Abstract
Changing speaker names consistently throughout a dialogue should not affect its meaning and corresponding outputs for text generation from dialogues. However, pre-trained language models, serving as the backbone for dialogue-processing tasks, have shown to be sensitive to nuances. This may result in unfairness in real-world applications. No comprehensive analysis of this problem has been done in the past. In this work, we propose to quantitatively measure a model's sensitivity on speaker names, and comprehensively evaluate a number of known methods for reducing speaker name sensitivity, including a novel approach of our own. Extensive experiments on multiple datasets provide a benchmark for this problem and show the favorable performance of our approach in sensitivity reduction and quality of generation.
INSTRUCTSCORE: Towards Explainable Text Generation Evaluation with Automatic Feedback
Authors: Wenda Xu, Danqing Wang, Liangming Pan, Zhenqiao Song, Markus Freitag, William Yang Wang, Lei Li
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
The field of automatic evaluation of text generation made tremendous progress in the last few years. In particular, since the advent of neural metrics, like COMET, BLEURT, and SEScore2, the newest generation of metrics show a high correlation with human judgment. Unfortunately, quality scores generated with neural metrics are not interpretable, and it is unclear which part of the generation output is criticized by the metrics. To address this limitation, we present INSTRUCTSCORE, an open-source, explainable evaluation metric for text generation. By harnessing both explicit human instruction and the implicit knowledge of GPT4, we fine-tune a LLAMA model to create an evaluative metric that can produce a diagnostic report aligned with human judgment. We evaluate INSTRUCTSCORE on the WMT22 Zh-En translation task, where our 7B model surpasses other LLM-based baselines, including those based on 175B GPT3. Impressively, our INSTRUCTSCORE, even without direct supervision from human-rated data, achieves performance levels on par with state-of-the-art metrics like COMET22, which was fine-tuned on human ratings.
QTSumm: A New Benchmark for Query-Focused Table Summarization
Abstract
People primarily consult tables to conduct data analysis or answer specific questions. Text generation systems that can provide accurate table summaries tailored to users' information needs can facilitate more efficient access to relevant data insights. However, existing table-to-text generation studies primarily focus on converting tabular data into coherent statements, rather than addressing information-seeking purposes. In this paper, we define a new query-focused table summarization task, where text generation models have to perform human-like reasoning and analysis over the given table to generate a tailored summary, and we introduce a new benchmark named QTSumm for this task. QTSumm consists of 5,625 human-annotated query-summary pairs over 2,437 tables on diverse topics. Moreover, we investigate state-of-the-art models (i.e., text generation, table-to-text generation, and large language models) on the QTSumm dataset. Experimental results and manual analysis reveal that our benchmark presents significant challenges in table-to-text generation for future research.
APPLS: A Meta-evaluation Testbed for Plain Language Summarization
Authors: Yue Guo, Tal August, Gondy Leroy, Trevor Cohen, Lucy Lu Wang
Abstract
While there has been significant development of models for Plain Language Summarization (PLS), evaluation remains a challenge. This is in part because PLS involves multiple, interrelated language transformations (e.g., adding background explanations, removing specialized terminology). No metrics are explicitly engineered for PLS, and the suitability of other text generation evaluation metrics remains unclear. To address these concerns, our study presents a granular meta-evaluation testbed, APPLS, designed to evaluate existing metrics for PLS. Drawing on insights from previous research, we define controlled perturbations for our testbed along four criteria that a metric of plain language should capture: informativeness, simplification, coherence, and faithfulness. Our analysis of metrics using this testbed reveals that current metrics fail to capture simplification, signaling a crucial gap. In response, we introduce POMME, a novel metric designed to assess text simplification in PLS. We demonstrate its correlation with simplification perturbations and validate across a variety of datasets. Our research contributes the first meta-evaluation testbed for PLS and a comprehensive evaluation of existing metrics, offering insights with relevance to other text generation tasks.
Keyword: machine translation
GATology for Linguistics: What Syntactic Dependencies It Knows
Abstract
Graph Attention Network (GAT) is a graph neural network which is one of the strategies for modeling and representing explicit syntactic knowledge and can work with pre-trained models, such as BERT, in downstream tasks. Currently, there is still a lack of investigation into how GAT learns syntactic knowledge from the perspective of model structure. As one of the strategies for modeling explicit syntactic knowledge, GAT and BERT have never been applied and discussed in Machine Translation (MT) scenarios. We design a dependency relation prediction task to study how GAT learns syntactic knowledge of three languages as a function of the number of attention heads and layers. We also use a paired t-test and F1-score to clarify the differences in syntactic dependency prediction between GAT and BERT fine-tuned by the MT task (MT-B). The experiments show that better performance can be achieved by appropriately increasing the number of attention heads with two GAT layers. With more than two layers, learning suffers. Moreover, GAT is more competitive in training speed and syntactic dependency prediction than MT-B, which may reveal a better incorporation of modeling explicit syntactic knowledge and the possibility of combining GAT and BERT in the MT tasks.
Syntactic Knowledge via Graph Attention with BERT in Machine Translation
Abstract
Although the Transformer model can effectively acquire context features via a self-attention mechanism, deeper syntactic knowledge is still not effectively modeled. To alleviate the above problem, we propose Syntactic knowledge via Graph attention with BERT (SGB) in Machine Translation (MT) scenarios. Graph Attention Network (GAT) and BERT jointly represent syntactic dependency feature as explicit knowledge of the source language to enrich source language representations and guide target language generation. Our experiments use gold syntax-annotation sentences and Quality Estimation (QE) model to obtain interpretability of translation quality improvement regarding syntactic knowledge without being limited to a BLEU score. Experiments show that the proposed SGB engines improve translation quality across the three MT tasks without sacrificing BLEU scores. We investigate what length of source sentences benefits the most and what dependencies are better identified by the SGB engines. We also find that learning of specific dependency relations by GAT can be reflected in the translation quality containing such relations and that syntax on the graph leads to new modeling of syntactic aspects of source sentences in the middle and bottom layers of BERT.
Neural Machine Translation for Code Generation
Authors: Dharma KC, Clayton T. Morrison
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
Neural machine translation (NMT) methods developed for natural language processing have been shown to be highly successful in automating translation from one natural language to another. Recently, these NMT methods have been adapted to the generation of program code. In NMT for code generation, the task is to generate output source code that satisfies constraints expressed in the input. In the literature, a variety of different input scenarios have been explored, including generating code based on natural language description, lower-level representations such as binary or assembly (neural decompilation), partial representations of source code (code completion and repair), and source code in another language (code translation). In this paper we survey the NMT for code generation literature, cataloging the variety of methods that have been explored according to input and output representations, model architectures, optimization techniques used, data sets, and evaluation methods. We discuss the limitations of existing methods and future research directions
Non-parametric, Nearest-neighbor-assisted Fine-tuning for Neural Machine Translation
Abstract
Non-parametric, k-nearest-neighbor algorithms have recently made inroads to assist generative models such as language models and machine translation decoders. We explore whether such non-parametric models can improve machine translation models at the fine-tuning stage by incorporating statistics from the kNN predictions to inform the gradient updates for a baseline translation model. There are multiple methods which could be used to incorporate kNN statistics and we investigate gradient scaling by a gating mechanism, the kNN's ground truth probability, and reinforcement learning. For four standard in-domain machine translation datasets, compared with classic fine-tuning, we report consistent improvements of all of the three methods by as much as 1.45 BLEU and 1.28 BLEU for German-English and English-German translations respectively. Through qualitative analysis, we found particular improvements when it comes to translating grammatical relations or function words, which results in increased fluency of our model.
Optimizing Non-Autoregressive Transformers with Contrastive Learning
Authors: Chenxin An, Jiangtao Feng, Fei Huang, xipeng Qiu
Abstract
Non-autoregressive Transformers (NATs) reduce the inference latency of Autoregressive Transformers (ATs) by predicting words all at once rather than in sequential order. They have achieved remarkable progress in machine translation as well as many other applications. However, a long-standing challenge for NATs is the learning of multi-modality data distribution, which is the main cause of the performance gap between NATs and ATs. In this paper, we propose to ease the difficulty of modality learning via sampling from the model distribution instead of the data distribution. We derive contrastive constraints to stabilize the training process and integrate this resulting objective with the state-of-the-art NAT architecture DA-Transformer. Our model \method is examined on 3 different tasks, including machine translation, text summarization, and paraphrasing with 5 benchmarks. Results show that our approach outperforms previous non-autoregressive baselines by a significant margin and establishes new state-of-the-art results for non-autoregressive transformers on all the benchmarks.
TeCS: A Dataset and Benchmark for Tense Consistency of Machine Translation
Abstract
Tense inconsistency frequently occurs in machine translation. However, there are few criteria to assess the model's mastery of tense prediction from a linguistic perspective. In this paper, we present a parallel tense test set, containing French-English 552 utterances. We also introduce a corresponding benchmark, tense prediction accuracy. With the tense test set and the benchmark, researchers are able to measure the tense consistency performance of machine translation systems for the first time.
Challenges in Context-Aware Neural Machine Translation
Authors: Linghao Jin, Jacqueline He, Jonathan May, Xuezhe Ma
Abstract
Context-aware neural machine translation involves leveraging information beyond sentence-level context to resolve inter-sentential discourse dependencies and improve document-level translation quality, and has given rise to a number of recent techniques. However, despite well-reasoned intuitions, most context-aware translation models show only modest improvements over sentence-level systems. In this work, we investigate several challenges that impede progress within this field, relating to discourse phenomena, context usage, model architectures, and document-level evaluation. To address these problems, we propose a more realistic setting for document-level translation, called paragraph-to-paragraph (para2para) translation, and collect a new dataset of Chinese-English novels to promote future research.
Condensing Multilingual Knowledge with Lightweight Language-Specific Modules
Authors: Haoran Xu, Weiting Tan, Shuyue Stella Li, Yunmo Chen, Benjamin Van Durme, Philipp Koehn, Kenton Murray
Abstract
Incorporating language-specific (LS) modules is a proven method to boost performance in multilingual machine translation. This approach bears similarity to Mixture-of-Experts (MoE) because it does not inflate FLOPs. However, the scalability of this approach to hundreds of languages (experts) tends to be unmanageable due to the prohibitive number of parameters introduced by full-rank matrices in fully-connected layers. In this work, we introduce the Language-Specific Matrix Synthesis (LMS) method. This approach constructs LS modules by generating low-rank matrices from two significantly smaller matrices to approximate the full-rank matrix. Furthermore, we condense multilingual knowledge from multiple LS modules into a single shared module with the Fuse Distillation (FD) technique to improve the efficiency of inference and model serialization. We show that our LMS method significantly outperforms previous LS methods and MoE methods with the same amount of extra parameters, e.g., 1.73 BLEU points over the Switch Transformer on many-to-many multilingual machine translation. Importantly, LMS is able to have comparable translation performance with much fewer parameters.
Sāmayik: A Benchmark and Dataset for English-Sanskrit Translation
Abstract
Sanskrit is a low-resource language with a rich heritage. Digitized Sanskrit corpora reflective of the contemporary usage of Sanskrit, specifically that too in prose, is heavily under-represented at present. Presently, no such English-Sanskrit parallel dataset is publicly available. We release a dataset, S\={a}mayik, of more than 42,000 parallel English-Sanskrit sentences, from four different corpora that aim to bridge this gap. Moreover, we also release benchmarks adapted from existing multilingual pretrained models for Sanskrit-English translation. We include training splits from our contemporary dataset and the Sanskrit-English parallel sentences from the training split of Itih\={a}sa, a previously released classical era machine translation dataset containing Sanskrit.
Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation
Abstract
Gender bias is a significant issue in machine translation, leading to ongoing research efforts in developing bias mitigation techniques. However, most works focus on debiasing of bilingual models without consideration for multilingual systems. In this paper, we specifically target the unambiguous gender bias issue of multilingual machine translation models and propose a new mitigation method based on a novel perspective on the problem. We hypothesize that the gender bias in unambiguous settings is due to the lack of gender information encoded into the non-explicit gender words and devise a scheme to encode correct gender information into their latent embeddings. Specifically, we employ Gender-Aware Contrastive Learning, GACL, based on gender pseudo-labels to encode gender information on the encoder embeddings. Our method is target-language-agnostic and applicable to already trained multilingual machine translation models through post-fine-tuning. Through multilingual evaluation, we show that our approach improves gender accuracy by a wide margin without hampering translation performance. We also observe that incorporated gender information transfers and benefits other target languages regarding gender accuracy. Finally, we demonstrate that our method is applicable and beneficial to models of various sizes.
Improving speech translation by fusing speech and text
Authors: Wenbiao Yin, Zhicheng Liu, Chengqi Zhao, Tao Wang, Jian Tong, Rong Ye
Subjects: Computation and Language (cs.CL); Sound (cs.SD); Audio and Speech Processing (eess.AS)
Abstract
In speech translation, leveraging multimodal data to improve model performance and address limitations of individual modalities has shown significant effectiveness. In this paper, we harness the complementary strengths of speech and text, which are disparate modalities. We observe three levels of modality gap between them, denoted by Modal input representation, Modal semantic, and Modal hidden states. To tackle these gaps, we propose \textbf{F}use-\textbf{S}peech-\textbf{T}ext (\textbf{FST}), a cross-modal model which supports three distinct input modalities for translation: speech, text, and fused speech-text. We leverage multiple techniques for cross-modal alignment and conduct a comprehensive analysis to assess its impact on speech translation, machine translation, and fused speech-text translation. We evaluate FST on MuST-C, GigaST, and newstest benchmark. Experiments show that the proposed FST achieves an average 34.0 BLEU on MuST-C En$\rightarrow$De/Es/Fr (vs SOTA +1.1 BLEU). Further experiments demonstrate that FST does not degrade on MT task, as observed in prior works. Instead, it yields an average improvement of 3.2 BLEU over the pre-trained MT model.
In-context Example Selection for Machine Translation Using Multiple Features
Abstract
Large language models have demonstrated the capability to perform well on many NLP tasks when the input is prompted with a few examples (in-context learning) including machine translation, which is the focus of this work. The quality of translation depends on various features of the selected examples, such as their quality and relevance. However, previous work has predominantly focused on individual features for example selection. We propose a general framework for combining different features influencing example selection. We learn a regression function that selects examples based on multiple features in order to maximize the translation quality. On multiple language pairs and language models, we show that our example selection method significantly outperforms random selection as well as strong single-factor baselines reported in the literature. Using our example selection method, we see an improvement of over 2.5 COMET points on average with respect to a strong BM25 retrieval-based baseline.
When Does Monolingual Data Help Multilingual Translation: The Role of Domain and Model Scale
Authors: Christos Baziotis, Biao Zhang, Alexandra Birch, Barry Haddow
Abstract
Multilingual machine translation (MMT), trained on a mixture of parallel and monolingual data, is key for improving translation in low-resource language pairs. However, the literature offers conflicting results on the performance of different methods. To resolve this, we examine how denoising autoencoding (DAE) and backtranslation (BT) impact MMT under different data conditions and model scales. Unlike prior studies, we use a realistic dataset of 100 directions and consider many domain combinations of monolingual and test data. We find that monolingual data generally helps MMT, but models are surprisingly brittle to domain mismatches, especially at smaller model scales. BT is beneficial when the parallel, monolingual, and test data sources are similar but can be detrimental otherwise, while DAE is less effective than previously reported. Next, we analyze the impact of scale (from 90M to 1.6B parameters) and find it is important for both methods, particularly DAE. As scale increases, DAE transitions from underperforming the parallel-only baseline at 90M to converging with BT performance at 1.6B, and even surpassing it in low-resource. These results offer new insights into how to best use monolingual data in MMT.
WYWEB: A NLP Evaluation Benchmark For Classical Chinese
Abstract
To fully evaluate the overall performance of different NLP models in a given domain, many evaluation benchmarks are proposed, such as GLUE, SuperGLUE and CLUE. The fi eld of natural language understanding has traditionally focused on benchmarks for various tasks in languages such as Chinese, English, and multilingua, however, there has been a lack of attention given to the area of classical Chinese, also known as "wen yan wen", which has a rich history spanning thousands of years and holds signifi cant cultural and academic value. For the prosperity of the NLP community, in this paper, we introduce the WYWEB evaluation benchmark, which consists of nine NLP tasks in classical Chinese, implementing sentence classifi cation, sequence labeling, reading comprehension, and machine translation. We evaluate the existing pre-trained language models, which are all struggling with this benchmark. We also introduce a number of supplementary datasets and additional tools to help facilitate further progress on classical Chinese NLU. The github repository is https://github.com/baudzhou/WYWEB.
Beyond Shared Vocabulary: Increasing Representational Word Similarities across Languages for Multilingual Machine Translation
Authors: Di Wu, Christof Monz
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
Using a shared vocabulary is common practice in Multilingual Neural Machine Translation (MNMT). In addition to its simple design, shared tokens play an important role in positive knowledge transfer, which manifests naturally when the shared tokens refer to similar meanings across languages. However, natural flaws exist in such a design as well: 1) when languages use different writing systems, transfer is inhibited, and 2) even if languages use similar writing systems, shared tokens may have completely different meanings in different languages, increasing ambiguity. In this paper, we propose a re-parameterized method for building embeddings to alleviate the first problem. More specifically, we define word-level information transfer pathways via word equivalence classes and rely on graph networks to fuse word embeddings across languages. Our experiments demonstrate the advantages of our approach: 1) the semantics of embeddings are better aligned across languages, 2) our method achieves significant BLEU improvements on high- and low-resource MNMT, and 3) only less than 1.0\% additional trainable parameters are required with a limited increase in computational costs.
Exploring Representational Disparities Between Multilingual and Bilingual Translation Models
Abstract
Multilingual machine translation has proven immensely useful for low-resource and zero-shot language pairs. However, language pairs in multilingual models sometimes see worse performance than in bilingual models, especially when translating in a one-to-many setting. To understand why, we examine the geometric differences in the representations from bilingual models versus those from one-to-many multilingual models. Specifically, we evaluate the isotropy of the representations, to measure how well they utilize the dimensions in their underlying vector space. Using the same evaluation data in both models, we find that multilingual model decoder representations tend to be less isotropic than bilingual model decoder representations. Additionally, we show that much of the anisotropy in multilingual decoder representations can be attributed to modeling language-specific information, therefore limiting remaining representational capacity.
Multilingual Large Language Models Are Not (Yet) Code-Switchers
Authors: Ruochen Zhang, Samuel Cahyawijaya, Jan Christian Blaise Cruz, Alham Fikri Aji
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
Multilingual Large Language Models (LLMs) have recently shown great capability in various tasks, exhibiting state-of-the-art performance using few-shot or zero-shot prompting methods. While these models have been extensively studied in tasks where inputs are assumed to be in a single language, less attention has been paid to exploring their performance when inputs involve code-switching (CSW). In this paper, we provide an extensive empirical study of various multilingual LLMs and benchmark their performance in three tasks: sentiment analysis, machine translation, and word-level language identification. Our findings indicate that despite multilingual LLMs showing promising outcomes in certain tasks when using zero-/few-shot prompting, their performance still falls short on average when compared to smaller finetuned models. We argue that LLMs that are "multilingual" are not necessarily code-switching compatible and extensive future research is required to fully bridge this gap.
Revisiting Machine Translation for Cross-lingual Classification
Authors: Mikel Artetxe, Vedanuj Goswami, Shruti Bhosale, Angela Fan, Luke Zettlemoyer
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Machine Translation (MT) has been widely used for cross-lingual classification, either by translating the test set into English and running inference with a monolingual model (translate-test), or translating the training set into the target languages and finetuning a multilingual model (translate-train). However, most research in the area focuses on the multilingual models rather than the MT component. We show that, by using a stronger MT system and mitigating the mismatch between training on original text and running inference on machine translated text, translate-test can do substantially better than previously assumed. The optimal approach, however, is highly task dependent, as we identify various sources of cross-lingual transfer gap that affect different tasks and approaches differently. Our work calls into question the dominance of multilingual models for cross-lingual classification, and prompts to pay more attention to MT-based baselines.
Pixel Representations for Multilingual Translation and Data-efficient Cross-lingual Transfer
Authors: Elizabeth Salesky, Neha Verma, Philipp Koehn, Matt Post
Abstract
We introduce and demonstrate how to effectively train multilingual machine translation models with pixel representations. We experiment with two different data settings with a variety of language and script coverage, and show performance competitive with subword embeddings. We analyze various properties of pixel representations to better understand where they provide potential benefits and the impact of different scripts and data representations. We observe that these properties not only enable seamless cross-lingual transfer to unseen scripts, but make pixel representations more data-efficient than alternatives such as vocabulary expansion. We hope this work contributes to more extensible multilingual models for all languages and scripts.
Ties Matter: Modifying Kendall's Tau for Modern Metric Meta-Evaluation
Authors: Daniel Deutsch, George Foster, Markus Freitag
Abstract
Kendall's tau is frequently used to meta-evaluate how well machine translation (MT) evaluation metrics score individual translations. Its focus on pairwise score comparisons is intuitive but raises the question of how ties should be handled, a gray area that has motivated different variants in the literature. We demonstrate that, in settings like modern MT meta-evaluation, existing variants have weaknesses arising from their handling of ties, and in some situations can even be gamed. We propose a novel variant that gives metrics credit for correctly predicting ties, as well as an optimization procedure that automatically introduces ties into metric scores, enabling fair comparison between metrics that do and do not predict ties. We argue and provide experimental evidence that these modifications lead to fairer Kendall-based assessments of metric performance.
Empowering LLM-based Machine Translation with Cultural Awareness
Authors: Binwei Yao, Ming Jiang, Diyi Yang, Junjie Hu
Abstract
Traditional neural machine translation (NMT) systems often fail to translate sentences that contain culturally specific information. Most previous NMT methods have incorporated external cultural knowledge during training, which requires fine-tuning on low-frequency items specific to the culture. Recent in-context learning utilizes lightweight prompts to guide large language models (LLMs) to perform machine translation, however, whether such an approach works in terms of injecting culture awareness into machine translation remains unclear. To this end, we introduce a new data curation pipeline to construct a culturally relevant parallel corpus, enriched with annotations of cultural-specific entities. Additionally, we design simple but effective prompting strategies to assist this LLM-based translation. Extensive experiments show that our approaches can largely help incorporate cultural knowledge into LLM-based machine translation, outperforming traditional NMT systems in translating cultural-specific sentences.
Keyword: non-autoregressive
Optimizing Non-Autoregressive Transformers with Contrastive Learning
Authors: Chenxin An, Jiangtao Feng, Fei Huang, xipeng Qiu
Abstract
Non-autoregressive Transformers (NATs) reduce the inference latency of Autoregressive Transformers (ATs) by predicting words all at once rather than in sequential order. They have achieved remarkable progress in machine translation as well as many other applications. However, a long-standing challenge for NATs is the learning of multi-modality data distribution, which is the main cause of the performance gap between NATs and ATs. In this paper, we propose to ease the difficulty of modality learning via sampling from the model distribution instead of the data distribution. We derive contrastive constraints to stabilize the training process and integrate this resulting objective with the state-of-the-art NAT architecture DA-Transformer. Our model \method is examined on 3 different tasks, including machine translation, text summarization, and paraphrasing with 5 benchmarks. Results show that our approach outperforms previous non-autoregressive baselines by a significant margin and establishes new state-of-the-art results for non-autoregressive transformers on all the benchmarks.
Keyword: abstractive summarization
Abstractive Text Summarization Using the BRIO Training Paradigm
Abstract
Summary sentences produced by abstractive summarization models may be coherent and comprehensive, but they lack control and rely heavily on reference summaries. The BRIO training paradigm assumes a non-deterministic distribution to reduce the model's dependence on reference summaries, and improve model performance during inference. This paper presents a straightforward but effective technique to improve abstractive summaries by fine-tuning pre-trained language models, and training them with the BRIO paradigm. We build a text summarization dataset for Vietnamese, called VieSum. We perform experiments with abstractive summarization models trained with the BRIO paradigm on the CNNDM and the VieSum datasets. The results show that the models, trained on basic hardware, outperform all existing abstractive summarization models, especially for Vietnamese.
Keyword: factual
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
Abstract
Automatic summarization generates concise summaries that contain key ideas of source documents. As the most mainstream datasets for the news sub-domain, CNN/DailyMail and BBC XSum have been widely used for performance benchmarking. However, the reference summaries of those datasets turn out to be noisy, mainly in terms of factual hallucination and information redundancy. To address this challenge, we first annotate new expert-writing Element-aware test sets following the "Lasswell Communication Model" proposed by Lasswell (1948), allowing reference summaries to focus on more fine-grained news elements objectively and comprehensively. Utilizing the new test sets, we observe the surprising zero-shot summary ability of LLMs, which addresses the issue of the inconsistent results between human preference and automatic evaluation metrics of LLMs' zero-shot summaries in prior work. Further, we propose a Summary Chain-of-Thought (SumCoT) technique to elicit LLMs to generate summaries step by step, which helps them integrate more fine-grained details of source documents into the final summaries that correlate with the human writing mindset. Experimental results show our method outperforms state-of-the-art fine-tuned PLMs and zero-shot LLMs by +4.33/+4.77 in ROUGE-L on the two datasets, respectively. Dataset and code are publicly available at https://github.com/Alsace08/SumCoT.
Abstract
Misinformation, i.e. factually incorrect information, is often conveyed in multiple modalities, e.g. an image accompanied by a caption. It is perceived as more credible by humans, and spreads faster and wider than its text-only counterparts. While an increasing body of research investigates automated fact-checking (AFC), previous surveys mostly focus on textual misinformation. In this survey, we conceptualise a framework for AFC including subtasks unique to multimodal misinformation. Furthermore, we discuss related terminological developed in different communities in the context of our framework. We focus on four modalities prevalent in real-world fact-checking: text, image, audio, and video. We survey benchmarks and models, and discuss limitations and promising directions for future research.
Improving Classifier Robustness through Active Generation of Pairwise Counterfactuals
Authors: Ananth Balashankar, Xuezhi Wang, Yao Qin, Ben Packer, Nithum Thain, Jilin Chen, Ed H. Chi, Alex Beutel
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
Counterfactual Data Augmentation (CDA) is a commonly used technique for improving robustness in natural language classifiers. However, one fundamental challenge is how to discover meaningful counterfactuals and efficiently label them, with minimal human labeling cost. Most existing methods either completely rely on human-annotated labels, an expensive process which limits the scale of counterfactual data, or implicitly assume label invariance, which may mislead the model with incorrect labels. In this paper, we present a novel framework that utilizes counterfactual generative models to generate a large number of diverse counterfactuals by actively sampling from regions of uncertainty, and then automatically label them with a learned pairwise classifier. Our key insight is that we can more correctly label the generated counterfactuals by training a pairwise classifier that interpolates the relationship between the original example and the counterfactual. We demonstrate that with a small amount of human-annotated counterfactual data (10%), we can generate a counterfactual augmentation dataset with learned labels, that provides an 18-20% improvement in robustness and a 14-21% reduction in errors on 6 out-of-domain datasets, comparable to that of a fully human-annotated counterfactual dataset for both sentiment classification and question paraphrase tasks.
Polyglot or Not? Measuring Multilingual Encyclopedic Knowledge Retrieval from Foundation Language Models
Abstract
In this work, we evaluate the capacity for foundation models to retrieve encyclopedic knowledge across a wide range of languages, topics, and contexts. To support this effort, we 1) produce a new dataset containing 303k factual associations in 20 different languages, 2) formulate a new counterfactual knowledge assessment, Polyglot or Not, and 3) benchmark 5 foundation models in a multilingual setting and a diverse set of 20 models in an English-only setting. We observed significant accuracy differences in models of interest, with Meta's LLaMA topping both the multilingual and English-only assessments. Error analysis reveals a significant deficiency in LLaMA's ability to retrieve facts in languages written in the Cyrillic script and gaps in its understanding of facts based on the location and gender of entailed subjects. Ultimately, we argue that the promise of utilizing foundation language models as bonafide polyglots is greatly diminished when they are tasked with retrieving information in languages other than English. Supporting code (https://github.com/daniel-furman/Polyglot-or-Not) and dataset (https://huggingface.co/datasets/Polyglot-or-Not/Fact-Completion) are openly released.
Improving Language Models via Plug-and-Play Retrieval Feedback
Abstract
Large language models (LLMs) exhibit remarkable performance across various NLP tasks. However, they often generate incorrect or hallucinated information, which hinders their practical applicability in real-world scenarios. Human feedback has been shown to effectively enhance the factuality and quality of generated content, addressing some of these limitations. However, this approach is resource-intensive, involving manual input and supervision, which can be time-consuming and expensive. Moreover, it cannot be provided during inference, further limiting its practical utility in dynamic and interactive applications. In this paper, we introduce ReFeed, a novel pipeline designed to enhance LLMs by providing automatic retrieval feedback in a plug-and-play framework without the need for expensive fine-tuning. ReFeed first generates initial outputs, then utilizes a retrieval model to acquire relevant information from large document collections, and finally incorporates the retrieved information into the in-context demonstration for output refinement, thereby addressing the limitations of LLMs in a more efficient and cost-effective manner. Experiments on four knowledge-intensive benchmark datasets demonstrate our proposed ReFeed could improve over +6.0% under zero-shot setting and +2.5% under few-shot setting, compared to baselines without using retrieval feedback.
IfQA: A Dataset for Open-domain Question Answering under Counterfactual Presuppositions
Authors: Wenhao Yu, Meng Jiang, Peter Clark, Ashish Sabharwal
Abstract
Although counterfactual reasoning is a fundamental aspect of intelligence, the lack of large-scale counterfactual open-domain question-answering (QA) benchmarks makes it difficult to evaluate and improve models on this ability. To address this void, we introduce the first such dataset, named IfQA, where each question is based on a counterfactual presupposition via an "if" clause. For example, if Los Angeles was on the east coast of the U.S., what would be the time difference between Los Angeles and Paris? Such questions require models to go beyond retrieving direct factual knowledge from the Web: they must identify the right information to retrieve and reason about an imagined situation that may even go against the facts built into their parameters. The IfQA dataset contains over 3,800 questions that were annotated annotated by crowdworkers on relevant Wikipedia passages. Empirical analysis reveals that the IfQA dataset is highly challenging for existing open-domain QA methods, including supervised retrieve-then-read pipeline methods (EM score 36.2), as well as recent few-shot approaches such as chain-of-thought prompting with GPT-3 (EM score 27.4). The unique challenges posed by the IfQA benchmark will push open-domain QA research on both retrieval and counterfactual reasoning fronts.
Evaluating Factual Consistency of Summaries with Large Language Models
Abstract
Detecting factual errors in summaries has been an important and challenging subject in summarization research. Inspired by the emergent ability of large language models (LLMs), we explore evaluating factual consistency of summaries by directly prompting LLMs. We present a comprehensive empirical study to assess the ability of LLMs as factual consistency evaluators, which consists of (1) analyzing different LLMs such as the GPT model series and Flan-T5; (2) investigating a variety of prompting methods including vanilla prompting, chain-of-thought prompting, and a sentence-by-sentence prompting method to tackle long summaries; and (3) evaluating on diverse summaries generated by multiple summarization systems, ranging from pre-transformer methods to SOTA pretrained models. Our experiments demonstrate that prompting LLMs is able to outperform the previous best factuality systems in all settings, by up to 12.2 absolute points in terms of the binary classification accuracy on inconsistency detection.
To Copy Rather Than Memorize: A Vertical Learning Paradigm for Knowledge Graph Completion
Abstract
Embedding models have shown great power in knowledge graph completion (KGC) task. By learning structural constraints for each training triple, these methods implicitly memorize intrinsic relation rules to infer missing links. However, this paper points out that the multi-hop relation rules are hard to be reliably memorized due to the inherent deficiencies of such implicit memorization strategy, making embedding models underperform in predicting links between distant entity pairs. To alleviate this problem, we present Vertical Learning Paradigm (VLP), which extends embedding models by allowing to explicitly copy target information from related factual triples for more accurate prediction. Rather than solely relying on the implicit memory, VLP directly provides additional cues to improve the generalization ability of embedding models, especially making the distant link prediction significantly easier. Moreover, we also propose a novel relative distance based negative sampling technique (ReD) for more effective optimization. Experiments demonstrate the validity and generality of our proposals on two standard benchmarks. Our code is available at https://github.com/rui9812/VLP.
Question Answering as Programming for Solving Time-Sensitive Questions
Authors: Xinyu Zhu, Cheng Yang, Bei Chen, Siheng Li, Jian-Guang Lou, Yujiu Yang
Abstract
In this work we try to apply Large Language Models (LLMs) to reframe the Question Answering task as Programming (QAaP). Due to the inherent dynamic nature of the real world, factual questions frequently involve a symbolic constraint: time, solving these questions necessitates not only extensive world knowledge, but also advanced reasoning ability to satisfy the temporal constraints. Despite the remarkable intelligence exhibited by LLMs in various NLP tasks, our experiments reveal that the aforementioned problems continue to pose a significant challenge to existing LLMs. To solve these time-sensitive factual questions, considering that modern LLMs possess superior ability in both natural language understanding and programming,we endeavor to leverage LLMs to represent diversely expressed text as well-structured code, and thereby grasp the desired knowledge along with the underlying symbolic constraint.
ManiTweet: A New Benchmark for Identifying Manipulation of News on Social Media
Authors: Kung-Hsiang Huang, Hou Pong Chan, Kathleen McKeown, Heng Ji
Abstract
Considerable advancements have been made to tackle the misrepresentation of information derived from reference articles in the domains of fact-checking and faithful summarization. However, an unaddressed aspect remains - the identification of social media posts that manipulate information within associated news articles. This task presents a significant challenge, primarily due to the prevalence of personal opinions in such posts. We present a novel task, identifying manipulation of news on social media, which aims to detect manipulation in social media posts and identify manipulated or inserted information. To study this task, we have proposed a data collection schema and curated a dataset called ManiTweet, consisting of 3.6K pairs of tweets and corresponding articles. Our analysis demonstrates that this task is highly challenging, with large language models (LLMs) yielding unsatisfactory performance. Additionally, we have developed a simple yet effective basic model that outperforms LLMs significantly on the ManiTweet dataset. Finally, we have conducted an exploratory analysis of human-written tweets, unveiling intriguing connections between manipulation and the domain and factuality of news articles, as well as revealing that manipulated sentences are more likely to encapsulate the main story or consequences of a news outlet.
FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
Authors: Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Wei Koh, Mohit Iyyer, Luke Zettlemoyer, Hannaneh Hajishirzi
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Evaluating the factuality of long-form text generated by large language models (LMs) is non-trivial because (1) generations often contain a mixture of supported and unsupported pieces of information, making binary judgments of quality inadequate, and (2) human evaluation is time-consuming and costly. In this paper, we introduce FActScore (Factual precision in Atomicity Score), a new evaluation that breaks a generation into a series of atomic facts and computes the percentage of atomic facts supported by a reliable knowledge source. We conduct an extensive human evaluation to obtain FActScores of people biographies generated by several state-of-the-art commercial LMs -- InstructGPT, ChatGPT, and the retrieval-augmented PerplexityAI -- and report new analysis demonstrating the need for such a fine-grained score (e.g., ChatGPT only achieves 58%). Since human evaluation is costly, we also introduce an automated model that estimates FActScore, using retrieval and a strong language model, with less than a 2% error rate. Finally, we use this automated metric to evaluate 6,500 generations from a new set of 13 recent LMs that would have cost $26K if evaluated by humans, with various findings: GPT-4 and ChatGPT are more factual than public models, and Vicuna and Alpaca are some of the best public models.
WikiChat: A Few-Shot LLM-Based Chatbot Grounded with Wikipedia
Authors: Sina J. Semnani, Violet Z. Yao, Heidi C. Zhang, Monica S. Lam
Abstract
Despite recent advances in Large Language Models (LLMs), users still cannot trust the information provided in their responses. LLMs cannot speak accurately about events that occurred after their training, which are often topics of great interest to users, and, as we show in this paper, they are highly prone to hallucination when talking about less popular (tail) topics. This paper presents WikiChat, a few-shot LLM-based chatbot that is grounded with live information from Wikipedia. Through many iterations of experimentation, we have crafte a pipeline based on information retrieval that (1) uses LLMs to suggest interesting and relevant facts that are individually verified against Wikipedia, (2) retrieves additional up-to-date information, and (3) composes coherent and engaging time-aware responses. We propose a novel hybrid human-and-LLM evaluation methodology to analyze the factuality and conversationality of LLM-based chatbots. We focus on evaluating important but previously neglected issues such as conversing about recent and tail topics. We evaluate WikiChat against strong fine-tuned and LLM-based baselines across a diverse set of conversation topics. We find that WikiChat outperforms all baselines in terms of the factual accuracy of its claims, by up to 12.1%, 28.3% and 32.7% on head, recent and tail topics, while matching GPT-3.5 in terms of providing natural, relevant, non-repetitive and informational responses.
WebIE: Faithful and Robust Information Extraction on the Web
Authors: Chenxi Whitehouse, Clara Vania, Alham Fikri Aji, Christos Christodoulopoulos, Andrea Pierleoni
Abstract
Extracting structured and grounded fact triples from raw text is a fundamental task in Information Extraction (IE). Existing IE datasets are typically collected from Wikipedia articles, using hyperlinks to link entities to the Wikidata knowledge base. However, models trained only on Wikipedia have limitations when applied to web domains, which often contain noisy text or text that does not have any factual information. We present WebIE, the first large-scale, entity-linked closed IE dataset consisting of 1.6M sentences automatically collected from the English Common Crawl corpus. WebIE also includes negative examples, i.e. sentences without fact triples, to better reflect the data on the web. We annotate ~25K triples from WebIE through crowdsourcing and introduce mWebIE, a translation of the annotated set in four other languages: French, Spanish, Portuguese, and Hindi. We evaluate the in-domain, out-of-domain, and zero-shot cross-lingual performance of generative IE models and find models trained on WebIE show better generalisability. We also propose three training strategies that use entity linking as an auxiliary task. Our experiments show that adding Entity-Linking objectives improves the faithfulness of our generative IE models.
USB: A Unified Summarization Benchmark Across Tasks and Domains
Authors: Kundan Krishna, Prakhar Gupta, Sanjana Ramprasad, Byron C. Wallace, Jeffrey P. Bigham, Zachary C. Lipton
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
An abundance of datasets exist for training and evaluating models on the task of summary generation.However, these datasets are often derived heuristically, and lack sufficient annotations to support research into all aspects of summarization, such as evidence extraction and controllable summarization. We introduce a benchmark comprising 8 tasks that require multi-dimensional understanding of summarization, e.g., surfacing evidence for a summary, assessing its correctness, and gauging its relevance to different topics. We compare various methods on this benchmark and discover that on multiple tasks, moderately-sized fine-tuned models consistently outperform much larger few-shot prompted language models. For factuality related tasks, we also evaluate existing heuristics to create training data and find that training on them performs worse than training on $20\times$ less human-labeled data. Our benchmark consists of data from 6 different domains, allowing us to study cross-domain performance of trained models. We find that for some tasks, the amount of training data matters more than the domain where it comes from, while for other tasks training specifically on data from the target domain, even if limited, is more beneficial. Our work fulfills the need for a well-annotated summarization benchmark with diverse tasks, and provides useful insights about the impact of the quality, size and domain of training data.
Improving Factuality and Reasoning in Language Models through Multiagent Debate
Authors: Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
Large language models (LLMs) have demonstrated remarkable capabilities in language generation, understanding, and few-shot learning in recent years. An extensive body of work has explored how their performance may be further improved through the tools of prompting, ranging from verification, self-consistency, or intermediate scratchpads. In this paper, we present a complementary approach to improve language responses where multiple language model instances propose and debate their individual responses and reasoning processes over multiple rounds to arrive at a common final answer. Our findings indicate that this approach significantly enhances mathematical and strategic reasoning across a number of tasks. We also demonstrate that our approach improves the factual validity of generated content, reducing fallacious answers and hallucinations that contemporary models are prone to. Our approach may be directly applied to existing black-box models and uses identical procedure and prompts for all tasks we investigate. Overall, our findings suggest that such "society of minds" approach has the potential to significantly advance the capabilities of LLMs and pave the way for further breakthroughs in language generation and understanding.
Evaluating and Modeling Attribution for Cross-Lingual Question Answering
Authors: Benjamin Muller, John Wieting, Jonathan H. Clark, Tom Kwiatkowski, Sebastian Ruder, Livio Baldini Soares, Roee Aharoni, Jonathan Herzig, Xinyi Wang
Abstract
Trustworthy answer content is abundant in many high-resource languages and is instantly accessible through question answering systems, yet this content can be hard to access for those that do not speak these languages. The leap forward in cross-lingual modeling quality offered by generative language models offers much promise, yet their raw generations often fall short in factuality. To improve trustworthiness in these systems, a promising direction is to attribute the answer to a retrieved source, possibly in a content-rich language different from the query. Our work is the first to study attribution for cross-lingual question answering. First, we collect data in 5 languages to assess the attribution level of a state-of-the-art cross-lingual QA system. To our surprise, we find that a substantial portion of the answers is not attributable to any retrieved passages (up to 50% of answers exactly matching a gold reference) despite the system being able to attend directly to the retrieved text. Second, to address this poor attribution level, we experiment with a wide range of attribution detection techniques. We find that Natural Language Inference models and PaLM 2 fine-tuned on a very small amount of attribution data can accurately detect attribution. Based on these models, we improve the attribution level of a cross-lingual question-answering system. Overall, we show that current academic generative cross-lingual QA systems have substantial shortcomings in attribution and we build tooling to mitigate these issues.
Keyword: knowledge distillation
One-stop Training of Multiple Capacity Models for Multilingual Machine Translation
Abstract
Training models with varying capacities can be advantageous for deploying them in different scenarios. While high-capacity models offer better performance, low-capacity models require fewer computing resources for training and inference. In this work, we propose a novel one-stop training framework consisting of two composite model architectures and a joint training algorithm called Two-Stage Joint-Training (TSJT). Unlike knowledge distillation, where multiple capacity models are trained from scratch separately, our approach integrates supervisions from different flexible-capacity models simultaneously, leading to faster and more efficient convergence. Extensive experiments on the WMT10 benchmark show that our method outperforms low-capacity baseline models and achieves comparable or better performance on high-capacity models. Notably, the analysis demonstrates that our method significantly influences the initial training process, leading to more efficient convergence and superior solutions.
Keyword: Hallucination
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
Abstract
Automatic summarization generates concise summaries that contain key ideas of source documents. As the most mainstream datasets for the news sub-domain, CNN/DailyMail and BBC XSum have been widely used for performance benchmarking. However, the reference summaries of those datasets turn out to be noisy, mainly in terms of factual hallucination and information redundancy. To address this challenge, we first annotate new expert-writing Element-aware test sets following the "Lasswell Communication Model" proposed by Lasswell (1948), allowing reference summaries to focus on more fine-grained news elements objectively and comprehensively. Utilizing the new test sets, we observe the surprising zero-shot summary ability of LLMs, which addresses the issue of the inconsistent results between human preference and automatic evaluation metrics of LLMs' zero-shot summaries in prior work. Further, we propose a Summary Chain-of-Thought (SumCoT) technique to elicit LLMs to generate summaries step by step, which helps them integrate more fine-grained details of source documents into the final summaries that correlate with the human writing mindset. Experimental results show our method outperforms state-of-the-art fine-tuned PLMs and zero-shot LLMs by +4.33/+4.77 in ROUGE-L on the two datasets, respectively. Dataset and code are publicly available at https://github.com/Alsace08/SumCoT.
How Language Model Hallucinations Can Snowball
Authors: Muru Zhang, Ofir Press, William Merrill, Alisa Liu, Noah A. Smith
Abstract
A major risk of using language models in practical applications is their tendency to hallucinate incorrect statements. Hallucinations are often attributed to knowledge gaps in LMs, but we hypothesize that in some cases, when justifying previously generated hallucinations, LMs output false claims that they can separately recognize as incorrect. We construct three question-answering datasets where ChatGPT and GPT-4 often state an incorrect answer and offer an explanation with at least one incorrect claim. Crucially, we find that ChatGPT and GPT-4 can identify 67% and 87% of their own mistakes, respectively. We refer to this phenomenon as hallucination snowballing: an LM over-commits to early mistakes, leading to more mistakes that it otherwise would not make.
Detecting and Mitigating Hallucinations in Multilingual Summarisation
Authors: Yifu Qiu, Yftah Ziser, Anna Korhonen, Edoardo M. Ponti, Shay B. Cohen
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Hallucinations pose a significant challenge to the reliability of neural models for abstractive summarisation. While automatically generated summaries may be fluent, they often lack faithfulness to the original document. This issue becomes even more pronounced in low-resource settings, such as cross-lingual transfer. With the existing faithful metrics focusing on English, even measuring the extent of this phenomenon in cross-lingual settings is hard. To address this, we first develop a novel metric, mFACT, evaluating the faithfulness of non-English summaries, leveraging translation-based transfer from multiple English faithfulness metrics. We then propose a simple but effective method to reduce hallucinations with a cross-lingual transfer, which weighs the loss of each training example by its faithfulness score. Through extensive experiments in multiple languages, we demonstrate that mFACT is the metric that is most suited to detect hallucinations. Moreover, we find that our proposed loss weighting method drastically increases both performance and faithfulness according to both automatic and human evaluation when compared to strong baselines for cross-lingual transfer such as MAD-X. Our code and dataset are available at https://github.com/yfqiu-nlp/mfact-summ.
Mitigating Language Model Hallucination with Interactive Question-Knowledge Alignment
Authors: Shuo Zhang, Liangming Pan, Junzhou Zhao, William Yang Wang
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
Despite the remarkable recent advances in language models, they still struggle with the hallucination problem and can generate misleading and unsupported responses. A common approach to mitigate the hallucination issue is retrieving and incorporating supporting evidence from a knowledge base. However, user questions usually do not align well with the stored knowledge, as they are unaware of the information available before asking questions. This misalignment can limit the language model's ability to locate and utilize the knowledge, potentially forcing it to hallucinate by ignoring or overriding the retrieved evidence. To address this issue, we introduce MixAlign, a framework that interacts with both the user and the knowledge base to obtain and integrate clarifications on how the user question relates to the stored information. MixAlign employs a language model to achieve automatic question-knowledge alignment and, if necessary, further enhances this alignment through human user clarifications. Experimental results demonstrate significant improvements over state-of-the-art methods, showcasing the effectiveness of MixAlign in mitigating language model hallucination.
Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models
Authors: Alfonso Amayuelas, Liangming Pan, Wenhu Chen, William Wang
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
This paper investigates the capabilities of Large Language Models (LLMs) in the context of understanding their own knowledge and measuring their uncertainty. We argue this is an important feature for mitigating hallucinations. Specifically, we focus on addressing \textit{known-unknown} questions, characterized by high uncertainty due to the absence of definitive answers. To facilitate our study, we collect a dataset with new Known-Unknown Questions (KUQ) and propose a novel categorization scheme to elucidate the sources of uncertainty. Subsequently, we assess the LLMs' ability to differentiate between known and unknown questions and classify them accordingly. Moreover, we evaluate the quality of their answers in an Open-Ended QA setting. To quantify the uncertainty expressed in the answers, we create a semantic evaluation method that measures the model's accuracy in expressing uncertainty between known vs unknown questions.
PaD: Program-aided Distillation Specializes Large Models in Reasoning
Abstract
While Large Language Models (LLMs) excel in several natural language processing tasks, their size and inaccessibility present challenges for extensive practical application. Previous studies acquire specialized skills through distillation on LLMs, which result in trading generic abilities, called model specialization. As for reasoning ability, chain-of-thought was synthesized to subsequent distillation. However, due to hallucination, synthetic chain-of-thought from LLMs contains faulty reasoning. These incorrect reasoning steps damage the reasoning capability. To tackle above issues, we propose Program-aided Distillation (PaD), which distills LLMs to obtain specialized small models in reasoning tasks. In PaD, we strengthen specialized models with program-aided reasoning, and help them overcome faulty reasoning steps with automated error checking. Experimental results demonstrate that, on the GSM8K benchmark, a 0.06B model using PaD can not only outperform certain LLMs (e.g., LLaMA), but also achieves a 10% improvement over baselines with a significantly smaller scale of parameters and data. Data pruning analysis reveals that PaD possesses higher training efficiency.
mmT5: Modular Multilingual Pre-Training Solves Source Language Hallucinations
Authors: Jonas Pfeiffer, Francesco Piccinno, Massimo Nicosia, Xinyi Wang, Machel Reid, Sebastian Ruder
Abstract
Multilingual sequence-to-sequence models perform poorly with increased language coverage and fail to consistently generate text in the correct target language in few-shot settings. To address these challenges, we propose mmT5, a modular multilingual sequence-to-sequence model. mmT5 utilizes language-specific modules during pre-training, which disentangle language-specific information from language-agnostic information. We identify representation drift during fine-tuning as a key limitation of modular generative models and develop strategies that enable effective zero-shot transfer. Our model outperforms mT5 at the same parameter sizes by a large margin on representative natural language understanding and generation tasks in 40+ languages. Compared to mT5, mmT5 raises the rate of generating text in the correct language under zero-shot settings from 7% to 99%, thereby greatly alleviating the source language hallucination problem.
WikiChat: A Few-Shot LLM-Based Chatbot Grounded with Wikipedia
Authors: Sina J. Semnani, Violet Z. Yao, Heidi C. Zhang, Monica S. Lam
Abstract
Despite recent advances in Large Language Models (LLMs), users still cannot trust the information provided in their responses. LLMs cannot speak accurately about events that occurred after their training, which are often topics of great interest to users, and, as we show in this paper, they are highly prone to hallucination when talking about less popular (tail) topics. This paper presents WikiChat, a few-shot LLM-based chatbot that is grounded with live information from Wikipedia. Through many iterations of experimentation, we have crafte a pipeline based on information retrieval that (1) uses LLMs to suggest interesting and relevant facts that are individually verified against Wikipedia, (2) retrieves additional up-to-date information, and (3) composes coherent and engaging time-aware responses. We propose a novel hybrid human-and-LLM evaluation methodology to analyze the factuality and conversationality of LLM-based chatbots. We focus on evaluating important but previously neglected issues such as conversing about recent and tail topics. We evaluate WikiChat against strong fine-tuned and LLM-based baselines across a diverse set of conversation topics. We find that WikiChat outperforms all baselines in terms of the factual accuracy of its claims, by up to 12.1%, 28.3% and 32.7% on head, recent and tail topics, while matching GPT-3.5 in terms of providing natural, relevant, non-repetitive and informational responses.
Improving Factuality and Reasoning in Language Models through Multiagent Debate
Authors: Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Abstract
Large language models (LLMs) have demonstrated remarkable capabilities in language generation, understanding, and few-shot learning in recent years. An extensive body of work has explored how their performance may be further improved through the tools of prompting, ranging from verification, self-consistency, or intermediate scratchpads. In this paper, we present a complementary approach to improve language responses where multiple language model instances propose and debate their individual responses and reasoning processes over multiple rounds to arrive at a common final answer. Our findings indicate that this approach significantly enhances mathematical and strategic reasoning across a number of tasks. We also demonstrate that our approach improves the factual validity of generated content, reducing fallacious answers and hallucinations that contemporary models are prone to. Our approach may be directly applied to existing black-box models and uses identical procedure and prompts for all tasks we investigate. Overall, our findings suggest that such "society of minds" approach has the potential to significantly advance the capabilities of LLMs and pave the way for further breakthroughs in language generation and understanding.
Keyword: evaluation
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
Abstract
Automatic summarization generates concise summaries that contain key ideas of source documents. As the most mainstream datasets for the news sub-domain, CNN/DailyMail and BBC XSum have been widely used for performance benchmarking. However, the reference summaries of those datasets turn out to be noisy, mainly in terms of factual hallucination and information redundancy. To address this challenge, we first annotate new expert-writing Element-aware test sets following the "Lasswell Communication Model" proposed by Lasswell (1948), allowing reference summaries to focus on more fine-grained news elements objectively and comprehensively. Utilizing the new test sets, we observe the surprising zero-shot summary ability of LLMs, which addresses the issue of the inconsistent results between human preference and automatic evaluation metrics of LLMs' zero-shot summaries in prior work. Further, we propose a Summary Chain-of-Thought (SumCoT) technique to elicit LLMs to generate summaries step by step, which helps them integrate more fine-grained details of source documents into the final summaries that correlate with the human writing mindset. Experimental results show our method outperforms state-of-the-art fine-tuned PLMs and zero-shot LLMs by +4.33/+4.77 in ROUGE-L on the two datasets, respectively. Dataset and code are publicly available at https://github.com/Alsace08/SumCoT.
clembench: Using Game Play to Evaluate Chat-Optimized Language Models as Conversational Agents
Authors: Kranti Chalamalasetti, Jana Götze, Sherzod Hakimov, Brielen Madureira, Philipp Sadler, David Schlangen
Abstract
Recent work has proposed a methodology for the systematic evaluation of "Situated Language Understanding Agents"-agents that operate in rich linguistic and non-linguistic contexts-through testing them in carefully constructed interactive settings. Other recent work has argued that Large Language Models (LLMs), if suitably set up, can be understood as (simulators of) such agents. A connection suggests itself, which this paper explores: Can LLMs be evaluated meaningfully by exposing them to constrained game-like settings that are built to challenge specific capabilities? As a proof of concept, this paper investigates five interaction settings, showing that current chat-optimised LLMs are, to an extent, capable to follow game-play instructions. Both this capability and the quality of the game play, measured by how well the objectives of the different games are met, follows the development cycle, with newer models performing better. The metrics even for the comparatively simple example games are far from being saturated, suggesting that the proposed instrument will remain to have diagnostic value. Our general framework for implementing and evaluating games with LLMs is available at https://github.com/clp-research/clembench.
Look-back Decoding for Open-Ended Text Generation
Authors: Nan Xu, Chunting Zhou, Asli Celikyilmaz, Xuezhe Ma
Abstract
Given a prefix (context), open-ended generation aims to decode texts that are coherent, which don't abruptly drift from previous topics, and informative, which don't suffer from undesired repetitions. In this paper, we propose Look-back, an improved decoding algorithm that leverages the Kullback-Leibler divergence to track the distribution distance between current and historical decoding steps. Thus Look-back can automatically predict potential repetitive phrase and topic drift, and remove tokens that may cause the failure modes, restricting the next token probability distribution within a plausible distance to the history. We perform decoding experiments on document continuation and story generation, and demonstrate that Look-back is able to generate more fluent and coherent text, outperforming other strong decoding methods significantly in both automatic and human evaluations.
Neural Machine Translation for Code Generation
Authors: Dharma KC, Clayton T. Morrison
Subjects: Computation and Language (cs.CL); Machine Learning (cs.LG)
Abstract
Neural machine translation (NMT) methods developed for natural language processing have been shown to be highly successful in automating translation from one natural language to another. Recently, these NMT methods have been adapted to the generation of program code. In NMT for code generation, the task is to generate output source code that satisfies constraints expressed in the input. In the literature, a variety of different input scenarios have been explored, including generating code based on natural language description, lower-level representations such as binary or assembly (neural decompilation), partial representations of source code (code completion and repair), and source code in another language (code translation). In this paper we survey the NMT for code generation literature, cataloging the variety of methods that have been explored according to input and output representations, model architectures, optimization techniques used, data sets, and evaluation methods. We discuss the limitations of existing methods and future research directions
The Grammar and Syntax Based Corpus Analysis Tool For The Ukrainian Language
Authors: Daria Stetsenko, Inez Okulska
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
This paper provides an overview of a text mining tool the StyloMetrix developed initially for the Polish language and further extended for English and recently for Ukrainian. The StyloMetrix is built upon various metrics crafted manually by computational linguists and researchers from literary studies to analyze grammatical, stylistic, and syntactic patterns. The idea of constructing the statistical evaluation of syntactic and grammar features is straightforward and familiar for the languages like English, Spanish, German, and others; it is yet to be developed for low-resource languages like Ukrainian. We describe the StyloMetrix pipeline and provide some experiments with this tool for the text classification task. We also describe our package's main limitations and the metrics' evaluation procedure.
Open-world Semi-supervised Generalized Relation Discovery Aligned in a Real-world Setting
Abstract
Open-world Relation Extraction (OpenRE) has recently garnered significant attention. However, existing approaches tend to oversimplify the problem by assuming that all unlabeled texts belong to novel classes, thereby limiting the practicality of these methods. We argue that the OpenRE setting should be more aligned with the characteristics of real-world data. Specifically, we propose two key improvements: (a) unlabeled data should encompass known and novel classes, including hard-negative instances; and (b) the set of novel classes should represent long-tail relation types. Furthermore, we observe that popular relations such as titles and locations can often be implicitly inferred through specific patterns, while long-tail relations tend to be explicitly expressed in sentences. Motivated by these insights, we present a novel method called KNoRD (Known and Novel Relation Discovery), which effectively classifies explicitly and implicitly expressed relations from known and novel classes within unlabeled data. Experimental evaluations on several Open-world RE benchmarks demonstrate that KNoRD consistently outperforms other existing methods, achieving significant performance gains.
EntRED: Benchmarking Relation Extraction with Fewer Shortcuts
Abstract
Entity names play an effective role in relation extraction (RE) and often influence model performance. As a result, the entity names in the benchmarks' test sets significantly influence the evaluation of RE models. In this work, we find that the standard RE benchmarks' datasets have a large portion of incorrect entity annotations, low entity name diversity, and are prone to have shortcuts from entity names to ground-truth relations. These issues make the standard benchmarks far from reflecting the real-world scenarios. Hence, in this work, we present EntRED, a challenging RE benchmark with reduced shortcuts and higher diversity of entities. To build EntRED, we propose an end-to-end entity replacement pipeline based on causal inference (CI): ERIC. ERIC performs type-constrained replacements on entities to reduce the shortcuts from entity bias to ground-truth relations. ERIC applies CI in two aspects: 1) targeting the instances that need entity replacements, and 2) determining the candidate entities for replacements. We apply ERIC on TACRED to produce EntRED. Our EntRED evaluates whether the RE model can correctly extract the relations from the text instead of relying on entity bias. Empirical results reveal that even the strong RE model has a significant performance drop on EntRED, which memorizes entity name patterns instead of reasoning from the textual context. We release ERIC's source code and the EntRED benchmark at https://github.com/wangywUST/ENTRED.
Cross-Attention is Not Enough: Incongruity-Aware Multimodal Sentiment Analysis and Emotion Recognition
Authors: Yaoting Wang, Yuanchao Li, Peter Bell, Catherine Lai
Subjects: Computation and Language (cs.CL); Multimedia (cs.MM); Audio and Speech Processing (eess.AS); Image and Video Processing (eess.IV)
Abstract
Fusing multiple modalities for affective computing tasks has proven effective for performance improvement. However, how multimodal fusion works is not well understood, and its use in the real world usually results in large model sizes. In this work, on sentiment and emotion analysis, we first analyze how the salient affective information in one modality can be affected by the other in crossmodal attention. We find that inter-modal incongruity exists at the latent level due to crossmodal attention. Based on this finding, we propose a lightweight model via Hierarchical Crossmodal Transformer with Modality Gating (HCT-MG), which determines a primary modality according to its contribution to the target task and then hierarchically incorporates auxiliary modalities to alleviate inter-modal incongruity and reduce information redundancy. The experimental evaluation on three benchmark datasets: CMU-MOSI, CMU-MOSEI, and IEMOCAP verifies the efficacy of our approach, showing that it: 1) outperforms major prior work by achieving competitive results and can successfully recognize hard samples; 2) mitigates the inter-modal incongruity at the latent level when modalities have mismatched affective tendencies; 3) reduces model size to less than 1M parameters while outperforming existing models of similar sizes.
ReSee: Responding through Seeing Fine-grained Visual Knowledge in Open-domain Dialogue
Authors: Haoqin Tu, Yitong Li, Fei Mi, Zhongliang Yang
Abstract
Incorporating visual knowledge into text-only dialogue systems has become a potential direction to imitate the way humans think, imagine, and communicate. However, existing multimodal dialogue systems are either confined by the scale and quality of available datasets or the coarse concept of visual knowledge. To address these issues, we provide a new paradigm of constructing multimodal dialogues as well as two datasets extended from text-only dialogues under such paradigm (ReSee-WoW, ReSee-DD). We propose to explicitly split the visual knowledge into finer granularity (turn-level'' andentity-level''). To further boost the accuracy and diversity of augmented visual information, we retrieve them from the Internet or a large image dataset. To demonstrate the superiority and universality of the provided visual knowledge, we propose a simple but effective framework ReSee to add visual representation into vanilla dialogue models by modality concatenations. We also conduct extensive experiments and ablations w.r.t. different model configurations and visual knowledge settings. Empirical, encouraging results not only demonstrate the effectiveness of introducing visual knowledge at both entity and turn level but also verify the proposed model ReSee outperforms several state-of-the-art methods on automatic and human evaluations. By leveraging text and vision knowledge, ReSee can produce informative responses with real-world visual concepts.
LLM-empowered Chatbots for Psychiatrist and Patient Simulation: Application and Evaluation
Abstract
Empowering chatbots in the field of mental health is receiving increasing amount of attention, while there still lacks exploration in developing and evaluating chatbots in psychiatric outpatient scenarios. In this work, we focus on exploring the potential of ChatGPT in powering chatbots for psychiatrist and patient simulation. We collaborate with psychiatrists to identify objectives and iteratively develop the dialogue system to closely align with real-world scenarios. In the evaluation experiments, we recruit real psychiatrists and patients to engage in diagnostic conversations with the chatbots, collecting their ratings for assessment. Our findings demonstrate the feasibility of using ChatGPT-powered chatbots in psychiatric scenarios and explore the impact of prompt designs on chatbot behavior and user experience.
Detecting and Mitigating Hallucinations in Multilingual Summarisation
Authors: Yifu Qiu, Yftah Ziser, Anna Korhonen, Edoardo M. Ponti, Shay B. Cohen
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Hallucinations pose a significant challenge to the reliability of neural models for abstractive summarisation. While automatically generated summaries may be fluent, they often lack faithfulness to the original document. This issue becomes even more pronounced in low-resource settings, such as cross-lingual transfer. With the existing faithful metrics focusing on English, even measuring the extent of this phenomenon in cross-lingual settings is hard. To address this, we first develop a novel metric, mFACT, evaluating the faithfulness of non-English summaries, leveraging translation-based transfer from multiple English faithfulness metrics. We then propose a simple but effective method to reduce hallucinations with a cross-lingual transfer, which weighs the loss of each training example by its faithfulness score. Through extensive experiments in multiple languages, we demonstrate that mFACT is the metric that is most suited to detect hallucinations. Moreover, we find that our proposed loss weighting method drastically increases both performance and faithfulness according to both automatic and human evaluation when compared to strong baselines for cross-lingual transfer such as MAD-X. Our code and dataset are available at https://github.com/yfqiu-nlp/mfact-summ.
AxomiyaBERTa: A Phonologically-aware Transformer Model for Assamese
Abstract
Despite their successes in NLP, Transformer-based language models still require extensive computing resources and suffer in low-resource or low-compute settings. In this paper, we present AxomiyaBERTa, a novel BERT model for Assamese, a morphologically-rich low-resource language (LRL) of Eastern India. AxomiyaBERTa is trained only on the masked language modeling (MLM) task, without the typical additional next sentence prediction (NSP) objective, and our results show that in resource-scarce settings for very low-resource languages like Assamese, MLM alone can be successfully leveraged for a range of tasks. AxomiyaBERTa achieves SOTA on token-level tasks like Named Entity Recognition and also performs well on "longer-context" tasks like Cloze-style QA and Wiki Title Prediction, with the assistance of a novel embedding disperser and phonological signals respectively. Moreover, we show that AxomiyaBERTa can leverage phonological signals for even more challenging tasks, such as a novel cross-document coreference task on a translated version of the ECB+ corpus, where we present a new SOTA result for an LRL. Our source code and evaluation scripts may be found at https://github.com/csu-signal/axomiyaberta.
Understanding compositional data augmentation in automatic morphological inflection
Abstract
Data augmentation techniques are widely used in low-resource automatic morphological inflection to address the issue of data sparsity. However, the full implications of these techniques remain poorly understood. In this study, we aim to shed light on the theoretical aspects of the data augmentation strategy StemCorrupt, a method that generates synthetic examples by randomly substituting stem characters in existing gold standard training examples. Our analysis uncovers that StemCorrupt brings about fundamental changes in the underlying data distribution, revealing inherent compositional concatenative structure. To complement our theoretical analysis, we investigate the data-efficiency of StemCorrupt. Through evaluation across a diverse set of seven typologically distinct languages, we demonstrate that selecting a subset of datapoints with both high diversity and high predictive uncertainty significantly enhances the data-efficiency of StemCorrupt compared to competitive baselines. Furthermore, we explore the impact of typological features on the choice of augmentation strategy and find that languages incorporating non-concatenativity, such as morphonological alternations, derive less benefit from synthetic examples with high predictive uncertainty. We attribute this effect to phonotactic violations induced by StemCorrupt, emphasizing the need for further research to ensure optimal performance across the entire spectrum of natural language morphology.
Prompt-Based Monte-Carlo Tree Search for Goal-Oriented Dialogue Policy Planning
Abstract
Planning for goal-oriented dialogue often requires simulating future dialogue interactions and estimating task progress. Many approaches thus consider training neural networks to perform look-ahead search algorithms such as A* search and Monte Carlo Tree Search (MCTS). However, this training often require abundant annotated data, which creates challenges when faced with noisy annotations or low-resource settings. We introduce GDP-Zero, an approach using Open-Loop MCTS to perform goal-oriented dialogue policy planning without any model training. GDP-Zero prompts a large language model to act as a policy prior, value function, user simulator, and system model during the tree search. We evaluate GDP-Zero on the goal-oriented task PersuasionForGood, and find that its responses are preferred over ChatGPT up to 59.32% of the time, and are rated more persuasive than ChatGPT during interactive evaluations.
Automated Metrics for Medical Multi-Document Summarization Disagree with Human Evaluations
Authors: Lucy Lu Wang, Yulia Otmakhova, Jay DeYoung, Thinh Hung Truong, Bailey E. Kuehl, Erin Bransom, Byron C. Wallace
Abstract
Evaluating multi-document summarization (MDS) quality is difficult. This is especially true in the case of MDS for biomedical literature reviews, where models must synthesize contradicting evidence reported across different documents. Prior work has shown that rather than performing the task, models may exploit shortcuts that are difficult to detect using standard n-gram similarity metrics such as ROUGE. Better automated evaluation metrics are needed, but few resources exist to assess metrics when they are proposed. Therefore, we introduce a dataset of human-assessed summary quality facets and pairwise preferences to encourage and support the development of better automated evaluation methods for literature review MDS. We take advantage of community submissions to the Multi-document Summarization for Literature Review (MSLR) shared task to compile a diverse and representative sample of generated summaries. We analyze how automated summarization evaluation metrics correlate with lexical features of generated summaries, to other automated metrics including several we propose in this work, and to aspects of human-assessed summary quality. We find that not only do automated metrics fail to capture aspects of quality as assessed by humans, in many cases the system rankings produced by these metrics are anti-correlated with rankings according to human annotators.
Using Textual Interface to Align External Knowledge for End-to-End Task-Oriented Dialogue Systems
Abstract
Traditional end-to-end task-oriented dialogue systems have been built with a modularized design. However, such design often causes misalignment between the agent response and external knowledge, due to inadequate representation of information. Furthermore, its evaluation metrics emphasize assessing the agent's pre-lexicalization response, neglecting the quality of the completed response. In this work, we propose a novel paradigm that uses a textual interface to align external knowledge and eliminate redundant processes. We demonstrate our paradigm in practice through MultiWOZ-Remake, including an interactive textual interface built for the MultiWOZ database and a correspondingly re-processed dataset. We train an end-to-end dialogue system to evaluate this new dataset. The experimental results show that our approach generates more natural final responses and achieves a greater task success rate compared to the previous models.
LLM-Eval: Unified Multi-Dimensional Automatic Evaluation for Open-Domain Conversations with Large Language Models
Authors: Yen-Ting Lin, Yun-Nung Chen
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
We propose LLM-Eval, a unified multi-dimensional automatic evaluation method for open-domain conversations with large language models (LLMs). Existing evaluation methods often rely on human annotations, ground-truth responses, or multiple LLM prompts, which can be expensive and time-consuming. To address these issues, we design a single prompt-based evaluation method that leverages a unified evaluation schema to cover multiple dimensions of conversation quality in a single model call. We extensively evaluate the performance of LLM-Eval on various benchmark datasets, demonstrating its effectiveness, efficiency, and adaptability compared to state-of-the-art evaluation methods. Our analysis also highlights the importance of choosing suitable LLMs and decoding strategies for accurate evaluation results. LLM-Eval offers a versatile and robust solution for evaluating open-domain conversation systems, streamlining the evaluation process and providing consistent performance across diverse scenarios.
Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models
Authors: Alfonso Amayuelas, Liangming Pan, Wenhu Chen, William Wang
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
This paper investigates the capabilities of Large Language Models (LLMs) in the context of understanding their own knowledge and measuring their uncertainty. We argue this is an important feature for mitigating hallucinations. Specifically, we focus on addressing \textit{known-unknown} questions, characterized by high uncertainty due to the absence of definitive answers. To facilitate our study, we collect a dataset with new Known-Unknown Questions (KUQ) and propose a novel categorization scheme to elucidate the sources of uncertainty. Subsequently, we assess the LLMs' ability to differentiate between known and unknown questions and classify them accordingly. Moreover, we evaluate the quality of their answers in an Open-Ended QA setting. To quantify the uncertainty expressed in the answers, we create a semantic evaluation method that measures the model's accuracy in expressing uncertainty between known vs unknown questions.
Challenges in Context-Aware Neural Machine Translation
Authors: Linghao Jin, Jacqueline He, Jonathan May, Xuezhe Ma
Abstract
Context-aware neural machine translation involves leveraging information beyond sentence-level context to resolve inter-sentential discourse dependencies and improve document-level translation quality, and has given rise to a number of recent techniques. However, despite well-reasoned intuitions, most context-aware translation models show only modest improvements over sentence-level systems. In this work, we investigate several challenges that impede progress within this field, relating to discourse phenomena, context usage, model architectures, and document-level evaluation. To address these problems, we propose a more realistic setting for document-level translation, called paragraph-to-paragraph (para2para) translation, and collect a new dataset of Chinese-English novels to promote future research.
Counterspeeches up my sleeve! Intent Distribution Learning and Persistent Fusion for Intent-Conditioned Counterspeech Generation
Abstract
Counterspeech has been demonstrated to be an efficacious approach for combating hate speech. While various conventional and controlled approaches have been studied in recent years to generate counterspeech, a counterspeech with a certain intent may not be sufficient in every scenario. Due to the complex and multifaceted nature of hate speech, utilizing multiple forms of counter-narratives with varying intents may be advantageous in different circumstances. In this paper, we explore intent-conditioned counterspeech generation. At first, we develop IntentCONAN, a diversified intent-specific counterspeech dataset with 6831 counterspeeches conditioned on five intents, i.e., informative, denouncing, question, positive, and humour. Subsequently, we propose QUARC, a two-stage framework for intent-conditioned counterspeech generation. QUARC leverages vector-quantized representations learned for each intent category along with PerFuMe, a novel fusion module to incorporate intent-specific information into the model. Our evaluation demonstrates that QUARC outperforms several baselines by an average of 10% across evaluation metrics. An extensive human evaluation supplements our hypothesis of better and more appropriate responses than comparative systems.
Asking Clarification Questions to Handle Ambiguity in Open-Domain QA
Authors: Dongryeol Lee, Segwang Kim, Minwoo Lee, Hwanhee Lee, Joonsuk Park, Sang-Woo Lee, Kyomin Jung
Abstract
Ambiguous questions persist in open-domain question answering, because formulating a precise question with a unique answer is often challenging. Previously, Min et al. (2020) have tackled this issue by generating disambiguated questions for all possible interpretations of the ambiguous question. This can be effective, but not ideal for providing an answer to the user. Instead, we propose to ask a clarification question, where the user's response will help identify the interpretation that best aligns with the user's intention. We first present CAMBIGNQ, a dataset consisting of 5,654 ambiguous questions, each with relevant passages, possible answers, and a clarification question. The clarification questions were efficiently created by generating them using InstructGPT and manually revising them as necessary. We then define a pipeline of tasks and design appropriate evaluation metrics. Lastly, we achieve 61.3 F1 on ambiguity detection and 40.5 F1 on clarification-based QA, providing strong baselines for future work.
Learn from Mistakes through Cooperative Interaction with Study Assistant
Abstract
Large language models have demonstrated their ability to self-reflect and refine their generation, which can further improve their performance. However, this feedback mechanism faces challenges such as no guarantee of correctness and the lack of global insight into the model's weaknesses. In this paper, we propose a novel framework, Study Assistant for Large Language Model (SALAM), to aid LLMs in the reflection and refinement process. Motivated by the human study assistant, this framework grades previous responses with the ground truth and collects mistakes in the training phase. During inference, it identifies common misunderstandings based on the mistake collections and provides guidelines for the model to help the model avoid similar mistakes during inference. SALAM is a model-agnostic framework, focusing on providing general feedback and can adapt to any base model. Our evaluation of SALAM on two challenging benchmarks demonstrated a significant improvement over various baselines.
Revealing User Familiarity Bias in Task-Oriented Dialogue via Interactive Evaluation
Authors: Takyoung Kim, Jamin Shin, Young-Ho Kim, Sanghwan Bae, Sungdong Kim
Abstract
Most task-oriented dialogue (TOD) benchmarks assume users that know exactly how to use the system by constraining the user behaviors within the system's capabilities via strict user goals, namely "user familiarity" bias. This data bias deepens when it combines with data-driven TOD systems, as it is impossible to fathom the effect of it with existing static evaluations. Hence, we conduct an interactive user study to unveil how vulnerable TOD systems are against realistic scenarios. In particular, we compare users with 1) detailed goal instructions that conform to the system boundaries (closed-goal) and 2) vague goal instructions that are often unsupported but realistic (open-goal). Our study reveals that conversations in open-goal settings lead to catastrophic failures of the system, in which 92% of the dialogues had significant issues. Moreover, we conduct a thorough analysis to identify distinctive features between the two settings through error annotation. From this, we discover a novel "pretending" behavior, in which the system pretends to handle the user requests even though they are beyond the system's capabilities. We discuss its characteristics and toxicity while emphasizing transparency and a fallback strategy for robust TOD systems.
Narrative XL: A Large-scale Dataset For Long-Term Memory Models
Authors: Arseny Moskvichev, Ky-Vinh Mai
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
Despite their tremendous successes, most large language models do not have any long-term memory mechanisms, which restricts their applications. Overcoming this limitation would not only require changes to the typical transformer architectures or training procedures, but also a dataset on which these new models could be trained and evaluated. We argue that existing resources lack a few key properties, and that at present, there are no naturalistic datasets of sufficient scale to train (and not only evaluate) long-term memory language models. We then present our solution that capitalizes on the advances in short-term memory language models to create such a dataset. Using GPT 3.5, we summarized each scene in 1500 hand-curated books from Project Gutenberg, which resulted in approximately 150 scene-level summaries per book. We then created a number of reading comprehension questions based on these summaries, including three types of multiple-choice scene recognition questions, as well as free-form narrative reconstruction questions. Each book is thus associated with more than 500 reading comprehension questions. Crucially, most questions have a known ``retention demand'', indicating how long-term of a memory is needed to answer it, which should aid long-term memory performance evaluation. We validate our data in three small-scale experiments: one with human labelers, and two with existing language models. We show that our questions 1) adequately represent the source material 2) can be used to diagnose the model's memory capacity 3) are not trivial for modern language models even when the memory demand does not exceed those models' context lengths. Lastly, we provide our code which can be used to further expand the dataset in an automated manner.
Preserving Knowledge Invariance: Rethinking Robustness Evaluation of Open Information Extraction
Authors: Ji Qi, Chuchun Zhang, Xiaozhi Wang, Kaisheng Zeng, Jifan Yu, Jinxin Liu, Jiuding Sun, Yuxiang Chen, Lei How, Juanzi Li, Bin Xu
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
The robustness to distribution changes ensures that NLP models can be successfully applied in the realistic world, especially for information extraction tasks. However, most prior evaluation benchmarks have been devoted to validating pairwise matching correctness, ignoring the crucial measurement of robustness. In this paper, we present the first benchmark that simulates the evaluation of open information extraction models in the real world, where the syntactic and expressive distributions under the same knowledge meaning may drift variously. We design and annotate a large-scale testbed in which each example is a knowledge-invariant clique that consists of sentences with structured knowledge of the same meaning but with different syntactic and expressive forms. By further elaborating the robustness metric, a model is judged to be robust if its performance is consistently accurate on the overall cliques. We perform experiments on typical models published in the last decade as well as a popular large language model, the results show that the existing successful models exhibit a frustrating degradation, with a maximum drop of 23.43 F1 score. Our resources and code will be publicly available.
Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation
Abstract
Gender bias is a significant issue in machine translation, leading to ongoing research efforts in developing bias mitigation techniques. However, most works focus on debiasing of bilingual models without consideration for multilingual systems. In this paper, we specifically target the unambiguous gender bias issue of multilingual machine translation models and propose a new mitigation method based on a novel perspective on the problem. We hypothesize that the gender bias in unambiguous settings is due to the lack of gender information encoded into the non-explicit gender words and devise a scheme to encode correct gender information into their latent embeddings. Specifically, we employ Gender-Aware Contrastive Learning, GACL, based on gender pseudo-labels to encode gender information on the encoder embeddings. Our method is target-language-agnostic and applicable to already trained multilingual machine translation models through post-fine-tuning. Through multilingual evaluation, we show that our approach improves gender accuracy by a wide margin without hampering translation performance. We also observe that incorporated gender information transfers and benefits other target languages regarding gender accuracy. Finally, we demonstrate that our method is applicable and beneficial to models of various sizes.
Assessing Linguistic Generalisation in Language Models: A Dataset for Brazilian Portuguese
Abstract
Much recent effort has been devoted to creating large-scale language models. Nowadays, the most prominent approaches are based on deep neural networks, such as BERT. However, they lack transparency and interpretability, and are often seen as black boxes. This affects not only their applicability in downstream tasks but also the comparability of different architectures or even of the same model trained using different corpora or hyperparameters. In this paper, we propose a set of intrinsic evaluation tasks that inspect the linguistic information encoded in models developed for Brazilian Portuguese. These tasks are designed to evaluate how different language models generalise information related to grammatical structures and multiword expressions (MWEs), thus allowing for an assessment of whether the model has learned different linguistic phenomena. The dataset that was developed for these tasks is composed of a series of sentences with a single masked word and a cue phrase that helps in narrowing down the context. This dataset is divided into MWEs and grammatical structures, and the latter is subdivided into 6 tasks: impersonal verbs, subject agreement, verb agreement, nominal agreement, passive and connectors. The subset for MWEs was used to test BERTimbau Large, BERTimbau Base and mBERT. For the grammatical structures, we used only BERTimbau Large, because it yielded the best results in the MWE task.
Out-of-Distribution Generalization in Text Classification: Past, Present, and Future
Authors: Linyi Yang, Yaoxiao Song, Xuan Ren, Chenyang Lyu, Yidong Wang, Lingqiao Liu, Jindong Wang, Jennifer Foster, Yue Zhang
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
Machine learning (ML) systems in natural language processing (NLP) face significant challenges in generalizing to out-of-distribution (OOD) data, where the test distribution differs from the training data distribution. This poses important questions about the robustness of NLP models and their high accuracy, which may be artificially inflated due to their underlying sensitivity to systematic biases. Despite these challenges, there is a lack of comprehensive surveys on the generalization challenge from an OOD perspective in text classification. Therefore, this paper aims to fill this gap by presenting the first comprehensive review of recent progress, methods, and evaluations on this topic. We furth discuss the challenges involved and potential future research directions. By providing quick access to existing work, we hope this survey will encourage future research in this area.
WYWEB: A NLP Evaluation Benchmark For Classical Chinese
Abstract
To fully evaluate the overall performance of different NLP models in a given domain, many evaluation benchmarks are proposed, such as GLUE, SuperGLUE and CLUE. The fi eld of natural language understanding has traditionally focused on benchmarks for various tasks in languages such as Chinese, English, and multilingua, however, there has been a lack of attention given to the area of classical Chinese, also known as "wen yan wen", which has a rich history spanning thousands of years and holds signifi cant cultural and academic value. For the prosperity of the NLP community, in this paper, we introduce the WYWEB evaluation benchmark, which consists of nine NLP tasks in classical Chinese, implementing sentence classifi cation, sequence labeling, reading comprehension, and machine translation. We evaluate the existing pre-trained language models, which are all struggling with this benchmark. We also introduce a number of supplementary datasets and additional tools to help facilitate further progress on classical Chinese NLU. The github repository is https://github.com/baudzhou/WYWEB.
How Old is GPT?: The HumBEL Framework for Evaluating Language Models using Human Demographic Dat
Authors: Anthony Sicilia, Jennifer C. Gates, Malihe Alikhani
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
While large pre-trained language models (LMs) find greater use across NLP, existing evaluation protocols do not consider how LM language use aligns with particular human demographic groups, which can be an important consideration in conversational AI applications. To remedy this gap, we consider how LM language skills can be measured and compared to human sub-populations. We suggest clinical techniques from Speech Language Pathology, which has well-established norms for acquisition of language skills, organized by (human) age. We conduct evaluation with a domain expert (i.e., a clinically licensed speech language pathologist), and also propose automated techniques to substitute clinical evaluation at scale. We find LM capability varies widely depending on task with GPT-3.5 mimicking the ability of a typical 6-9 year old at tasks requiring inference about word meanings and simultaneously outperforming a typical 21 year old at memorization. GPT-3.5 (InstructGPT) also has trouble with social language use, exhibiting less than 50\% of the tested pragmatic skills. It shows errors in understanding particular word parts-of-speech and associative word relations, among other lexical features. Ultimately, findings reiterate the importance of considering demographic alignment and conversational goals when using these models as public-facing tools. Our framework will be publicly available via code, data, and a python package.
ZeroSCROLLS: A Zero-Shot Benchmark for Long Text Understanding
Authors: Uri Shaham, Maor Ivgi, Avia Efrat, Jonathan Berant, Omer Levy
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Machine Learning (stat.ML)
Abstract
We introduce ZeroSCROLLS, a zero-shot benchmark for natural language understanding over long texts, which contains only test sets, without training or development data. We adapt six tasks from the SCROLLS benchmark, and add four new datasets, including two novel information fusing tasks, such as aggregating the percentage of positive reviews. Using ZeroSCROLLS, we conduct a comprehensive evaluation of both open-source and closed large language models, finding that Claude outperforms ChatGPT, and that GPT-4 achieves the highest average score. However, there is still room for improvement on multiple open challenges in ZeroSCROLLS, such as aggregation tasks, where models struggle to pass the naive baseline. As the state of the art is a moving target, we invite researchers to evaluate their ideas on the live ZeroSCROLLS leaderboard
Exploring Representational Disparities Between Multilingual and Bilingual Translation Models
Abstract
Multilingual machine translation has proven immensely useful for low-resource and zero-shot language pairs. However, language pairs in multilingual models sometimes see worse performance than in bilingual models, especially when translating in a one-to-many setting. To understand why, we examine the geometric differences in the representations from bilingual models versus those from one-to-many multilingual models. Specifically, we evaluate the isotropy of the representations, to measure how well they utilize the dimensions in their underlying vector space. Using the same evaluation data in both models, we find that multilingual model decoder representations tend to be less isotropic than bilingual model decoder representations. Additionally, we show that much of the anisotropy in multilingual decoder representations can be attributed to modeling language-specific information, therefore limiting remaining representational capacity.
Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
Abstract
Fine-tuning on instruction data has been widely validated as an effective practice for implementing chat language models like ChatGPT. Scaling the diversity and quality of such data, although straightforward, stands a great chance of leading to improved performance. This paper aims to improve the upper bound of open-source models further. We first provide a systematically designed, diverse, informative, large-scale dataset of instructional conversations, UltraChat, which does not involve human queries. Our objective is to capture the breadth of interactions that a human might have with an AI assistant and employs a comprehensive framework to generate multi-turn conversation iteratively. UltraChat contains 1.5 million high-quality multi-turn dialogues and covers a wide range of topics and instructions. Our statistical analysis of UltraChat reveals its superiority in various key metrics, including scale, average length, diversity, coherence, etc., solidifying its position as a leading open-source dataset. Building upon UltraChat, we fine-tune a LLaMA model to create a powerful conversational model, UltraLLaMA. Our evaluations indicate that UltraLLaMA consistently outperforms other open-source models, including Vicuna, the previously recognized state-of-the-art open-source model. The dataset and the model will be publicly released\footnote{\url{https://github.com/thunlp/UltraChat}}.
On Learning to Summarize with Large Language Models as References
Authors: Yixin Liu, Alexander R. Fabbri, Pengfei Liu, Dragomir Radev, Arman Cohan
Abstract
Recent studies have found that summaries generated by large language models (LLMs) are favored by human annotators over the original reference summaries in commonly used summarization datasets. Therefore, we investigate a new learning paradigm of text summarization models that considers the LLMs as the reference or the gold-standard oracle on commonly used summarization datasets such as the CNN/DailyMail dataset. To examine the standard practices that are aligned with the new learning setting, we propose a novel training method that is based on contrastive learning with LLMs as a summarization quality evaluator. For this reward-based training method, we investigate two different methods of utilizing LLMs for summary quality evaluation, namely GPTScore and GPTRank. Our experiments on the CNN/DailyMail dataset demonstrate that smaller summarization models trained by our proposed method can achieve performance equal to or surpass that of the reference LLMs, as evaluated by the LLMs themselves. This underscores the efficacy of our proposed paradigm in enhancing model performance over the standard maximum likelihood estimation (MLE) training method, and its efficiency since it only requires a small budget to access the LLMs. We release the training scripts, model outputs, and LLM-based evaluation results to facilitate future studies.
FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
Authors: Sewon Min, Kalpesh Krishna, Xinxi Lyu, Mike Lewis, Wen-tau Yih, Pang Wei Koh, Mohit Iyyer, Luke Zettlemoyer, Hannaneh Hajishirzi
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Evaluating the factuality of long-form text generated by large language models (LMs) is non-trivial because (1) generations often contain a mixture of supported and unsupported pieces of information, making binary judgments of quality inadequate, and (2) human evaluation is time-consuming and costly. In this paper, we introduce FActScore (Factual precision in Atomicity Score), a new evaluation that breaks a generation into a series of atomic facts and computes the percentage of atomic facts supported by a reliable knowledge source. We conduct an extensive human evaluation to obtain FActScores of people biographies generated by several state-of-the-art commercial LMs -- InstructGPT, ChatGPT, and the retrieval-augmented PerplexityAI -- and report new analysis demonstrating the need for such a fine-grained score (e.g., ChatGPT only achieves 58%). Since human evaluation is costly, we also introduce an automated model that estimates FActScore, using retrieval and a strong language model, with less than a 2% error rate. Finally, we use this automated metric to evaluate 6,500 generations from a new set of 13 recent LMs that would have cost $26K if evaluated by humans, with various findings: GPT-4 and ChatGPT are more factual than public models, and Vicuna and Alpaca are some of the best public models.
Learning to Generate Novel Scientific Directions with Contextualized Literature-based Discovery
Authors: Qingyun Wang, Doug Downey, Heng Ji, Tom Hope
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
Abstract
Literature-Based Discovery (LBD) aims to discover new scientific knowledge by mining papers and generating hypotheses. Standard LBD is limited to predicting pairwise relations between discrete concepts (e.g., drug-disease links). LBD also ignores critical contexts like experimental settings (e.g., a specific patient population where a drug is evaluated) and background knowledge and motivations that human scientists consider (e.g., to find a drug candidate without specific side effects). We address these limitations with a novel formulation of contextualized-LBD (C-LBD): generating scientific hypotheses in natural language, while grounding them in a context that controls the hypothesis search space. We present a new modeling framework using retrieval of ``inspirations'' from a heterogeneous network of citations and knowledge graph relations, and create a new dataset derived from papers. In automated and human evaluations, our models improve over baselines, including powerful large language models (LLMs), but also reveal challenges on the road to building machines that generate new scientific knowledge.
R2H: Building Multimodal Navigation Helpers that Respond to Help
Authors: Yue Fan, Kaizhi Zheng, Jing Gu, Xin Eric Wang
Abstract
The ability to assist humans during a navigation task in a supportive role is crucial for intelligent agents. Such agents, equipped with environment knowledge and conversational abilities, can guide individuals through unfamiliar terrains by generating natural language responses to their inquiries, grounded in the visual information of their surroundings. However, these multimodal conversational navigation helpers are still underdeveloped. This paper proposes a new benchmark, Respond to Help (R2H), to build multimodal navigation helpers that can respond to help, based on existing dialog-based embodied datasets. R2H mainly includes two tasks: (1) Respond to Dialog History (RDH), which assesses the helper agent's ability to generate informative responses based on a given dialog history, and (2) Respond during Interaction (RdI), which evaluates the helper agent's ability to maintain effective and consistent cooperation with a task performer agent during navigation in real-time. Furthermore, we propose a novel task-oriented multimodal response generation model that can see and respond, named SeeRee, as the navigation helper to guide the task performer in embodied tasks. Through both automatic and human evaluations, we show that SeeRee produces more effective and informative responses than baseline methods in assisting the task performer with different navigation tasks. Project website: https://sites.google.com/view/respond2help/home.
Weakly-Supervised Learning of Visual Relations in Multimodal Pretraining
Authors: Emanuele Bugliarello, Aida Nematzadeh, Lisa Anne Hendricks
Subjects: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
Abstract
Recent work in vision-and-language pretraining has investigated supervised signals from object detection data to learn better, fine-grained multimodal representations. In this work, we take a step further and explore how we add supervision from small-scale visual relation data. In particular, we propose two pretraining approaches to contextualise visual entities in a multimodal setup. With verbalised scene graphs, we transform visual relation triplets into structured captions, and treat them as additional views of images. With masked relation prediction, we further encourage relating entities from visually masked contexts. When applied to strong baselines pretrained on large amounts of Web data, zero-shot evaluations on both coarse-grained and fine-grained tasks show the efficacy of our methods in learning multimodal representations from weakly-supervised relations data.
INSTRUCTSCORE: Towards Explainable Text Generation Evaluation with Automatic Feedback
Authors: Wenda Xu, Danqing Wang, Liangming Pan, Zhenqiao Song, Markus Freitag, William Yang Wang, Lei Li
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Abstract
The field of automatic evaluation of text generation made tremendous progress in the last few years. In particular, since the advent of neural metrics, like COMET, BLEURT, and SEScore2, the newest generation of metrics show a high correlation with human judgment. Unfortunately, quality scores generated with neural metrics are not interpretable, and it is unclear which part of the generation output is criticized by the metrics. To address this limitation, we present INSTRUCTSCORE, an open-source, explainable evaluation metric for text generation. By harnessing both explicit human instruction and the implicit knowledge of GPT4, we fine-tune a LLAMA model to create an evaluative metric that can produce a diagnostic report aligned with human judgment. We evaluate INSTRUCTSCORE on the WMT22 Zh-En translation task, where our 7B model surpasses other LLM-based baselines, including those based on 175B GPT3. Impressively, our INSTRUCTSCORE, even without direct supervision from human-rated data, achieves performance levels on par with state-of-the-art metrics like COMET22, which was fine-tuned on human ratings.
Query Rewriting for Retrieval-Augmented Large Language Models
Authors: Xinbei Ma, Yeyun Gong, Pengcheng He, Hai Zhao, Nan Duan
Abstract
Large Language Models (LLMs) play a powerful \textit{Reader} of the \textit{Retrieve-then-Read} pipeline, making great progress in knowledge-based open-domain tasks. This work introduces a new framework, \textit{Rewrite-Retrieve-Read} that improves the retrieval-augmented method from the perspective of the query rewriting. Prior studies mostly contribute to adapt the retriever or stimulate the reader. Different from them, our approach pay attention of the query adaptation. Because the original query can not be always optimal to retrieve for the LLM, especially in the real world.(1) We first prompt an LLM to rewrite the queries, then conduct retrieval-augmented reading. (2) We further apply a small language model as a trainable rewriter, which rewrite the search query to cater to the frozen retriever and the LLM reader. To fine-tune the rewriter, we first use a pseudo data to conduct supervised warm-up training. Then the \textit{Retrieve-then-Read} pipeline is modeled as a reinforcement learning context. The rewriter is further trained as a policy model by maximize the reward of the pipeline performance. Evaluation is performed on two downstream tasks, open-domain QA and multiple choice. Our framework is proved effective and scalable.
LLM-powered Data Augmentation for Enhanced Crosslingual Performance
Authors: Chenxi Whitehouse, Monojit Choudhury, Alham Fikri Aji
Abstract
This paper aims to explore the potential of leveraging Large Language Models (LLMs) for data augmentation in crosslingual commonsense reasoning datasets, where the available training data is extremely limited. To achieve this, we employ several LLMs including Dolly-v2, StableVicuna, ChatGPT, and GPT-4 to augment three datasets: XCOPA, XWinograd, and XStoryCloze. Subsequently, we assess the effectiveness of fine-tuning smaller crosslingual models, mBERT and XLMR, using the synthesised data. We compare the performance of training with data generated in English and target languages, as well as translating the English-generated data into the target languages. Our experiments reveal the overall advantages of incorporating data generated by LLMs. Training on synthetic data generated by GPT-4, whether English or multilingual, improves performance consistently compared to the baseline. Other models also exhibit an overall increase in performance, however, their effectiveness decreases in some settings. We also ask native speakers to evaluate the naturalness and logical soundness of the generated examples for different languages. Human evaluation reveals that LLMs like ChatGPT and GPT-4 excel at generating natural text in most languages, except a few such as Tamil. Moreover, ChatGPT trails behind in generating plausible alternatives in comparison to the original dataset, while GPT-4 demonstrates competitive logic consistency in the synthesised data.
Evaluation of African American Language Bias in Natural Language Generation
Abstract
We evaluate how well LLMs understand African American Language (AAL) in comparison to their performance on White Mainstream English (WME), the encouraged "standard" form of English taught in American classrooms. We measure LLM performance using automatic metrics and human judgments for two tasks: a counterpart generation task, where a model generates AAL (or WME) given WME (or AAL), and a masked span prediction (MSP) task, where models predict a phrase that was removed from their input. Our contributions include: (1) evaluation of six pre-trained, large language models on the two language generation tasks; (2) a novel dataset of AAL text from multiple contexts (social media, hip-hop lyrics, focus groups, and linguistic interviews) with human-annotated counterparts in WME; and (3) documentation of model performance gaps that suggest bias and identification of trends in lack of understanding of AAL features.
WikiChat: A Few-Shot LLM-Based Chatbot Grounded with Wikipedia
Authors: Sina J. Semnani, Violet Z. Yao, Heidi C. Zhang, Monica S. Lam
Abstract
Despite recent advances in Large Language Models (LLMs), users still cannot trust the information provided in their responses. LLMs cannot speak accurately about events that occurred after their training, which are often topics of great interest to users, and, as we show in this paper, they are highly prone to hallucination when talking about less popular (tail) topics. This paper presents WikiChat, a few-shot LLM-based chatbot that is grounded with live information from Wikipedia. Through many iterations of experimentation, we have crafte a pipeline based on information retrieval that (1) uses LLMs to suggest interesting and relevant facts that are individually verified against Wikipedia, (2) retrieves additional up-to-date information, and (3) composes coherent and engaging time-aware responses. We propose a novel hybrid human-and-LLM evaluation methodology to analyze the factuality and conversationality of LLM-based chatbots. We focus on evaluating important but previously neglected issues such as conversing about recent and tail topics. We evaluate WikiChat against strong fine-tuned and LLM-based baselines across a diverse set of conversation topics. We find that WikiChat outperforms all baselines in terms of the factual accuracy of its claims, by up to 12.1%, 28.3% and 32.7% on head, recent and tail topics, while matching GPT-3.5 in terms of providing natural, relevant, non-repetitive and informational responses.
RET-LLM: Towards a General Read-Write Memory for Large Language Models
Authors: Ali Modarressi, Ayyoob Imani, Mohsen Fayyaz, Hinrich Schütze
Abstract
Large language models (LLMs) have significantly advanced the field of natural language processing (NLP) through their extensive parameters and comprehensive data utilization. However, existing LLMs lack a dedicated memory unit, limiting their ability to explicitly store and retrieve knowledge for various tasks. In this paper, we propose RET-LLM a novel framework that equips LLMs with a general write-read memory unit, allowing them to extract, store, and recall knowledge from the text as needed for task performance. Inspired by Davidsonian semantics theory, we extract and save knowledge in the form of triplets. The memory unit is designed to be scalable, aggregatable, updatable, and interpretable. Through qualitative evaluations, we demonstrate the superiority of our proposed framework over baseline approaches in question answering tasks. Moreover, our framework exhibits robust performance in handling temporal-based question answering tasks, showcasing its ability to effectively manage time-dependent information.
ChatCoT: Tool-Augmented Chain-of-Thought Reasoning on\ Chat-based Large Language Models
Authors: Zhipeng Chen, Kun Zhou, Beichen Zhang, Zheng Gong, Wayne Xin Zhao, Ji-Rong Wen
Abstract
Although large language models (LLMs) have achieved excellent performance in a variety of evaluation benchmarks, they still struggle in complex reasoning tasks which require specific knowledge and multi-hop reasoning. To improve the reasoning abilities, we propose \textbf{ChatCoT}, a tool-augmented chain-of-thought reasoning framework for chat-based LLMs. In ChatCoT, we model the chain-of-thought~(CoT) reasoning as multi-turn conversations, to utilize tools in a more natural way through chatting. At each turn, LLMs can either interact with tools or perform the reasoning. %framework to decompose reasoning and manipulate tools for complex reasoning tasks via multi-round dialogue. %Our idea is to decompose the complex problems into several sub-problems and solve these sub-problems with the help of external tools. Our approach can effectively leverage the multi-turn conversation ability of chat-based LLMs, and integrate the thought chain following and tools manipulation in a unified way. Specially, we initialize the early turns of the conversation by the tools, tasks and reasoning format, and propose an iterative \emph{tool-augmented reasoning} step to perform step-by-step tool-augmented reasoning. The experiment results on two complex reasoning datasets (MATH and HotpotQA) have shown the effectiveness of ChatCoT on complex reasoning tasks, achieving a 6.8\% relative improvement over the state-of-the-art baseline. %and ChatCoT can achieve performance on Math. Our code and data are available at: \url{https://github.com/RUCAIBOX/ChatCoT}.
Ties Matter: Modifying Kendall's Tau for Modern Metric Meta-Evaluation
Authors: Daniel Deutsch, George Foster, Markus Freitag
Abstract
Kendall's tau is frequently used to meta-evaluate how well machine translation (MT) evaluation metrics score individual translations. Its focus on pairwise score comparisons is intuitive but raises the question of how ties should be handled, a gray area that has motivated different variants in the literature. We demonstrate that, in settings like modern MT meta-evaluation, existing variants have weaknesses arising from their handling of ties, and in some situations can even be gamed. We propose a novel variant that gives metrics credit for correctly predicting ties, as well as an optimization procedure that automatically introduces ties into metric scores, enabling fair comparison between metrics that do and do not predict ties. We argue and provide experimental evidence that these modifications lead to fairer Kendall-based assessments of metric performance.
Anchor Prediction: Automatic Refinement of Internet Links
Authors: Nelson F. Liu, Kenton Lee, Kristina Toutanova
Subjects: Computation and Language (cs.CL); Information Retrieval (cs.IR)
Abstract
Internet links enable users to deepen their understanding of a topic by providing convenient access to related information. However, the majority of links are unanchored -- they link to a target webpage as a whole, and readers may expend considerable effort localizing the specific parts of the target webpage that enrich their understanding of the link's source context. To help readers effectively find information in linked webpages, we introduce the task of anchor prediction, where the goal is to identify the specific part of the linked target webpage that is most related to the source linking context. We release the AuthorAnchors dataset, a collection of 34K naturally-occurring anchored links, which reflect relevance judgments by the authors of the source article. To model reader relevance judgments, we annotate and release ReaderAnchors, an evaluation set of anchors that readers find useful. Our analysis shows that effective anchor prediction often requires jointly reasoning over lengthy source and target webpages to determine their implicit relations and identify parts of the target webpage that are related but not redundant. We benchmark a performant T5-based ranking approach to establish baseline performance on the task, finding ample room for improvement.
APPLS: A Meta-evaluation Testbed for Plain Language Summarization
Authors: Yue Guo, Tal August, Gondy Leroy, Trevor Cohen, Lucy Lu Wang
Abstract
While there has been significant development of models for Plain Language Summarization (PLS), evaluation remains a challenge. This is in part because PLS involves multiple, interrelated language transformations (e.g., adding background explanations, removing specialized terminology). No metrics are explicitly engineered for PLS, and the suitability of other text generation evaluation metrics remains unclear. To address these concerns, our study presents a granular meta-evaluation testbed, APPLS, designed to evaluate existing metrics for PLS. Drawing on insights from previous research, we define controlled perturbations for our testbed along four criteria that a metric of plain language should capture: informativeness, simplification, coherence, and faithfulness. Our analysis of metrics using this testbed reveals that current metrics fail to capture simplification, signaling a crucial gap. In response, we introduce POMME, a novel metric designed to assess text simplification in PLS. We demonstrate its correlation with simplification perturbations and validate across a variety of datasets. Our research contributes the first meta-evaluation testbed for PLS and a comprehensive evaluation of existing metrics, offering insights with relevance to other text generation tasks.
Keyword: text generation
Reducing Sensitivity on Speaker Names for Text Generation from Dialogues
INSTRUCTSCORE: Towards Explainable Text Generation Evaluation with Automatic Feedback
QTSumm: A New Benchmark for Query-Focused Table Summarization
APPLS: A Meta-evaluation Testbed for Plain Language Summarization
Keyword: machine translation
GATology for Linguistics: What Syntactic Dependencies It Knows
Syntactic Knowledge via Graph Attention with BERT in Machine Translation
Neural Machine Translation for Code Generation
Non-parametric, Nearest-neighbor-assisted Fine-tuning for Neural Machine Translation
Optimizing Non-Autoregressive Transformers with Contrastive Learning
TeCS: A Dataset and Benchmark for Tense Consistency of Machine Translation
Challenges in Context-Aware Neural Machine Translation
Condensing Multilingual Knowledge with Lightweight Language-Specific Modules
Sāmayik: A Benchmark and Dataset for English-Sanskrit Translation
Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation
Improving speech translation by fusing speech and text
In-context Example Selection for Machine Translation Using Multiple Features
When Does Monolingual Data Help Multilingual Translation: The Role of Domain and Model Scale
WYWEB: A NLP Evaluation Benchmark For Classical Chinese
Beyond Shared Vocabulary: Increasing Representational Word Similarities across Languages for Multilingual Machine Translation
Exploring Representational Disparities Between Multilingual and Bilingual Translation Models
Multilingual Large Language Models Are Not (Yet) Code-Switchers
Revisiting Machine Translation for Cross-lingual Classification
Pixel Representations for Multilingual Translation and Data-efficient Cross-lingual Transfer
Ties Matter: Modifying Kendall's Tau for Modern Metric Meta-Evaluation
Empowering LLM-based Machine Translation with Cultural Awareness
Keyword: non-autoregressive
Optimizing Non-Autoregressive Transformers with Contrastive Learning
Keyword: abstractive summarization
Abstractive Text Summarization Using the BRIO Training Paradigm
Keyword: factual
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
Multimodal Automated Fact-Checking: A Survey
Improving Classifier Robustness through Active Generation of Pairwise Counterfactuals
Polyglot or Not? Measuring Multilingual Encyclopedic Knowledge Retrieval from Foundation Language Models
Improving Language Models via Plug-and-Play Retrieval Feedback
IfQA: A Dataset for Open-domain Question Answering under Counterfactual Presuppositions
Evaluating Factual Consistency of Summaries with Large Language Models
To Copy Rather Than Memorize: A Vertical Learning Paradigm for Knowledge Graph Completion
Question Answering as Programming for Solving Time-Sensitive Questions
ManiTweet: A New Benchmark for Identifying Manipulation of News on Social Media
FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
WikiChat: A Few-Shot LLM-Based Chatbot Grounded with Wikipedia
WebIE: Faithful and Robust Information Extraction on the Web
USB: A Unified Summarization Benchmark Across Tasks and Domains
Improving Factuality and Reasoning in Language Models through Multiagent Debate
Evaluating and Modeling Attribution for Cross-Lingual Question Answering
Keyword: knowledge distillation
One-stop Training of Multiple Capacity Models for Multilingual Machine Translation
Keyword: Hallucination
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
How Language Model Hallucinations Can Snowball
Detecting and Mitigating Hallucinations in Multilingual Summarisation
Mitigating Language Model Hallucination with Interactive Question-Knowledge Alignment
Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models
PaD: Program-aided Distillation Specializes Large Models in Reasoning
mmT5: Modular Multilingual Pre-Training Solves Source Language Hallucinations
WikiChat: A Few-Shot LLM-Based Chatbot Grounded with Wikipedia
Improving Factuality and Reasoning in Language Models through Multiagent Debate
Keyword: evaluation
Element-aware Summarization with Large Language Models: Expert-aligned Evaluation and Chain-of-Thought Method
clembench: Using Game Play to Evaluate Chat-Optimized Language Models as Conversational Agents
Look-back Decoding for Open-Ended Text Generation
Neural Machine Translation for Code Generation
The Grammar and Syntax Based Corpus Analysis Tool For The Ukrainian Language
Open-world Semi-supervised Generalized Relation Discovery Aligned in a Real-world Setting
EntRED: Benchmarking Relation Extraction with Fewer Shortcuts
Cross-Attention is Not Enough: Incongruity-Aware Multimodal Sentiment Analysis and Emotion Recognition
ReSee: Responding through Seeing Fine-grained Visual Knowledge in Open-domain Dialogue
turn-level'' andentity-level''). To further boost the accuracy and diversity of augmented visual information, we retrieve them from the Internet or a large image dataset. To demonstrate the superiority and universality of the provided visual knowledge, we propose a simple but effective framework ReSee to add visual representation into vanilla dialogue models by modality concatenations. We also conduct extensive experiments and ablations w.r.t. different model configurations and visual knowledge settings. Empirical, encouraging results not only demonstrate the effectiveness of introducing visual knowledge at both entity and turn level but also verify the proposed model ReSee outperforms several state-of-the-art methods on automatic and human evaluations. By leveraging text and vision knowledge, ReSee can produce informative responses with real-world visual concepts.LLM-empowered Chatbots for Psychiatrist and Patient Simulation: Application and Evaluation
Detecting and Mitigating Hallucinations in Multilingual Summarisation
AxomiyaBERTa: A Phonologically-aware Transformer Model for Assamese
Understanding compositional data augmentation in automatic morphological inflection
Prompt-Based Monte-Carlo Tree Search for Goal-Oriented Dialogue Policy Planning
Automated Metrics for Medical Multi-Document Summarization Disagree with Human Evaluations
Using Textual Interface to Align External Knowledge for End-to-End Task-Oriented Dialogue Systems
LLM-Eval: Unified Multi-Dimensional Automatic Evaluation for Open-Domain Conversations with Large Language Models
Knowledge of Knowledge: Exploring Known-Unknowns Uncertainty with Large Language Models
Challenges in Context-Aware Neural Machine Translation
Counterspeeches up my sleeve! Intent Distribution Learning and Persistent Fusion for Intent-Conditioned Counterspeech Generation
Asking Clarification Questions to Handle Ambiguity in Open-Domain QA
Learn from Mistakes through Cooperative Interaction with Study Assistant
Revealing User Familiarity Bias in Task-Oriented Dialogue via Interactive Evaluation
Narrative XL: A Large-scale Dataset For Long-Term Memory Models
Preserving Knowledge Invariance: Rethinking Robustness Evaluation of Open Information Extraction
Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translation
Assessing Linguistic Generalisation in Language Models: A Dataset for Brazilian Portuguese
Out-of-Distribution Generalization in Text Classification: Past, Present, and Future
WYWEB: A NLP Evaluation Benchmark For Classical Chinese
How Old is GPT?: The HumBEL Framework for Evaluating Language Models using Human Demographic Dat
ZeroSCROLLS: A Zero-Shot Benchmark for Long Text Understanding
Exploring Representational Disparities Between Multilingual and Bilingual Translation Models
Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
On Learning to Summarize with Large Language Models as References
FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation
Learning to Generate Novel Scientific Directions with Contextualized Literature-based Discovery
R2H: Building Multimodal Navigation Helpers that Respond to Help
Weakly-Supervised Learning of Visual Relations in Multimodal Pretraining
INSTRUCTSCORE: Towards Explainable Text Generation Evaluation with Automatic Feedback
Query Rewriting for Retrieval-Augmented Large Language Models
LLM-powered Data Augmentation for Enhanced Crosslingual Performance
Evaluation of African American Language Bias in Natural Language Generation
WikiChat: A Few-Shot LLM-Based Chatbot Grounded with Wikipedia
RET-LLM: Towards a General Read-Write Memory for Large Language Models
ChatCoT: Tool-Augmented Chain-of-Thought Reasoning on\ Chat-based Large Language Models
Ties Matter: Modifying Kendall's Tau for Modern Metric Meta-Evaluation
Anchor Prediction: Automatic Refinement of Internet Links
APPLS: A Meta-evaluation Testbed for Plain Language Summarization