifying the targets even when they are not explicitly stated. To address that task, we collect and annotate target spans in three prominent implicit hate speech datasets: SBIC, DynaHate, and IHC. We call the resulting merged collection Implicit-Target-Span. The collection is achieved using an innovative pooling method with matching scores based on human annotations and Large Language Models (LLMs). Our experiments indicate that Implicit-Target-Span provides a challenging test bed for target span detection methods.
Authors: Niklas Stoehr, Mitchell Gordon, Chiyuan Zhang, Owen Lewis
Abstract: Can we localize the weights and mechanisms used by a language model to memorize and recite entire paragraphs of its training data? In this paper, we show that while memorization is spread across multiple layers and model components, gradients of memorized paragraphs have a distinguishable spatial pattern, being larger in lower model layers than gradients of non-memorized examples. Moreover, the memorized examples can be unlearned by fine-tuning only the high-gradient weights. We localize a low-layer attention head that appears to be especially involved in paragraph memorization. This head is predominantly focusing its attention on distinctive, rare tokens that are least frequent in a corpus-level unigram distribution. Next, we study how localized memorization is across the tokens in the prefix by perturbing tokens and measuring the caused change in the decoding. A few distinctive tokens early in a prefix can often corrupt the entire continuation. Overall, memorized continuations are not only harder to unlearn, but also to corrupt than non-memorized ones.
Authors: Opher Lieber, Barak Lenz, Hofit Bata, Gal Cohen, Jhonathan Osin, Itay Dalmedigos, Erez Safahi, Shaked Meirom, Yonatan Belinkov, Shai Shalev-Shwartz, Omri Abend, Raz Alon, Tomer Asida, Amir Bergman, Roman Glozman, Michael Gokhman, Avashalom Manevich, Nir Ratner, Noam Rozen, Erez Shwartz, Mor Zusman, Yoav Shoham
Abstract: We present Jamba, a new base large language model based on a novel hybrid Transformer-Mamba mixture-of-experts (MoE) architecture. Specifically, Jamba interleaves blocks of Transformer and Mamba layers, enjoying the benefits of both model families. MoE is added in some of these layers to increase model capacity while keeping active parameter usage manageable. This flexible architecture allows resource- and objective-specific configurations. In the particular configuration we have implemented, we end up with a powerful model that fits in a single 80GB GPU. Built at large scale, Jamba provides high throughput and small memory footprint compared to vanilla Transformers, and at the same time state-of-the-art performance on standard language model benchmarks and long-context evaluations. Remarkably, the model presents strong results for up to 256K tokens context length. We study various architectural decisions, such as how to combine Transformer and Mamba layers, and how to mix experts, and show that some of them are crucial in large scale modeling. We also describe several interesting properties of these architectures which the training and evaluation of Jamba have revealed, and plan to release checkpoints from various ablation runs, to encourage further exploration of this novel architecture. We make the weights of our implementation of Jamba publicly available under a permissive license.
Authors: Shengjie Liu, Jing Wu, Jingyuan Bao, Wenyi Wang, Naira Hovakimyan, Christopher G Healey
Abstract: This paper describes an investigation of the robustness of large language models (LLMs) for retrieval augmented generation (RAG)-based summarization tasks. While LLMs provide summarization capabilities, their performance in complex, real-world scenarios remains under-explored. Our first contribution is LogicSumm, an innovative evaluation framework incorporating realistic scenarios to assess LLM robustness during RAG-based summarization. Based on limitations identified by LogiSumm, we then developed SummRAG, a comprehensive system to create training dialogues and fine-tune a model to enhance robustness within LogicSumm's scenarios. SummRAG is an example of our goal of defining structured methods to test the capabilities of an LLM, rather than addressing issues in a one-off fashion. Experimental results confirm the power of SummRAG, showcasing improved logical coherence and summarization quality. Data, corresponding model weights, and Python code are available online.
Authors: Peng Ding, Jiading Fang, Peng Li, Kangrui Wang, Xiaochen Zhou, Mo Yu, Jing Li, Matthew R. Walter, Hongyuan Mei
Abstract: Large language models such as ChatGPT and GPT-4 have recently achieved astonishing performance on a variety of natural language processing tasks. In this paper, we propose MANGO, a benchmark to evaluate their capabilities to perform text-based mapping and navigation. Our benchmark includes 53 mazes taken from a suite of textgames: each maze is paired with a walkthrough that visits every location but does not cover all possible paths. The task is question-answering: for each maze, a large language model reads the walkthrough and answers hundreds of mapping and navigation questions such as "How should you go to Attic from West of House?" and "Where are we if we go north and east from Cellar?". Although these questions are easy to humans, it turns out that even GPT-4, the best-to-date language model, performs poorly at answering them. Further, our experiments suggest that a strong mapping and navigation ability would benefit large language models in performing relevant downstream tasks, such as playing textgames. Our MANGO benchmark will facilitate future research on methods that improve the mapping and navigation capabilities of language models. We host our leaderboard, data, code, and evaluation program at https://mango.ttic.edu and https://github.com/oaklight/mango/.
URLs: https://mango.ttic.edu, https://github.com/oaklight/mango/.
Authors: Hanting Chen, Zhicheng Liu, Xutao Wang, Yuchuan Tian, Yunhe Wang
Abstract: In an effort to reduce the computational load of Transformers, research on linear attention has gained significant momentum. However, the improvement strategies for attention mechanisms typically necessitate extensive retraining, which is impractical for large language models with a vast array of parameters. In this paper, we present DiJiang, a novel Frequency Domain Kernelization approach that enables the transformation of a pre-trained vanilla Transformer into a linear complexity model with little training costs. By employing a weighted Quasi-Monte Carlo method for sampling, the proposed approach theoretically offers superior approximation efficiency. To further reduce the training computational complexity, our kernelization is based on Discrete Cosine Transform (DCT) operations. Extensive experiments demonstrate that the proposed method achieves comparable performance to the original Transformer, but with significantly reduced training costs and much faster inference speeds. Our DiJiang-7B achieves comparable performance with LLaMA2-7B on various benchmark while requires only about 1/50 training cost. Code is available at https://github.com/YuchuanTian/DiJiang.
Authors: Shulin Liu, Chengcheng Xu, Hao Liu, Tinghao Yu, Tao Yang
Abstract: The recent success of Large Language Models (LLMs) has garnered significant attention in both academia and industry. Prior research on LLMs has primarily focused on enhancing or leveraging their generalization capabilities in zero- and few-shot settings. However, there has been limited investigation into effectively fine-tuning LLMs for a specific natural language understanding task in supervised settings. In this study, we conduct an experimental analysis by fine-tuning LLMs for the task of Chinese short text matching. We explore various factors that influence performance when fine-tuning LLMs, including task modeling methods, prompt formats, and output formats.
Authors: Hao Wu, Fan Xu
Abstract: Building natural language interfaces typically uses a semantic parser to parse the user's natural language and convert it into structured \textbf{S}emantic \textbf{L}ogic \textbf{F}orms (SLFs). The mainstream approach is to adopt a sequence-to-sequence framework, which requires that natural language commands and SLFs must be represented serially. Since a single natural language may have multiple SLFs or multiple natural language commands may have the same SLF, training a sequence-to-sequence model is sensitive to the choice among them, a phenomenon recorded as "order matters". To solve this problem, we propose a novel neural network, SLFNet, which firstly incorporates dependent syntactic information as prior knowledge and can capture the long-range interactions between contextual information and words. Secondly construct semantic probability graphs to obtain local dependencies between predictor variables. Finally we propose the Multi-Head SLF Attention mechanism to synthesize SLFs from natural language commands based on Sequence-to-Slots. Experiments show that SLFNet achieves state-of-the-art performance on the ChineseQCI-TS and Okapi datasets, and competitive performance on the ATIS dataset.
Authors: Qinhao Zhou, Zihan Zhang, Xiang Xiang, Ke Wang, Yuchuan Wu, Yongbin Li
Abstract: Open-source pre-trained Large Language Models (LLMs) exhibit strong language understanding and generation capabilities, making them highly successful in a variety of tasks. However, when used as agents for dealing with complex problems in the real world, their performance is far inferior to large commercial models such as ChatGPT and GPT-4. As intelligent agents, LLMs need to have the capabilities of task planning, long-term memory, and the ability to leverage external tools to achieve satisfactory performance. Various methods have been proposed to enhance the agent capabilities of LLMs. On the one hand, methods involve constructing agent-specific data and fine-tuning the models. On the other hand, some methods focus on designing prompts that effectively activate the reasoning abilities of the LLMs. We explore both strategies on the 7B and 13B models. We propose a comprehensive method for constructing agent-specific data using GPT-4. Through supervised fine-tuning with constructed data, we find that for these models with a relatively small number of parameters, supervised fine-tuning can significantly reduce hallucination outputs and formatting errors in agent tasks. Furthermore, techniques such as multi-path reasoning and task decomposition can effectively decrease problem complexity and enhance the performance of LLMs as agents. We evaluate our method on five agent tasks of AgentBench and achieve satisfactory results.
Authors: Shuyao Xu, Long Qin, Tianyang Chen, Zhenzhou Zha, Bingxue Qiu, Weizhi Wang
Abstract: In second language learning, scenario-based conversation practice is important for language learners to achieve fluency in speaking, but students often lack sufficient opportunities to practice their conversational skills with qualified instructors or native speakers. To bridge this gap, we propose situational dialogue models for students to engage in conversational practice. Our situational dialogue models are fine-tuned on large language models (LLMs), with the aim of combining the engaging nature of an open-ended conversation with the focused practice of scenario-based tasks. Leveraging the generalization capabilities of LLMs, we demonstrate that our situational dialogue models perform effectively not only on training topics but also on topics not encountered during training. This offers a promising solution to support a wide range of conversational topics without extensive manual work. Additionally, research in the field of dialogue systems still lacks reliable automatic evaluation metrics, leading to human evaluation as the gold standard (Smith et al., 2022), which is typically expensive. To address the limitations of existing evaluation methods, we present a novel automatic evaluation method that employs fine-tuned LLMs to efficiently and effectively assess the performance of situational dialogue models.
Authors: Che Jiang, Biqing Qi, Xiangyu Hong, Dayuan Fu, Yang Cheng, Fandong Meng, Mo Yu, Bowen Zhou, Jie Zhou
Abstract: Large language models are successful in answering factoid questions but are also prone to hallucination.We investigate the phenomenon of LLMs possessing correct answer knowledge yet still hallucinating from the perspective of inference dynamics, an area not previously covered in studies on hallucinations.We are able to conduct this analysis via two key ideas.First, we identify the factual questions that query the same triplet knowledge but result in different answers. The difference between the model behaviors on the correct and incorrect outputs hence suggests the patterns when hallucinations happen. Second, to measure the pattern, we utilize mappings from the residual streams to vocabulary space. We reveal the different dynamics of the output token probabilities along the depths of layers between the correct and hallucinated cases. In hallucinated cases, the output token's information rarely demonstrates abrupt increases and consistent superiority in the later stages of the model. Leveraging the dynamic curve as a feature, we build a classifier capable of accurately detecting hallucinatory predictions with an 88\% success rate. Our study shed light on understanding the reasons for LLMs' hallucinations on their known facts, and more importantly, on accurately predicting when they are hallucinating.
Authors: Juhwan Choi, YoungBin Kim
Abstract: In the field of text data augmentation, rule-based methods are widely adopted for real-world applications owing to their cost-efficiency. However, conventional rule-based approaches suffer from the possibility of losing the original semantics of the given text. We propose a novel text data augmentation strategy that avoids such phenomena through a straightforward deletion of adverbs, which play a subsidiary role in the sentence. Our comprehensive experiments demonstrate the efficiency and effectiveness of our proposed approach for not just single text classification, but also natural language inference that requires semantic preservation. We publicly released our source code for reproducibility.
Authors: Luchang Li, Sheng Qian, Jie Lu, Lunxi Yuan, Rui Wang, Qin Xie
Abstract: The Large Language Model (LLM) is widely employed for tasks such as intelligent assistants, text summarization, translation, and multi-modality on mobile phones. However, the current methods for on-device LLM deployment maintain slow inference speed, which causes poor user experience. To facilitate high-efficiency LLM deployment on device GPUs, we propose four optimization techniques: (a) a symbolic expression-based approach to support dynamic shape model inference; (b) operator optimizations and execution priority setting to enhance inference speed and reduce phone lagging; (c) an FP4 quantization method termed M0E4 to reduce dequantization overhead; (d) a sub-tensor-based technique to eliminate the need for copying KV cache after LLM inference. Furthermore, we implement these methods in our mobile inference engine, Transformer-Lite, which is compatible with both Qualcomm and MTK processors. We evaluated Transformer-Lite's performance using LLMs with varied architectures and parameters ranging from 2B to 14B. Specifically, we achieved prefill and decoding speeds of 121 token/s and 14 token/s for ChatGLM2 6B, and 330 token/s and 30 token/s for smaller Gemma 2B, respectively. Compared with CPU-based FastLLM and GPU-based MLC-LLM, our engine attains over 10x speedup for the prefill speed and 2~3x speedup for the decoding speed.
Authors: Yongqi Tong, Dawei Li, Sizhe Wang, Yujia Wang, Fei Teng, Jingbo Shang
Abstract: Recent works have shown the benefits to LLMs from fine-tuning golden-standard Chain-of-Thought (CoT) rationales or using them as correct examples in few-shot prompting. While humans can indeed imitate correct examples, learning from our mistakes is another vital aspect of human cognition. Hence, a question naturally arises: \textit{can LLMs learn and benefit from their mistakes, especially for their reasoning? } This study investigates this problem from both the prompting and model-tuning perspectives. We begin by introducing \textsc{CoTErrorSet}, a new benchmark with 609,432 questions, each designed with both correct and error references, and demonstrating the types and reasons for making such mistakes. To explore the effectiveness of those mistakes, we design two methods: (1) \textbf{Self-rethinking} prompting guides LLMs to rethink whether they have made similar previous mistakes; and (2) \textbf{Mistake tuning} involves finetuning models in both correct and incorrect reasoning domains, rather than only tuning models to learn ground truth in traditional methodology. We conduct a series of experiments to prove LLMs can obtain benefits from mistakes in both directions. Our two methods offer potentially cost-effective strategies by leveraging errors to enhance reasoning capabilities, which costs significantly less than creating meticulously hand-crafted golden references. We ultimately make a thorough analysis of the reasons behind LLMs' errors, which provides directions that future research needs to overcome. \textsc{CoTErrorSet} will be published soon on \texttt{Anonymity Link}.
Authors: Shadi Manafi, Nikhil Krishnaswamy
Abstract: Multilingual Language Models (MLLMs) exhibit robust cross-lingual transfer capabilities, or the ability to leverage information acquired in a source language and apply it to a target language. These capabilities find practical applications in well-established Natural Language Processing (NLP) tasks such as Named Entity Recognition (NER). This study aims to investigate the effectiveness of a source language when applied to a target language, particularly in the context of perturbing the input test set. We evaluate on 13 pairs of languages, each including one high-resource language (HRL) and one low-resource language (LRL) with a geographic, genetic, or borrowing relationship. We evaluate two well-known MLLMs--MBERT and XLM-R--on these pairs, in native LRL and cross-lingual transfer settings, in two tasks, under a set of different perturbations. Our findings indicate that NER cross-lingual transfer depends largely on the overlap of entity chunks. If a source and target language have more entities in common, the transfer ability is stronger. Models using cross-lingual transfer also appear to be somewhat more robust to certain perturbations of the input, perhaps indicating an ability to leverage stronger representations derived from the HRL. Our research provides valuable insights into cross-lingual transfer and its implications for NLP applications, and underscores the need to consider linguistic nuances and potential limitations when employing MLLMs across distinct languages.
Authors: Kanij Fatema, Fazle Dawood Haider, Nirzona Ferdousi Turpa, Tanveer Azmal, Sourav Ahmed, Navid Hasan, Mohammad Akhlaqur Rahman, Biplab Kumar Sarkar, Afrar Jahin, Md. Rezuwan Hassan, Md Foriduzzaman Zihad, Rubayet Sabbir Faruque, Asif Sushmit, Mashrur Imtiaz, Farig Sadeque, Syed Shahrier Rahman
Abstract: The International Phonetic Alphabet (IPA) serves to systematize phonemes in language, enabling precise textual representation of pronunciation. In Bengali phonology and phonetics, ongoing scholarly deliberations persist concerning the IPA standard and core Bengali phonemes. This work examines prior research, identifies current and potential issues, and suggests a framework for a Bengali IPA standard, facilitating linguistic analysis and NLP resource creation and downstream technology development. In this work, we present a comprehensive study of Bengali IPA transcription and introduce a novel IPA transcription framework incorporating a novel dataset with DL-based benchmarks.
Authors: Fahim Faisal, Antonios Anastasopoulos
Abstract: The capacity and effectiveness of pre-trained multilingual models (MLMs) for zero-shot cross-lingual transfer is well established. However, phenomena of positive or negative transfer, and the effect of language choice still need to be fully understood, especially in the complex setting of massively multilingual LMs. We propose an \textit{efficient} method to study transfer language influence in zero-shot performance on another target language. Unlike previous work, our approach disentangles downstream tasks from language, using dedicated adapter units. Our findings suggest that some languages do not largely affect others, while some languages, especially ones unseen during pre-training, can be extremely beneficial or detrimental for different target languages. We find that no transfer language is beneficial for all target languages. We do, curiously, observe languages previously unseen by MLMs consistently benefit from transfer from almost any language. We additionally use our modular approach to quantify negative interference efficiently and categorize languages accordingly. Furthermore, we provide a list of promising transfer-target language configurations that consistently lead to target language performance improvements. Code and data are publicly available: https://github.com/ffaisal93/neg_inf
Authors: Benjamin Townsend, Madison May, Christopher Wells
Abstract: We introduce RealKIE, a benchmark of five challenging datasets aimed at advancing key information extraction methods, with an emphasis on enterprise applications. The datasets include a diverse range of documents including SEC S1 Filings, US Non-disclosure Agreements, UK Charity Reports, FCC Invoices, and Resource Contracts. Each presents unique challenges: poor text serialization, sparse annotations in long documents, and complex tabular layouts. These datasets provide a realistic testing ground for key information extraction tasks like investment analysis and legal data processing. In addition to presenting these datasets, we offer an in-depth description of the annotation process, document processing techniques, and baseline modeling approaches. This contribution facilitates the development of NLP models capable of handling practical challenges and supports further research into information extraction technologies applicable to industry-specific problems. The annotated data and OCR outputs are available to download at https://indicodatasolutions.github.io/RealKIE/ code to reproduce the baselines will be available shortly.
Authors: Helena Bonaldi, Yi-Ling Chung, Gavin Abercrombie, Marco Guerini
Abstract: In recent years, counterspeech has emerged as one of the most promising strategies to fight online hate. These non-escalatory responses tackle online abuse while preserving the freedom of speech of the users, and can have a tangible impact in reducing online and offline violence. Recently, there has been growing interest from the Natural Language Processing (NLP) community in addressing the challenges of analysing, collecting, classifying, and automatically generating counterspeech, to reduce the huge burden of manually producing it. In particular, researchers have taken different directions in addressing these challenges, thus providing a variety of related tasks and resources. In this paper, we provide a guide for doing research on counterspeech, by describing - with detailed examples - the steps to undertake, and providing best practices that can be learnt from the NLP studies on this topic. Finally, we discuss open challenges and future directions of counterspeech research in NLP.
Authors: Megan Su, Yuwei Bao
Abstract: Interactive Artificial Intelligent(AI) assistant systems are designed to offer timely guidance to help human users to complete a variety tasks. One of the remaining challenges is to understand user's mental states during the task for more personalized guidance. In this work, we analyze users' mental states during task executions and investigate the capabilities and challenges for large language models to interpret user profiles for more personalized user guidance.
Authors: Manjeet Yadav, Nilesh Kumar Sahu, Mudita Chaturvedi, Snehil Gupta, Haroon R Lone
Abstract: Improving mental health support in developing countries is a pressing need. One potential solution is the development of scalable, automated systems to conduct diagnostic screenings, which could help alleviate the burden on mental health professionals. In this work, we evaluate several state-of-the-art Large Language Models (LLMs), with and without fine-tuning, on our custom dataset for generating concise summaries from mental state examinations. We rigorously evaluate four different models for summary generation using established ROUGE metrics and input from human evaluators. The results highlight that our top-performing fine-tuned model outperforms existing models, achieving ROUGE-1 and ROUGE-L values of 0.810 and 0.764, respectively. Furthermore, we assessed the fine-tuned model's generalizability on a publicly available D4 dataset, and the outcomes were promising, indicating its potential applicability beyond our custom dataset.
Authors: Nihar Ranjan Sahoo, Pranamya Prashant Kulkarni, Narjis Asad, Arif Ahmad, Tanu Goyal, Aparna Garimella, Pushpak Bhattacharyya
Abstract: The pervasive influence of social biases in language data has sparked the need for benchmark datasets that capture and evaluate these biases in Large Language Models (LLMs). Existing efforts predominantly focus on English language and the Western context, leaving a void for a reliable dataset that encapsulates India's unique socio-cultural nuances. To bridge this gap, we introduce IndiBias, a comprehensive benchmarking dataset designed specifically for evaluating social biases in the Indian context. We filter and translate the existing CrowS-Pairs dataset to create a benchmark dataset suited to the Indian context in Hindi language. Additionally, we leverage LLMs including ChatGPT and InstructGPT to augment our dataset with diverse societal biases and stereotypes prevalent in India. The included bias dimensions encompass gender, religion, caste, age, region, physical appearance, and occupation. We also build a resource to address intersectional biases along three intersectional dimensions. Our dataset contains 800 filtered sentences from the CrowS-Pairs dataset and tuples for bias measurement across different demographics. It is made available in English and Hindi languages, providing a size comparable to existing benchmark datasets. Furthermore, using IndiBias we compare ten different language models on multiple bias measurement metrics. We observed that the language models exhibit more bias across a majority of the intersectional groups.
Authors: Francois Meyer, Jan Buys
Abstract: Multilingual modelling can improve machine translation for low-resource languages, partly through shared subword representations. This paper studies the role of subword segmentation in cross-lingual transfer. We systematically compare the efficacy of several subword methods in promoting synergy and preventing interference across different linguistic typologies. Our findings show that subword regularisation boosts synergy in multilingual modelling, whereas BPE more effectively facilitates transfer during cross-lingual fine-tuning. Notably, our results suggest that differences in orthographic word boundary conventions (the morphological granularity of written words) may impede cross-lingual transfer more significantly than linguistic unrelatedness. Our study confirms that decisions around subword modelling can be key to optimising the benefits of multilingual modelling.
Authors: Zehao Wen, Rabih Younes
Abstract: In our rapidly evolving digital sphere, the ability to discern media bias becomes crucial as it can shape public sentiment and influence pivotal decisions. The advent of large language models (LLMs), such as ChatGPT, noted for their broad utility in various natural language processing (NLP) tasks, invites exploration of their efficacy in media bias detection. Can ChatGPT detect media bias? This study seeks to answer this question by leveraging the Media Bias Identification Benchmark (MBIB) to assess ChatGPT's competency in distinguishing six categories of media bias, juxtaposed against fine-tuned models such as BART, ConvBERT, and GPT-2. The findings present a dichotomy: ChatGPT performs at par with fine-tuned models in detecting hate speech and text-level context bias, yet faces difficulties with subtler elements of other bias detections, namely, fake news, racial, gender, and cognitive biases.
Authors: Po-Heng Chen, Sijia Cheng, Wei-Lin Chen, Yen-Ting Lin, Yun-Nung Chen
Abstract: The evaluation of large language models (LLMs) has drawn substantial attention in the field recently. This work focuses on evaluating LLMs in a Chinese context, specifically, for Traditional Chinese which has been largely underrepresented in existing benchmarks. We present TMLU, a holistic evaluation suit tailored for assessing the advanced knowledge and reasoning capability in LLMs, under the context of Taiwanese Mandarin. TMLU consists of an array of 37 subjects across social science, STEM, humanities, Taiwan-specific content, and others, ranging from middle school to professional levels. In addition, we curate chain-of-thought-like few-shot explanations for each subject to facilitate the evaluation of complex reasoning skills. To establish a comprehensive baseline, we conduct extensive experiments and analysis on 24 advanced LLMs. The results suggest that Chinese open-weight models demonstrate inferior performance comparing to multilingual proprietary ones, and open-weight models tailored for Taiwanese Mandarin lag behind the Simplified-Chinese counterparts. The findings indicate great headrooms for improvement, and emphasize the goal of TMLU to foster the development of localized Taiwanese-Mandarin LLMs. We release the benchmark and evaluation scripts for the community to promote future research.
Authors: Yingxue Fu
Abstract: Existing discourse corpora are annotated based on different frameworks, which show significant dissimilarities in definitions of arguments and relations and structural constraints. Despite surface differences, these frameworks share basic understandings of discourse relations. The relationship between these frameworks has been an open research question, especially the correlation between relation inventories utilized in different frameworks. Better understanding of this question is helpful for integrating discourse theories and enabling interoperability of discourse corpora annotated under different frameworks. However, studies that explore correlations between discourse relation inventories are hindered by different criteria of discourse segmentation, and expert knowledge and manual examination are typically needed. Some semi-automatic methods have been proposed, but they rely on corpora annotated in multiple frameworks in parallel. In this paper, we introduce a fully automatic approach to address the challenges. Specifically, we extend the label-anchored contrastive learning method introduced by Zhang et al. (2022b) to learn label embeddings during a classification task. These embeddings are then utilized to map discourse relations from different frameworks. We show experimental results on RST-DT (Carlson et al., 2001) and PDTB 3.0 (Prasad et al., 2018).
Authors: Abed Alhakim Freihat, Hadi Khalilia, G\'abor Bella, Fausto Giunchiglia
Abstract: High-quality WordNets are crucial for achieving high-quality results in NLP applications that rely on such resources. However, the wordnets of most languages suffer from serious issues of correctness and completeness with respect to the words and word meanings they define, such as incorrect lemmas, missing glosses and example sentences, or an inadequate, Western-centric representation of the morphology and the semantics of the language. Previous efforts have largely focused on increasing lexical coverage while ignoring other qualitative aspects. In this paper, we focus on the Arabic language and introduce a major revision of the Arabic WordNet that addresses multiple dimensions of lexico-semantic resource quality. As a result, we updated more than 58% of the synsets of the existing Arabic WordNet by adding missing information and correcting errors. In order to address issues of language diversity and untranslatability, we also extended the wordnet structure by new elements: phrasets and lexical gaps.
Authors: Keiichi Namikoshi, Alex Filipowicz, David A. Shamma, Rumen Iliev, Candice L. Hogan, Nikos Arechiga
Abstract: We consider the problem of aligning a large language model (LLM) to model the preferences of a human population. Modeling the beliefs, preferences, and behaviors of a specific population can be useful for a variety of different applications, such as conducting simulated focus groups for new products, conducting virtual surveys, and testing behavioral interventions, especially for interventions that are expensive, impractical, or unethical. Existing work has had mixed success using LLMs to accurately model human behavior in different contexts. We benchmark and evaluate two well-known fine-tuning approaches and evaluate the resulting populations on their ability to match the preferences of real human respondents on a survey of preferences for battery electric vehicles (BEVs). We evaluate our models against their ability to match population-wide statistics as well as their ability to match individual responses, and we investigate the role of temperature in controlling the trade-offs between these two. Additionally, we propose and evaluate a novel loss term to improve model performance on responses that require a numeric response.
Authors: Thibaut Thonet, Jos Rozen, Laurent Besacier
Abstract: Research on Large Language Models (LLMs) has recently witnessed an increasing interest in extending models' context size to better capture dependencies within long documents. While benchmarks have been proposed to assess long-range abilities, existing efforts primarily considered generic tasks that are not necessarily aligned with real-world applications. In contrast, our work proposes a new benchmark for long-context LLMs focused on a practical meeting assistant scenario. In this scenario, the long contexts consist of transcripts obtained by automatic speech recognition, presenting unique challenges for LLMs due to the inherent noisiness and oral nature of such data. Our benchmark, named ELITR-Bench, augments the existing ELITR corpus' transcripts with 271 manually crafted questions and their ground-truth answers. Our experiments with recent long-context LLMs on ELITR-Bench highlight a gap between open-source and proprietary models, especially when questions are asked sequentially within a conversation. We also provide a thorough analysis of our GPT-4-based evaluation method, encompassing insights from a crowdsourcing study. Our findings suggest that while GPT-4's evaluation scores are correlated with human judges', its ability to differentiate among more than three score levels may be limited.
Authors: Julen Etxaniz, Oscar Sainz, Naiara Perez, Itziar Aldabe, German Rigau, Eneko Agirre, Aitor Ormazabal, Mikel Artetxe, Aitor Soroa
Abstract: We introduce Latxa, a family of large language models for Basque ranging from 7 to 70 billion parameters. Latxa is based on Llama 2, which we continue pretraining on a new Basque corpus comprising 4.3M documents and 4.2B tokens. Addressing the scarcity of high-quality benchmarks for Basque, we further introduce 4 multiple choice evaluation datasets: EusProficiency, comprising 5,169 questions from official language proficiency exams; EusReading, comprising 352 reading comprehension questions; EusTrivia, comprising 1,715 trivia questions from 5 knowledge areas; and EusExams, comprising 16,774 questions from public examinations. In our extensive evaluation, Latxa outperforms all previous open models we compare to by a large margin. In addition, it is competitive with GPT-4 Turbo in language proficiency and understanding, despite lagging behind in reading comprehension and knowledge-intensive tasks. Both the Latxa family of models, as well as our new pretraining corpora and evaluation datasets, are publicly available under open licenses at https://github.com/hitz-zentroa/latxa. Our suite enables reproducible research on methods to build LLMs for low-resource languages.
Authors: Caiqi Zhang, Fangyu Liu, Marco Basaldella, Nigel Collier
Abstract: Large Language Models (LLMs) have demonstrated remarkable capability in a variety of NLP tasks. Despite their effectiveness, these models are prone to generate nonfactual content. Uncertainty Quantification (UQ) is pivotal in enhancing our understanding of a model's confidence in its generated content, thereby aiding in the mitigation of nonfactual outputs. Existing research on UQ predominantly targets short text generation, typically yielding brief, word-limited responses. However, real-world applications frequently necessitate much longer responses. Our study first highlights the limitations of current UQ methods in handling long text generation. We then introduce \textsc{Luq}, a novel sampling-based UQ approach specifically designed for long text. Our findings reveal that \textsc{Luq} outperforms existing baseline methods in correlating with the model's factuality scores (negative coefficient of -0.85 observed for Gemini Pro). With \textsc{Luq} as the tool for UQ, we investigate behavior patterns of several popular LLMs' response confidence spectrum and how that interplays with the response' factuality. We identify that LLMs lack confidence in generating long text for rare facts and a factually strong model (i.e. GPT-4) tends to reject questions it is not sure about. To further improve the factual accuracy of LLM responses, we propose a method called \textsc{Luq-Ensemble} that ensembles responses from multiple models and selects the response with the least uncertainty. The ensembling method greatly improves the response factuality upon the best standalone LLM.
Authors: Taha ValizadehAslani, Hualou Liang
Abstract: Fine-tuning a pre-trained model, such as Bidirectional Encoder Representations from Transformers (BERT), has been proven to be an effective method for solving many natural language processing (NLP) tasks. However, due to the large number of parameters in many state-of-the-art NLP models, including BERT, the process of fine-tuning is computationally expensive. One attractive solution to this issue is parameter-efficient fine-tuning, which involves modifying only a minimal segment of the model while keeping the remainder unchanged. Yet, it remains unclear which segment of the BERT model is crucial for fine-tuning. In this paper, we first analyze different components in the BERT model to pinpoint which one undergoes the most significant changes after fine-tuning. We find that output LayerNorm changes more than any other components when fine-tuned for different General Language Understanding Evaluation (GLUE) tasks. Then we show that only fine-tuning the LayerNorm can reach comparable, or in some cases better, performance to full fine-tuning and other parameter-efficient fine-tuning methods. Moreover, we use Fisher information to determine the most critical subset of LayerNorm and demonstrate that many NLP tasks in the GLUE benchmark can be solved by fine-tuning only a small portion of LayerNorm with negligible performance degradation.
Authors: Burcu Sayin, Pasquale Minervini, Jacopo Staiano, Andrea Passerini
Abstract: We explore the potential of Large Language Models (LLMs) to assist and potentially correct physicians in medical decision-making tasks. We evaluate several LLMs, including Meditron, Llama2, and Mistral, to analyze the ability of these models to interact effectively with physicians across different scenarios. We consider questions from PubMedQA and several tasks, ranging from binary (yes/no) responses to long answer generation, where the answer of the model is produced after an interaction with a physician. Our findings suggest that prompt design significantly influences the downstream accuracy of LLMs and that LLMs can provide valuable feedback to physicians, challenging incorrect diagnoses and contributing to more accurate decision-making. For example, when the physician is accurate 38% of the time, Mistral can produce the correct answer, improving accuracy up to 74% depending on the prompt being used, while Llama2 and Meditron models exhibit greater sensitivity to prompt choice. Our analysis also uncovers the challenges of ensuring that LLM-generated suggestions are pertinent and useful, emphasizing the need for further research in this area.
Authors: Fangxu Yu, Junjie Guo, Zhen Wu, Xinyu Dai
Abstract: Emotion Recognition in Conversation (ERC) involves detecting the underlying emotion behind each utterance within a conversation. Effectively generating representations for utterances remains a significant challenge in this task. Recent works propose various models to address this issue, but they still struggle with differentiating similar emotions such as excitement and happiness. To alleviate this problem, We propose an Emotion-Anchored Contrastive Learning (EACL) framework that can generate more distinguishable utterance representations for similar emotions. To achieve this, we utilize label encodings as anchors to guide the learning of utterance representations and design an auxiliary loss to ensure the effective separation of anchors for similar emotions. Moreover, an additional adaptation process is proposed to adapt anchors to serve as effective classifiers to improve classification performance. Across extensive experiments, our proposed EACL achieves state-of-the-art emotion recognition performance and exhibits superior performance on similar emotions. Our code is available at https://github.com/Yu-Fangxu/EACL.
Authors: Rowan Hall Maudslay, Simone Teufel, Francis Bond, James Pustejovsky
Abstract: The senses of a word exhibit rich internal structure. In a typical lexicon, this structure is overlooked: a word's senses are encoded as a list without inter-sense relations. We present ChainNet, a lexical resource which for the first time explicitly identifies these structures. ChainNet expresses how senses in the Open English Wordnet are derived from one another: every nominal sense of a word is either connected to another sense by metaphor or metonymy, or is disconnected in the case of homonymy. Because WordNet senses are linked to resources which capture information about their meaning, ChainNet represents the first dataset of grounded metaphor and metonymy.
Authors: Neema Kotonya, Francesca Toni
Abstract: As deep neural models in NLP become more complex, and as a consequence opaque, the necessity to interpret them becomes greater. A burgeoning interest has emerged in rationalizing explanations to provide short and coherent justifications for predictions. In this position paper, we advocate for a formal framework for key concepts and properties about rationalizing explanations to support their evaluation systematically. We also outline one such formal framework, tailored to rationalizing explanations of increasingly complex structures, from free-form explanations to deductive explanations, to argumentative explanations (with the richest structure). Focusing on the automated fact verification task, we provide illustrations of the use and usefulness of our formalization for evaluating explanations, tailored to their varying structures.
Authors: Jinhyuk Lee, Zhuyun Dai, Xiaoqi Ren, Blair Chen, Daniel Cer, Jeremy R. Cole, Kai Hui, Michael Boratko, Rajvi Kapadia, Wen Ding, Yi Luan, Sai Meher Karthik Duddu, Gustavo Hernandez Abrego, Weiqiang Shi, Nithi Gupta, Aditya Kusupati, Prateek Jain, Siddhartha Reddy Jonnalagadda, Ming-Wei Chang, Iftekhar Naim
Abstract: We present Gecko, a compact and versatile text embedding model. Gecko achieves strong retrieval performance by leveraging a key idea: distilling knowledge from large language models (LLMs) into a retriever. Our two-step distillation process begins with generating diverse, synthetic paired data using an LLM. Next, we further refine the data quality by retrieving a set of candidate passages for each query, and relabeling the positive and hard negative passages using the same LLM. The effectiveness of our approach is demonstrated by the compactness of the Gecko. On the Massive Text Embedding Benchmark (MTEB), Gecko with 256 embedding dimensions outperforms all existing entries with 768 embedding size. Gecko with 768 embedding dimensions achieves an average score of 66.31, competing with 7x larger models and 5x higher dimensional embeddings.
Authors: Joel Ruben Antony Moniz, Soundarya Krishnan, Melis Ozyildirim, Prathamesh Saraf, Halim Cagri Ates, Yuan Zhang, Hong Yu, Nidhi Rajshree
Abstract: Reference resolution is an important problem, one that is essential to understand and successfully handle context of different kinds. This context includes both previous turns and context that pertains to non-conversational entities, such as entities on the user's screen or those running in the background. While LLMs have been shown to be extremely powerful for a variety of tasks, their use in reference resolution, particularly for non-conversational entities, remains underutilized. This paper demonstrates how LLMs can be used to create an extremely effective system to resolve references of various types, by showing how reference resolution can be converted into a language modeling problem, despite involving forms of entities like those on screen that are not traditionally conducive to being reduced to a text-only modality. We demonstrate large improvements over an existing system with similar functionality across different types of references, with our smallest model obtaining absolute gains of over 5% for on-screen references. We also benchmark against GPT-3.5 and GPT-4, with our smallest model achieving performance comparable to that of GPT-4, and our larger models substantially outperforming it.
Authors: Allison Chen, Ilia Sucholutsky, Olga Russakovsky, Thomas L. Griffiths
Abstract: Does language help make sense of the visual world? How important is it to actually see the world rather than having it described with words? These basic questions about the nature of intelligence have been difficult to answer because we only had one example of an intelligent system -- humans -- and limited access to cases that isolated language or vision. However, the development of sophisticated Vision-Language Models (VLMs) by artificial intelligence researchers offers us new opportunities to explore the contributions that language and vision make to learning about the world. We ablate components from the cognitive architecture of these models to identify their contributions to learning new tasks from limited data. We find that a language model leveraging all components recovers a majority of a VLM's performance, despite its lack of visual input, and that language seems to allow this by providing access to prior knowledge and reasoning.
Authors: Tsendsuren Munkhdalai, Youzheng Chen, Khe Chai Sim, Fadi Biadsy, Tara Sainath, Pedro Moreno Mengibar
Abstract: Parameter efficient adaptation methods have become a key mechanism to train large pre-trained models for downstream tasks. However, their per-task parameter overhead is considered still high when the number of downstream tasks to adapt for is large. We introduce an adapter module that has a better efficiency in large scale multi-task adaptation scenario. Our adapter is hierarchical in terms of how the adapter parameters are allocated. The adapter consists of a single shared controller network and multiple task-level adapter heads to reduce the per-task parameter overhead without performance regression on downstream tasks. The adapter is also recurrent so the entire adapter parameters are reused across different layers of the pre-trained model. Our Hierarchical Recurrent Adapter (HRA) outperforms the previous adapter-based approaches as well as full model fine-tuning baseline in both single and multi-task adaptation settings when evaluated on automatic speech recognition tasks.
Authors: Dominic Widdows, Willie Aboumrad, Dohun Kim, Sayonee Ray, Jonathan Mei
Abstract: Language processing is at the heart of current developments in artificial intelligence, and quantum computers are becoming available at the same time. This has led to great interest in quantum natural language processing, and several early proposals and experiments. This paper surveys the state of this area, showing how NLP-related techniques including word embeddings, sequential models, attention, and grammatical parsing have been used in quantum language processing. We introduce a new quantum design for the basic task of text encoding (representing a string of characters in memory), which has not been addressed in detail before. As well as motivating new technologies, quantum theory has made key contributions to the challenging questions of 'What is uncertainty?' and 'What is intelligence?' As these questions are taking on fresh urgency with artificial systems, the paper also considers some of the ways facts are conceptualized and presented in language. In particular, we argue that the problem of 'hallucinations' arises through a basic misunderstanding: language expresses any number of plausible hypotheses, only a few of which become actual, a distinction that is ignored in classical mechanics, but present (albeit confusing) in quantum mechanics.
Authors: Ravi Mangal, Nina Narodytska, Divya Gopinath, Boyue Caroline Hu, Anirban Roy, Susmit Jha, Corina Pasareanu
Abstract: Formal analysis of vision-based deep neural networks (DNNs) is highly desirable but it is very challenging due to the difficulty of expressing formal specifications for vision tasks and the lack of efficient verification procedures. In this paper, we propose to leverage emerging multimodal, vision-language, foundation models (VLMs) as a lens through which we can reason about vision models. VLMs have been trained on a large body of images accompanied by their textual description, and are thus implicitly aware of high-level, human-understandable concepts describing the images. We describe a logical specification language $\texttt{Con}_{\texttt{spec}}$ designed to facilitate writing specifications in terms of these concepts. To define and formally check $\texttt{Con}_{\texttt{spec}}$ specifications, we leverage a VLM, which provides a means to encode and efficiently check natural-language properties of vision models. We demonstrate our techniques on a ResNet-based classifier trained on the RIVAL-10 dataset leveraging CLIP as the multimodal model.
Authors: Jing Wu, Zhixin Lai, Suiyao Chen, Ran Tao, Pan Zhao, Naira Hovakimyan
Abstract: Crop management plays a crucial role in determining crop yield, economic profitability, and environmental sustainability. Despite the availability of management guidelines, optimizing these practices remains a complex and multifaceted challenge. In response, previous studies have explored using reinforcement learning with crop simulators, typically employing simple neural-network-based reinforcement learning (RL) agents. Building on this foundation, this paper introduces a more advanced intelligent crop management system. This system uniquely combines RL, a language model (LM), and crop simulations facilitated by the Decision Support System for Agrotechnology Transfer (DSSAT). We utilize deep RL, specifically a deep Q-network, to train management policies that process numerous state variables from the simulator as observations. A novel aspect of our approach is the conversion of these state variables into more informative language, facilitating the language model's capacity to understand states and explore optimal management practices. The empirical results reveal that the LM exhibits superior learning capabilities. Through simulation experiments with maize crops in Florida (US) and Zaragoza (Spain), the LM not only achieves state-of-the-art performance under various evaluation metrics but also demonstrates a remarkable improvement of over 49\% in economic profit, coupled with reduced environmental impact when compared to baseline methods. Our code is available at \url{https://github.com/jingwu6/LM_AG}.
Authors: Prasanna Vijayaraghavan, Jeffrey Frederic Queisser, Sergio Verduzco Flores, Jun Tani
Abstract: Humans excel at applying learned behavior to unlearned situations. A crucial component of this generalization behavior is our ability to compose/decompose a whole into reusable parts, an attribute known as compositionality. One of the fundamental questions in robotics concerns this characteristic. "How can linguistic compositionality be developed concomitantly with sensorimotor skills through associative learning, particularly when individuals only learn partial linguistic compositions and their corresponding sensorimotor patterns?" To address this question, we propose a brain-inspired neural network model that integrates vision, proprioception, and language into a framework of predictive coding and active inference, based on the free-energy principle. The effectiveness and capabilities of this model were assessed through various simulation experiments conducted with a robot arm. Our results show that generalization in learning to unlearned verb-noun compositions, is significantly enhanced when training variations of task composition are increased. We attribute this to self-organized compositional structures in linguistic latent state space being influenced significantly by sensorimotor learning. Ablation studies show that visual attention and working memory are essential to accurately generate visuo-motor sequences to achieve linguistically represented goals. These insights advance our understanding of mechanisms underlying development of compositionality through interactions of linguistic and sensorimotor experience.
Authors: Tonghui Ren, Yuankai Fan, Zhenying He, Ren Huang, Jiaqi Dai, Can Huang, Yinan Jing, Kai Zhang, Yifan Yang, X. Sean Wang
Abstract: Large Language Model (LLM) techniques play an increasingly important role in Natural Language to SQL (NL2SQL) translation. LLMs trained by extensive corpora have strong natural language understanding and basic SQL generation abilities without additional tuning specific to NL2SQL tasks. Existing LLMs-based NL2SQL approaches try to improve the translation by enhancing the LLMs with an emphasis on user intention understanding. However, LLMs sometimes fail to generate appropriate SQL due to their lack of knowledge in organizing complex logical operator composition. A promising method is to input the LLMs with demonstrations, which include known NL2SQL translations from various databases. LLMs can learn to organize operator compositions from the input demonstrations for the given task. In this paper, we propose PURPLE (Pre-trained models Utilized to Retrieve Prompts for Logical Enhancement), which improves accuracy by retrieving demonstrations containing the requisite logical operator composition for the NL2SQL task on hand, thereby guiding LLMs to produce better SQL translation. PURPLE achieves a new state-of-the-art performance of 80.5% exact-set match accuracy and 87.8% execution match accuracy on the validation set of the popular NL2SQL benchmark Spider. PURPLE maintains high accuracy across diverse benchmarks, budgetary constraints, and various LLMs, showing robustness and cost-effectiveness.
Authors: Shuang Li, Jiahua Wang, Lijie Wen
Abstract: Multi-modal reasoning plays a vital role in bridging the gap between textual and visual information, enabling a deeper understanding of the context. This paper presents the Feature Swapping Multi-modal Reasoning (FSMR) model, designed to enhance multi-modal reasoning through feature swapping. FSMR leverages a pre-trained visual-language model as an encoder, accommodating both text and image inputs for effective feature representation from both modalities. It introduces a unique feature swapping module, enabling the exchange of features between identified objects in images and corresponding vocabulary words in text, thereby enhancing the model's comprehension of the interplay between images and text. To further bolster its multi-modal alignment capabilities, FSMR incorporates a multi-modal cross-attention mechanism, facilitating the joint modeling of textual and visual information. During training, we employ image-text matching and cross-entropy losses to ensure semantic consistency between visual and language elements. Extensive experiments on the PMR dataset demonstrate FSMR's superiority over state-of-the-art baseline models across various performance metrics.
Authors: Tuan Nguyen, Corinne Fredouille, Alain Ghio, Mathieu Balaguer, Virginie Woisard
Abstract: Automatic speech quality assessment has raised more attention as an alternative or support to traditional perceptual clinical evaluation. However, most research so far only gains good results on simple tasks such as binary classification, largely due to data scarcity. To deal with this challenge, current works tend to segment patients' audio files into many samples to augment the datasets. Nevertheless, this approach has limitations, as it indirectly relates overall audio scores to individual segments. This paper introduces a novel approach where the system learns at the audio level instead of segments despite data scarcity. This paper proposes to use the pre-trained Wav2Vec2 architecture for both SSL, and ASR as feature extractor in speech assessment. Carried out on the HNC dataset, our ASR-driven approach established a new baseline compared with other approaches, obtaining average $MSE=0.73$ and $MSE=1.15$ for the prediction of intelligibility and severity scores respectively, using only 95 training samples. It shows that the ASR based Wav2Vec2 model brings the best results and may indicate a strong correlation between ASR and speech quality assessment. We also measure its ability on variable segment durations and speech content, exploring factors influencing its decision.
Authors: Aleksandr V. Petrov, Sean MacAvaney, Craig Macdonald
Abstract: Transformer-based Cross-Encoders achieve state-of-the-art effectiveness in text retrieval. However, Cross-Encoders based on large transformer models (such as BERT or T5) are computationally expensive and allow for scoring only a small number of documents within a reasonably small latency window. However, keeping search latencies low is important for user satisfaction and energy usage. In this paper, we show that weaker shallow transformer models (i.e., transformers with a limited number of layers) actually perform better than full-scale models when constrained to these practical low-latency settings since they can estimate the relevance of more documents in the same time budget. We further show that shallow transformers may benefit from the generalized Binary Cross-Entropy (gBCE) training scheme, which has recently demonstrated success for recommendation tasks. Our experiments with TREC Deep Learning passage ranking query sets demonstrate significant improvements in shallow and full-scale models in low-latency scenarios. For example, when the latency limit is 25ms per query, MonoBERT-Large (a cross-encoder based on a full-scale BERT model) is only able to achieve NDCG@10 of 0.431 on TREC DL 2019, while TinyBERT-gBCE (a cross-encoder based on TinyBERT trained with gBCE) reaches NDCG@10 of 0.652, a +51% gain over MonoBERT-Large. We also show that shallow Cross-Encoders are effective even when used without a GPU (e.g., with CPU inference, NDCG@10 decreases only by 3% compared to GPU inference with 50ms latency), which makes Cross-Encoders practical to run even without specialized hardware acceleration.
Authors: Atsuyuki Miyai, Jingkang Yang, Jingyang Zhang, Yifei Ming, Qing Yu, Go Irie, Yixuan Li, Hai Li, Ziwei Liu, Kiyoharu Aizawa
Abstract: This paper introduces a novel and significant challenge for Vision Language Models (VLMs), termed Unsolvable Problem Detection (UPD). UPD examines the VLM's ability to withhold answers when faced with unsolvable problems in the context of Visual Question Answering (VQA) tasks. UPD encompasses three distinct settings: Absent Answer Detection (AAD), Incompatible Answer Set Detection (IASD), and Incompatible Visual Question Detection (IVQD). To deeply investigate the UPD problem, extensive experiments indicate that most VLMs, including GPT-4V and LLaVA-Next-34B, struggle with our benchmarks to varying extents, highlighting significant room for the improvements. To address UPD, we explore both training-free and training-based solutions, offering new insights into their effectiveness and limitations. We hope our insights, together with future efforts within the proposed UPD settings, will enhance the broader understanding and development of more practical and reliable VLMs.
Authors: Qinghao Ye, Haiyang Xu, Guohai Xu, Jiabo Ye, Ming Yan, Yiyang Zhou, Junyang Wang, Anwen Hu, Pengcheng Shi, Yaya Shi, Chenliang Li, Yuanhong Xu, Hehong Chen, Junfeng Tian, Qi Qian, Ji Zhang, Fei Huang, Jingren Zhou
Abstract: Large language models (LLMs) have demonstrated impressive zero-shot abilities on a variety of open-ended tasks, while recent research has also explored the use of LLMs for multi-modal generation. In this study, we introduce mPLUG-Owl, a novel training paradigm that equips LLMs with multi-modal abilities through modularized learning of foundation LLM, a visual knowledge module, and a visual abstractor module. This approach can support multiple modalities and facilitate diverse unimodal and multimodal abilities through modality collaboration. The training paradigm of mPLUG-Owl involves a two-stage method for aligning image and text, which learns visual knowledge with the assistance of LLM while maintaining and even improving the generation abilities of LLM. In the first stage, the visual knowledge module and abstractor module are trained with a frozen LLM module to align the image and text. In the second stage, language-only and multi-modal supervised datasets are used to jointly fine-tune a low-rank adaption (LoRA) module on LLM and the abstractor module by freezing the visual knowledge module. We carefully build a visually-related instruction evaluation set OwlEval. Experimental results show that our model outperforms existing multi-modal models, demonstrating mPLUG-Owl's impressive instruction and visual understanding ability, multi-turn conversation ability, and knowledge reasoning ability. Besides, we observe some unexpected and exciting abilities such as multi-image correlation and scene text understanding, which makes it possible to leverage it for harder real scenarios, such as vision-only document comprehension. Our code, pre-trained model, instruction-tuned models, and evaluation set are available at https://github.com/X-PLUG/mPLUG-Owl. The online demo is available at https://www.modelscope.cn/studios/damo/mPLUG-Owl.
URLs: https://github.com/X-PLUG/mPLUG-Owl., https://www.modelscope.cn/studios/damo/mPLUG-Owl.
Authors: Ayush Maheshwari, Ashim Gupta, Amrith Krishna, Atul Kumar Singh, Ganesh Ramakrishnan, G. Anil Kumar, Jitin Singla
Abstract: We release S\={a}mayik, a dataset of around 53,000 parallel English-Sanskrit sentences, written in contemporary prose. Sanskrit is a classical language still in sustenance and has a rich documented heritage. However, due to the limited availability of digitized content, it still remains a low-resource language. Existing Sanskrit corpora, whether monolingual or bilingual, have predominantly focused on poetry and offer limited coverage of contemporary written materials. S\={a}mayik is curated from a diverse range of domains, including language instruction material, textual teaching pedagogy, and online tutorials, among others. It stands out as a unique resource that specifically caters to the contemporary usage of Sanskrit, with a primary emphasis on prose writing. Translation models trained on our dataset demonstrate statistically significant improvements when translating out-of-domain contemporary corpora, outperforming models trained on older classical-era poetry datasets. Finally, we also release benchmark models by adapting four multilingual pre-trained models, three of them have not been previously exposed to Sanskrit for translating between English and Sanskrit while one of them is multi-lingual pre-trained translation model including English and Sanskrit. The dataset and source code is present at https://github.com/ayushbits/saamayik.
Authors: Zhaofeng Wu, Linlu Qiu, Alexis Ross, Ekin Aky\"urek, Boyuan Chen, Bailin Wang, Najoung Kim, Jacob Andreas, Yoon Kim
Abstract: The impressive performance of recent language models across a wide range of tasks suggests that they possess a degree of abstract reasoning skills. Are these skills general and transferable, or specialized to specific tasks seen during pretraining? To disentangle these effects, we propose an evaluation framework based on "counterfactual" task variants that deviate from the default assumptions underlying standard tasks. Across a suite of 11 tasks, we observe nontrivial performance on the counterfactual variants, but nevertheless find that performance substantially and consistently degrades compared to the default conditions. This suggests that while current LMs may possess abstract task-solving skills to an extent, they often also rely on narrow, non-transferable procedures for task-solving. These results motivate a more careful interpretation of language model performance that teases apart these aspects of behavior.
Authors: Tianhua Zhang, Jiaxin Ge, Hongyin Luo, Yung-Sung Chuang, Mingye Gao, Yuan Gong, Xixin Wu, Yoon Kim, Helen Meng, James Glass
Abstract: How can we perform computations over natural language representations to solve tasks that require symbolic and numeric reasoning? We propose natural language embedded programs (NLEP) as a unifying framework for addressing math/symbolic reasoning, natural language understanding, and instruction following tasks. Our approach prompts a language model to generate full Python programs that define functions over data structures which contain natural language representations of structured knowledge. A Python interpreter then executes the generated code and prints the output. Despite using a task-general prompt, we find that this approach can improve upon strong baselines across a range of different tasks including math and symbolic reasoning, text classification, question answering, and instruction following. We found that the generated programs are interpretable since they outline the exact reasoning process followed by the program interpreter.
Authors: Lauren Hong, Ting Wang
Abstract: Parameter-efficient fine-tuning (PEFT) enables efficient adaptation of pre-trained language models (PLMs) to specific tasks. By tuning only a minimal set of (extra) parameters, PEFT achieves performance that is comparable to standard fine-tuning. However, despite its prevalent use, the security implications of PEFT remain largely unexplored. In this paper, we take the initial steps and present PETA, a novel trojan attack that compromises the weights of PLMs by accounting for downstream adaptation through bilevel optimization: the upper-level objective embeds the backdoor into a model while the lower-level objective simulates PEFT to both retain the PLM's task-specific performance and ensure that the backdoor persists after fine-tuning. With extensive evaluation across a variety of downstream tasks and trigger designs, we demonstrate PETA's effectiveness in terms of both attack success rate and clean accuracy, even when the attacker does not have full knowledge of the victim user's training process.
Authors: Shengnan An, Zexiong Ma, Zeqi Lin, Nanning Zheng, Jian-Guang Lou, Weizhu Chen
Abstract: Large language models (LLMs) recently exhibited remarkable reasoning capabilities on solving math problems. To further improve their reasoning capabilities, this work explores whether LLMs can LEarn from MistAkes (LEMA), akin to the human learning process. Consider a human student who failed to solve a math problem, he will learn from what mistake he has made and how to correct it. Mimicking this error-driven learning process, LEMA incorporates mistake-correction data pairs during fine-tuning LLMs. Specifically, we first collect inaccurate reasoning paths from various LLMs, and then employ GPT-4 as a ''corrector'' to identify the mistake step, explain the reason for the mistake, correct the mistake and generate the final answer. In addition, we apply a correction-centric evolution strategy that effectively expands the question set for generating correction data. Experiments across various LLMs and reasoning tasks show that LEMA effectively improves CoT-alone fine-tuning. Our further ablations shed light on the non-homogeneous effectiveness between CoT data and correction data. These results suggest a significant potential for LLMs to improve through learning from their mistakes. Our code, models and prompts are publicly available at https://github.com/microsoft/LEMA.
Authors: Jiao Ou, Junda Lu, Che Liu, Yihong Tang, Fuzheng Zhang, Di Zhang, Kun Gai
Abstract: Large language models (LLMs) have achieved remarkable breakthroughs in new dialogue capabilities by leveraging instruction tuning, which refreshes human impressions of dialogue systems. The long-standing goal of dialogue systems is to be human-like enough to establish long-term connections with users. Therefore, there has been an urgent need to evaluate LLMs as human-like dialogue systems. In this paper, we propose DialogBench, a dialogue evaluation benchmark that contains 12 dialogue tasks to probe the capabilities of LLMs as human-like dialogue systems should have. Specifically, we prompt GPT-4 to generate evaluation instances for each task. We first design the basic prompt based on widely used design principles and further mitigate the existing biases to generate higher-quality evaluation instances. Our extensive tests on English and Chinese DialogBench of 26 LLMs show that instruction tuning improves the human likeness of LLMs to a certain extent, but most LLMs still have much room for improvement as human-like dialogue systems. Interestingly, results also show that the positioning of assistant AI can make instruction tuning weaken the human emotional perception of LLMs and their mastery of information about human daily life.
Authors: HyunJin Kim, Young Jin Kim, JinYeong Bak
Abstract: Pre-trained language models (PLMs) show impressive performance in various downstream NLP tasks. However, pre-training large language models demands substantial memory and training compute. Furthermore, due to the substantial resources required, many PLM weights are confidential. Consequently, users are compelled to share their data with model owners for fine-tuning specific tasks. To overcome the limitations, we introduce Plug-in External Memory Adaptation (PEMA), a Parameter-Efficient Fine-Tuning (PEFT) method, enabling PLM fine-tuning without requiring access to all the weights. PEMA integrates with context representations from test data during inference to perform downstream tasks. It uses external memory to store PLM-generated context representations mapped with target tokens. Our method utilizes weight matrices of LoRA-like bottlenecked adapter in the PLM's final layer to enhance efficiency. Our approach also includes Gradual Unrolling, a novel interpolation strategy to improve generation quality. We validate PEMA's effectiveness through experiments on syntactic and real datasets for machine translation and style transfer. Our findings show that PEMA outperforms other PEFT approaches in memory and latency efficiency for training, and also excels in maintaining sentence meaning and generating appropriate language and styles.
Authors: Lihui Liu, Blaine Hill, Boxin Du, Fei Wang, Hanghang Tong
Abstract: Conversational question answering (convQA) over knowledge graphs (KGs) involves answering multi-turn natural language questions about information contained in a KG. State-of-the-art methods of ConvQA often struggle with inexplicit question-answer pairs. These inputs are easy for human beings to understand given a conversation history, but hard for a machine to interpret, which can degrade ConvQA performance. To address this problem, we propose a reinforcement learning (RL) based model, CornNet, which utilizes question reformulations generated by large language models (LLMs) to improve ConvQA performance. CornNet adopts a teacher-student architecture where a teacher model learns question representations using human writing reformulations, and a student model to mimic the teacher model's output via reformulations generated by LLMs. The learned question representation is then used by an RL model to locate the correct answer in a KG. Extensive experimental results show that CornNet outperforms state-of-the-art convQA models.
Authors: Aditya Joshi, Raj Dabre, Diptesh Kanojia, Zhuang Li, Haolan Zhan, Gholamreza Haffari, Doris Dippold
Abstract: State-of-the-art natural language processing (NLP) models are trained on massive training corpora, and report a superlative performance on evaluation datasets. This survey delves into an important attribute of these datasets: the dialect of a language. Motivated by the performance degradation of NLP models for dialectic datasets and its implications for the equity of language technologies, we survey past research in NLP for dialects in terms of datasets, and approaches. We describe a wide range of NLP tasks in terms of two categories: natural language understanding (NLU) (for tasks such as dialect classification, sentiment analysis, parsing, and NLU benchmarks) and natural language generation (NLG) (for summarisation, machine translation, and dialogue systems). The survey is also broad in its coverage of languages which include English, Arabic, German among others. We observe that past work in NLP concerning dialects goes deeper than mere dialect classification, and . This includes early approaches that used sentence transduction that lead to the recent approaches that integrate hypernetworks into LoRA. We expect that this survey will be useful to NLP researchers interested in building equitable language technologies by rethinking LLM benchmarks and model architectures.
Authors: Manuel Faysse, Patrick Fernandes, Nuno M. Guerreiro, Ant\'onio Loison, Duarte M. Alves, Caio Corro, Nicolas Boizard, Jo\~ao Alves, Ricardo Rei, Pedro H. Martins, Antoni Bigata Casademunt, Fran\c{c}ois Yvon, Andr\'e F. T. Martins, Gautier Viaud, C\'eline Hudelot, Pierre Colombo
Abstract: We introduce CroissantLLM, a 1.3B language model pretrained on a set of 3T English and French tokens, to bring to the research and industrial community a high-performance, fully open-sourced bilingual model that runs swiftly on consumer-grade local hardware. To that end, we pioneer the approach of training an intrinsically bilingual model with a 1:1 English-to-French pretraining data ratio, a custom tokenizer, and bilingual finetuning datasets. We release the training dataset, notably containing a French split with manually curated, high-quality, and varied data sources. To assess performance outside of English, we craft a novel benchmark, FrenchBench, consisting of an array of classification and generation tasks, covering various orthogonal aspects of model performance in the French Language. Additionally, rooted in transparency and to foster further Large Language Model research, we release codebases, and dozens of checkpoints across various model sizes, training data distributions, and training steps, as well as fine-tuned Chat models, and strong translation models. We evaluate our model through the FMTI framework, and validate 81 % of the transparency criteria, far beyond the scores of even most open initiatives. This work enriches the NLP landscape, breaking away from previous English-centric work in order to strengthen our understanding of multilinguality in language models.
Authors: Chen Ling, Xujiang Zhao, Xuchao Zhang, Wei Cheng, Yanchi Liu, Yiyou Sun, Mika Oishi, Takao Osaki, Katsushi Matsuda, Jie Ji, Guangji Bai, Liang Zhao, Haifeng Chen
Abstract: In-context learning has emerged as a groundbreaking ability of Large Language Models (LLMs) and revolutionized various fields by providing a few task-relevant demonstrations in the prompt. However, trustworthy issues with LLM's response, such as hallucination, have also been actively discussed. Existing works have been devoted to quantifying the uncertainty in LLM's response, but they often overlook the complex nature of LLMs and the uniqueness of in-context learning. In this work, we delve into the predictive uncertainty of LLMs associated with in-context learning, highlighting that such uncertainties may stem from both the provided demonstrations (aleatoric uncertainty) and ambiguities tied to the model's configurations (epistemic uncertainty). We propose a novel formulation and corresponding estimation method to quantify both types of uncertainties. The proposed method offers an unsupervised way to understand the prediction of in-context learning in a plug-and-play fashion. Extensive experiments are conducted to demonstrate the effectiveness of the decomposition. The code and data are available at: https://github.com/lingchen0331/UQ_ICL.
Authors: Jaylen Jones, Lingbo Mo, Eric Fosler-Lussier, Huan Sun
Abstract: Counter narratives - informed responses to hate speech contexts designed to refute hateful claims and de-escalate encounters - have emerged as an effective hate speech intervention strategy. While previous work has proposed automatic counter narrative generation methods to aid manual interventions, the evaluation of these approaches remains underdeveloped. Previous automatic metrics for counter narrative evaluation lack alignment with human judgment as they rely on superficial reference comparisons instead of incorporating key aspects of counter narrative quality as evaluation criteria. To address prior evaluation limitations, we propose a novel evaluation framework prompting LLMs to provide scores and feedback for generated counter narrative candidates using 5 defined aspects derived from guidelines from counter narrative specialized NGOs. We found that LLM evaluators achieve strong alignment to human-annotated scores and feedback and outperform alternative metrics, indicating their potential as multi-aspect, reference-free and interpretable evaluators for counter narrative evaluation.
Authors: Antonio San Mart\'in
Abstract: This paper examines the impact of Generative Artificial Intelligence (GenAI) tools like ChatGPT on the creation and consumption of terminological definitions. From the terminologist's point of view, the strategic use of GenAI tools can streamline the process of crafting definitions, reducing both time and effort, while potentially enhancing quality. GenAI tools enable AI-assisted terminography, notably post-editing terminography, where the machine produces a definition that the terminologist then corrects or refines. However, the potential of GenAI tools to fulfill all the terminological needs of a user, including term definitions, challenges the very existence of terminological definitions and resources as we know them. Unlike terminological definitions, GenAI tools can describe the knowledge activated by a term in a specific context. However, a main drawback of these tools is that their output can contain errors. For this reason, users requiring reliability will likely still resort to terminological resources for definitions. Nevertheless, with the inevitable integration of AI into terminology work, the distinction between human-created and AI-created content will become increasingly blurred.
Authors: Zihan Wang, Peiyi Wang, Houfeng Wang
Abstract: Hierarchical text classification (HTC) is a challenging subtask of multi-label classification due to its complex taxonomic structure. Nearly all recent HTC works focus on how the labels are structured but ignore the sub-structure of ground-truth labels according to each input text which contains fruitful label co-occurrence information. In this work, we introduce this local hierarchy with an adversarial framework. We propose a HiAdv framework that can fit in nearly all HTC models and optimize them with the local hierarchy as auxiliary information. We test on two typical HTC models and find that HiAdv is effective in all scenarios and is adept at dealing with complex taxonomic hierarchies. Further experiments demonstrate that the promotion of our framework indeed comes from the local hierarchy and the local hierarchy is beneficial for rare classes which have insufficient training data.
Authors: Timothee Mickus, Elaine Zosa, Ra\'ul V\'azquez, Teemu Vahtola, J\"org Tiedemann, Vincent Segonne, Alessandro Raganato, Marianna Apidianaki
Abstract: This paper presents the results of the SHROOM, a shared task focused on detecting hallucinations: outputs from natural language generation (NLG) systems that are fluent, yet inaccurate. Such cases of overgeneration put in jeopardy many NLG applications, where correctness is often mission-critical. The shared task was conducted with a newly constructed dataset of 4000 model outputs labeled by 5 annotators each, spanning 3 NLP tasks: machine translation, paraphrase generation and definition modeling. The shared task was tackled by a total of 58 different users grouped in 42 teams, out of which 27 elected to write a system description paper; collectively, they submitted over 300 prediction sets on both tracks of the shared task. We observe a number of key trends in how this approach was tackled -- many participants rely on a handful of model, and often rely either on synthetic data for fine-tuning or zero-shot prompting strategies. While a majority of the teams did outperform our proposed baseline system, the performances of top-scoring systems are still consistent with a random handling of the more challenging items.
Authors: Joonwon Jang, Sanghwan Jang, Wonbin Kweon, Minjin Jeon, Hwanjo Yu
Abstract: Large language models (LLMs) are able to solve various tasks with only a few demonstrations utilizing their in-context learning (ICL) abilities. However, LLMs often rely on their pre-trained semantic priors of demonstrations rather than on the input-label relationships to proceed with ICL prediction. In this work, we term this phenomenon as the 'Demonstration Shortcut'. While previous works have primarily focused on improving ICL prediction results for predefined tasks, we aim to rectify the Demonstration Shortcut, thereby enabling the LLM to effectively learn new input-label relationships from demonstrations. To achieve this, we introduce In-Context Calibration, a demonstration-aware calibration method. We evaluate the effectiveness of the proposed method in two settings: (1) the Original ICL Task using the standard label space and (2) the Task Learning setting, where the label space is replaced with semantically unrelated tokens. In both settings, In-Context Calibration demonstrates substantial improvements, with results generalized across three LLM families (OPT, GPT, and Llama2) under various configurations.
Authors: Zhishuai Li, Xiang Wang, Jingjing Zhao, Sun Yang, Guoqing Du, Xiaoru Hu, Bin Zhang, Yuxiao Ye, Ziyue Li, Rui Zhao, Hangyu Mao
Abstract: Recent advancements in Text-to-SQL (Text2SQL) emphasize stimulating the large language models (LLM) on in-context learning, achieving significant results. Nevertheless, they face challenges when dealing with verbose database information and complex user intentions. This paper presents a two-stage framework to enhance the performance of current LLM-based natural language to SQL systems. We first introduce a novel prompt representation, called reference-enhanced representation, which includes schema information and randomly sampled cell values from tables to instruct LLMs in generating SQL queries. Then, in the first stage, question-SQL pairs are retrieved as few-shot demonstrations, prompting the LLM to generate a preliminary SQL (PreSQL). After that, the mentioned entities in PreSQL are parsed to conduct schema linking, which can significantly compact the useful information. In the second stage, with the linked schema, we simplify the prompt's schema information and instruct the LLM to produce the final SQL. Finally, as the post-refinement module, we propose using cross-consistency across different LLMs rather than self-consistency within a particular LLM. Our methods achieve new SOTA results on the Spider benchmark, with an execution accuracy of 87.6%.
Authors: Mahdi Rezapour
Abstract: In this study, we explore the application of transformer-based models for emotion classification on text data. We train and evaluate several pre-trained transformer models, on the Emotion dataset using different variants of transformers. The paper also analyzes some factors that in-fluence the performance of the model, such as the fine-tuning of the transformer layer, the trainability of the layer, and the preprocessing of the text data. Our analysis reveals that commonly applied techniques like removing punctuation and stop words can hinder model performance. This might be because transformers strength lies in understanding contextual relationships within text. Elements like punctuation and stop words can still convey sentiment or emphasis and removing them might disrupt this context.
Authors: Chinmaya Andukuri, Jan-Philipp Fr\"anken, Tobias Gerstenberg, Noah D. Goodman
Abstract: When prompting language models to complete a task, users often leave important aspects unsaid. While asking questions could resolve this ambiguity (GATE; Li et al., 2023), models often struggle to ask good questions. We explore a language model's ability to self-improve (STaR; Zelikman et al., 2022) by rewarding the model for generating useful questions-a simple method we dub STaR-GATE. We generate a synthetic dataset of 25,500 unique persona-task prompts to simulate conversations between a pretrained language model-the Questioner-and a Roleplayer whose preferences are unknown to the Questioner. By asking questions, the Questioner elicits preferences from the Roleplayer. The Questioner is iteratively finetuned on questions that increase the probability of high-quality responses to the task, which are generated by an Oracle with access to the Roleplayer's latent preferences. After two iterations of self-improvement, the Questioner asks better questions, allowing it to generate responses that are preferred over responses from the initial model on 72% of tasks. Our results indicate that teaching a language model to ask better questions leads to better personalized responses.
Authors: Song Wang, Yiliang Zhou, Ziqiang Han, Cui Tao, Yunyu Xiao, Ying Ding, Joydeep Ghosh, Yifan Peng
Abstract: Data accuracy is essential for scientific research and policy development. The National Violent Death Reporting System (NVDRS) data is widely used for discovering the patterns and causes of death. Recent studies suggested the annotation inconsistencies within the NVDRS and the potential impact on erroneous suicide-cause attributions. We present an empirical Natural Language Processing (NLP) approach to detect annotation inconsistencies and adopt a cross-validation-like paradigm to identify problematic instances. We analyzed 267,804 suicide death incidents between 2003 and 2020 from the NVDRS. Our results showed that incorporating the target state's data into training the suicide-crisis classifier brought an increase of 5.4% to the F-1 score on the target state's test set and a decrease of 1.1% on other states' test set. To conclude, we demonstrated the annotation inconsistencies in NVDRS's death investigation notes, identified problematic instances, evaluated the effectiveness of correcting problematic instances, and eventually proposed an NLP improvement solution.
Authors: Ruijie Wang, Luca Rossetto, Michael Cochez, Abraham Bernstein
Abstract: Multi-relation question answering (QA) is a challenging task, where given questions usually require long reasoning chains in KGs that consist of multiple relations. Recently, methods with explicit multi-step reasoning over KGs have been prominently used in this task and have demonstrated promising performance. Examples include methods that perform stepwise label propagation through KG triples and methods that navigate over KG triples based on reinforcement learning. A main weakness of these methods is that their reasoning mechanisms are usually complex and difficult to implement or train. In this paper, we argue that multi-relation QA can be achieved via end-to-end single-step implicit reasoning, which is simpler, more efficient, and easier to adopt. We propose QAGCN -- a Question-Aware Graph Convolutional Network (GCN)-based method that includes a novel GCN architecture with controlled question-dependent message propagation for the implicit reasoning. Extensive experiments have been conducted, where QAGCN achieved competitive and even superior performance compared to state-of-the-art explicit-reasoning methods. Our code and pre-trained models are available in the repository: https://github.com/ruijie-wang-uzh/QAGCN
Authors: Young-Jun Lee, Byungsoo Ko, Han-Gyu Kim, Jonghwan Hyeon, Ho-Jin Choi
Abstract: As sharing images in an instant message is a crucial factor, there has been active research on learning an image-text multi-modal dialogue models. However, training a well-generalized multi-modal dialogue model remains challenging due to the low quality and limited diversity of images per dialogue in existing multi-modal dialogue datasets. In this paper, we propose an automated pipeline to construct a multi-modal dialogue dataset, ensuring both dialogue quality and image diversity without requiring minimum human effort. In our pipeline, to guarantee the coherence between images and dialogue, we prompt GPT-4 to infer potential image-sharing moments - specifically, the utterance, speaker, rationale, and image description. Furthermore, we leverage CLIP similarity to maintain consistency between aligned multiple images to the utterance. Through this pipeline, we introduce DialogCC, a high-quality and diverse multi-modal dialogue dataset that surpasses existing datasets in terms of quality and diversity in human evaluation. Our comprehensive experiments highlight that when multi-modal dialogue models are trained using our dataset, their generalization performance on unseen dialogue datasets is significantly enhanced. We make our source code and dataset publicly available.
Authors: Stephen Choi, William Gazeley, Siu Ho Wong, Tingting Li
Abstract: With the exponential growth in large language models (LLMs), leveraging their emergent properties for specialized domains like finance merits exploration. However, regulated fields such as finance pose unique constraints, requiring domain-optimized frameworks. We present ConFIRM, an LLM-based conversational financial information retrieval model tailored for query intent classification and knowledge base labeling. ConFIRM comprises two modules: 1) a method to synthesize finance domain-specific question-answer pairs, and 2) evaluation of parameter efficient fine-tuning approaches for the query classification task. We generate a dataset of over 4000 samples, assessing accuracy on a separate test set. ConFIRM achieved over 90% accuracy, essential for regulatory compliance. ConFIRM provides a data-efficient solution to extract precise query intent for financial dialog systems.
Authors: Chancharik Mitra, Brandon Huang, Trevor Darrell, Roei Herzig
Abstract: The combination of strong visual backbones and Large Language Model (LLM) reasoning has led to Large Multimodal Models (LMMs) becoming the current standard for a wide range of vision and language (VL) tasks. However, recent research has shown that even the most advanced LMMs still struggle to capture aspects of compositional visual reasoning, such as attributes and relationships between objects. One solution is to utilize scene graphs (SGs)--a formalization of objects and their relations and attributes that has been extensively used as a bridge between the visual and textual domains. Yet, scene graph data requires scene graph annotations, which are expensive to collect and thus not easily scalable. Moreover, finetuning an LMM based on SG data can lead to catastrophic forgetting of the pretraining objective. To overcome this, inspired by chain-of-thought methods, we propose Compositional Chain-of-Thought (CCoT), a novel zero-shot Chain-of-Thought prompting method that utilizes SG representations in order to extract compositional knowledge from an LMM. Specifically, we first generate an SG using the LMM, and then use that SG in the prompt to produce a response. Through extensive experiments, we find that the proposed CCoT approach not only improves LMM performance on several vision and language VL compositional benchmarks but also improves the performance of several popular LMMs on general multimodal benchmarks, without the need for fine-tuning or annotated ground-truth SGs. Code: https://github.com/chancharikmitra/CCoT
Authors: Mohammad Reza Taesiri, Tianjun Feng, Anh Nguyen, Cor-Paul Bezemer
Abstract: Large multimodal models (LMMs) have evolved from large language models (LLMs) to integrate multiple input modalities, such as visual inputs. This integration augments the capacity of LLMs for tasks requiring visual comprehension and reasoning. However, the extent and limitations of their enhanced abilities are not fully understood, especially when it comes to real-world tasks. To address this gap, we introduce GlitchBench, a novel benchmark derived from video game quality assurance tasks, to test and evaluate the reasoning capabilities of LMMs. Our benchmark is curated from a variety of unusual and glitched scenarios from video games and aims to challenge both the visual and linguistic reasoning powers of LMMs in detecting and interpreting out-of-the-ordinary events. We evaluate multiple state-of-the-art LMMs, and we show that GlitchBench presents a new challenge for these models. Code and data are available at: https://glitchbench.github.io/
Authors: Shuang Wu, Liwen Zhu, Tao Yang, Shiwei Xu, Qiang Fu, Yang Wei, Haobo Fu
Abstract: This paper presents an innovative framework that integrates Large Language Models (LLMs) with an external Thinker module to enhance the reasoning capabilities of LLM-based agents. Unlike augmenting LLMs with prompt engineering, Thinker directly harnesses knowledge from databases and employs various optimization techniques. The framework forms a reasoning hierarchy where LLMs handle intuitive System-1 tasks such as natural language processing, while the Thinker focuses on cognitive System-2 tasks that require complex logical analysis and domain-specific knowledge. Our framework is presented using a 9-player Werewolf game that demands dual-system reasoning. We introduce a communication protocol between LLMs and the Thinker, and train the Thinker using data from 18800 human sessions and reinforcement learning. Experiments demonstrate the framework's effectiveness in deductive reasoning, speech generation, and online game evaluation. Additionally, we fine-tune a 6B LLM to surpass GPT4 when integrated with the Thinker. This paper also contributes the largest dataset for social deduction games to date.
Authors: Shuai Zhao, Leilei Gan, Luu Anh Tuan, Jie Fu, Lingjuan Lyu, Meihuizi Jia, Jinming Wen
Abstract: Recently, various parameter-efficient fine-tuning (PEFT) strategies for application to language models have been proposed and successfully implemented. However, this raises the question of whether PEFT, which only updates a limited set of model parameters, constitutes security vulnerabilities when confronted with weight-poisoning backdoor attacks. In this study, we show that PEFT is more susceptible to weight-poisoning backdoor attacks compared to the full-parameter fine-tuning method, with pre-defined triggers remaining exploitable and pre-defined targets maintaining high confidence, even after fine-tuning. Motivated by this insight, we developed a Poisoned Sample Identification Module (PSIM) leveraging PEFT, which identifies poisoned samples through confidence, providing robust defense against weight-poisoning backdoor attacks. Specifically, we leverage PEFT to train the PSIM with randomly reset sample labels. During the inference process, extreme confidence serves as an indicator for poisoned samples, while others are clean. We conduct experiments on text classification tasks, five fine-tuning strategies, and three weight-poisoning backdoor attack methods. Experiments show near 100% success rates for weight-poisoning backdoor attacks when utilizing PEFT. Furthermore, our defensive approach exhibits overall competitive performance in mitigating weight-poisoning backdoor attacks.
Authors: Seil Kang, Donghyun Kim, Junhyeok Kim, Hyo Kyung Lee, Seong Jae Hwang
Abstract: Significant methodological strides have been made toward Chest X-ray (CXR) understanding via modern vision-language models (VLMs), demonstrating impressive Visual Question Answering (VQA) and CXR report generation abilities. However, existing CXR understanding frameworks still possess several procedural caveats. (1) Previous methods solely use CXR reports, which are insufficient for comprehensive Visual Question Answering (VQA), especially when additional health-related data like medication history and prior diagnoses are needed. (2) Previous methods use raw CXR reports, which are often arbitrarily structured. While modern language models can understand various text formats, restructuring reports for clearer, organized anatomy-based information could enhance their usefulness. (3) Current evaluation methods for CXR-VQA primarily emphasize linguistic correctness, lacking the capability to offer nuanced assessments of the generated answers. In this work, to address the aforementioned caveats, we introduce WoLF, a Wide-scope Large Language Model Framework for CXR understanding. To resolve (1), we capture multi-faceted records of patients, which are utilized for accurate diagnoses in real-world clinical scenarios. Specifically, we adopt the Electronic Health Records (EHR) to generate instruction-following data suited for CXR understanding. Regarding (2), we enhance report generation performance by decoupling knowledge in CXR reports based on anatomical structure even within the attention step via masked attention. To address (3), we introduce an AI-evaluation protocol optimized for assessing the capabilities of LLM. Through extensive experimental validations, WoLF demonstrates superior performance over other models on MIMIC-CXR in the AI-evaluation arena about VQA (up to +9.47%p mean score) and by metrics about report generation (+7.3%p BLEU-1).
Authors: Zhixin Lai, Jing Wu, Suiyao Chen, Yucheng Zhou, Naira Hovakimyan
Abstract: In this study, we uncover the unexpected efficacy of residual-based large language models (LLMs) as part of encoders for biomedical imaging tasks, a domain traditionally devoid of language or textual data. The approach diverges from established methodologies by utilizing a frozen transformer block, extracted from pre-trained LLMs, as an innovative encoder layer for the direct processing of visual tokens. This strategy represents a significant departure from the standard multi-modal vision-language frameworks, which typically hinge on language-driven prompts and inputs. We found that these LLMs could boost performance across a spectrum of biomedical imaging applications, including both 2D and 3D visual classification tasks, serving as plug-and-play boosters. More interestingly, as a byproduct, we found that the proposed framework achieved superior performance, setting new state-of-the-art results on extensive, standardized datasets in MedMNIST-2D and 3D. Through this work, we aim to open new avenues for employing LLMs in biomedical imaging and enriching the understanding of their potential in this specialized domain.