Existing metrics for evaluating the quality of automatically generated questions such as BLEU, ROUGE, BERTScore, and BLEURT compare the reference and predicted questions, providing a high score when there is a considerable lexical overlap or semantic similarity between the candidate and the reference questions. This approach has two major shortcomings. First, we need expensive human-provided reference questions. Second, it penalises valid questions that may not have high lexical or semantic similarity to the reference questions. In this paper, we propose a new metric, RQUGE, based on the answerability of the candidate question given the context. The metric consists of a question-answering and a span scorer modules, using pre-trained models from existing literature, thus it can be used without any further training. We demonstrate that RQUGE has a higher correlation with human judgment without relying on the reference question. Additionally, RQUGE is shown to be more robust to several adversarial corruptions. Furthermore, we illustrate that we can significantly improve the performance of QA models on out-of-domain datasets by fine-tuning on synthetic data generated by a question generation model and re-ranked by RQUGE.
相關內容
Modern language modeling tasks are often underspecified: for a given token prediction, many words may satisfy the user's intent of producing natural language at inference time, however only one word would minimize the task's loss function at training time. We provide a simple yet plausible causal mechanism describing the role underspecification plays in the generation of spurious correlations. Despite its simplicity, our causal model directly informs the development of two lightweight black-box evaluation methods, that we apply to gendered pronoun resolution tasks on a wide range of LLMs to 1) aid in the detection of inference-time task underspecification by exploiting 2) previously unreported gender vs. time and gender vs. location spurious correlations on LLMs with a range of A) sizes: from BERT-base to GPT 3.5, B) pre-training objectives: from masked & autoregressive language modeling to a mixture of these objectives, and C) training stages: from pre-training only to reinforcement learning from human feedback (RLHF). Code and open-source demos available at https: //github.com/2dot71mily/sib_paper.
This paper studies the possibilities made open by the use of Lazy Clause Generation (LCG) based approaches to Constraint Programming (CP) for tackling sequential classical planning. We propose a novel CP model based on seminal ideas on so-called lifted causal encodings for planning as satisfiability, that does not require grounding, as choosing groundings for functions and action schemas becomes an integral part of the problem of designing valid plans. This encoding does not require encoding frame axioms, and does not explicitly represent states as decision variables for every plan step. We also present a propagator procedure that illustrates the possibilities of LCG to widen the kind of inference methods considered to be feasible in planning as (iterated) CSP solving. We test encodings and propagators over classic IPC and recently proposed benchmarks for lifted planning, and report that for planning problem instances requiring fewer plan steps our methods compare very well with the state-of-the-art in optimal sequential planning.
Recent image harmonization methods have demonstrated promising results. However, due to their heavy reliance on a large number of composite images, these works are expensive in the training phase and often fail to generalize to unseen images. In this paper, we draw lessons from human behavior and come up with a zero-shot image harmonization method. Specifically, in the harmonization process, a human mainly utilizes his long-term prior on harmonious images and makes a composite image close to that prior. To imitate that, we resort to pretrained generative models for the prior of natural images. For the guidance of the harmonization direction, we propose an Attention-Constraint Text which is optimized to well illustrate the image environments. Some further designs are introduced for preserving the foreground content structure. The resulting framework, highly consistent with human behavior, can achieve harmonious results without burdensome training. Extensive experiments have demonstrated the effectiveness of our approach, and we have also explored some interesting applications.
Evaluating Natural Language Generation (NLG) outputs is crucial but laborious and expensive. While various automatic NLG assessment methods have been proposed, they often are quite task-specific and have to be engineered with a particular domain and attribute in mind. In this work, we propose a robust zero-shot approach to NLG evaluation using pairwise comparative judgment with open-source Large Language Models (LLMs). The motivation for this approach is that even as humans, it is easier to determine which of two options are better, than it is to independently objectively score each option. We use this insight and leverage the emergent abilities of LLMs, where we probe FlanT5 to determine which of two candidate responses is better, rather than assigning absolute scores. Our results demonstrate that comparative assessment is a more effective approach than absolute scoring, enabling smaller open-source LLMs to achieve comparable performance to larger public access APIs. We evaluate systems on both summary evaluation and dialogue response generation, and show that opensource LLMs can lead to good correlations with human scores for a range of different attributes.
Off-policy evaluation (OPE) aims to estimate the benefit of following a counterfactual sequence of actions, given data collected from executed sequences. However, existing OPE estimators often exhibit high bias and high variance in problems involving large, combinatorial action spaces. We investigate how to mitigate this issue using factored action spaces i.e. expressing each action as a combination of independent sub-actions from smaller action spaces. This approach facilitates a finer-grained analysis of how actions differ in their effects. In this work, we propose a new family of "decomposed" importance sampling (IS) estimators based on factored action spaces. Given certain assumptions on the underlying problem structure, we prove that the decomposed IS estimators have less variance than their original non-decomposed versions, while preserving the property of zero bias. Through simulations, we empirically verify our theoretical results, probing the validity of various assumptions. Provided with a technique that can derive the action space factorisation for a given problem, our work shows that OPE can be improved "for free" by utilising this inherent problem structure.
Before deploying a language model (LM) within a given domain, it is important to measure its tendency to generate factually incorrect information in that domain. Existing factual generation evaluation methods focus on facts sampled from the LM itself, and thus do not control the set of evaluated facts and might under-represent rare and unlikely facts. We propose FACTOR: Factual Assessment via Corpus TransfORmation, a scalable approach for evaluating LM factuality. FACTOR automatically transforms a factual corpus of interest into a benchmark evaluating an LM's propensity to generate true facts from the corpus vs. similar but incorrect statements. We use our framework to create two benchmarks: Wiki-FACTOR and News-FACTOR. We show that: (i) our benchmark scores increase with model size and improve when the LM is augmented with retrieval; (ii) benchmark score correlates with perplexity, but the two metrics do not always agree on model ranking; and (iii) when perplexity and benchmark score disagree, the latter better reflects factuality in open-ended generation, as measured by human annotators. We make our data and code publicly available in //github.com/AI21Labs/factor.
While summarization has been extensively researched in natural language processing (NLP), cross-lingual cross-temporal summarization (CLCTS) is a largely unexplored area that has the potential to improve cross-cultural accessibility and understanding. This paper comprehensively addresses the CLCTS task, including dataset creation, modeling, and evaluation. We build the first CLCTS corpus, leveraging historical fictive texts and Wikipedia summaries in English and German, and examine the effectiveness of popular transformer end-to-end models with different intermediate finetuning tasks. Additionally, we explore the potential of ChatGPT for CLCTS as a summarizer and an evaluator. Overall, we report evaluations from humans, ChatGPT, and several recent automatic evaluation metrics where we find that our intermediate task finetuned end-to-end models generate bad to moderate quality summaries; ChatGPT as a summarizer (without any finetuning) provides moderate to good quality outputs and as an evaluator correlates moderately with human evaluations but is prone to giving lower scores. ChatGPT also seems very adept at normalizing historical text and outperforms context-unaware spelling normalization tools such as Norma. We finally test ChatGPT in a scenario with adversarially attacked and unseen source documents and find that ChatGPT profits from its prior knowledge to a certain degree, with better performances for omission and entity swap than negation against its prior knowledge. This benefit inflates its assessed quality as ChatGPT performs slightly worse for unseen source documents compared to seen documents. We additionally introspect our models' performances to find that longer, older and more complex source texts (all of which are more characteristic for historical language variants) are harder to summarize for all models, indicating the difficulty of the CLCTS task.
Existing evaluation metrics for natural language generation (NLG) tasks face the challenges on generalization ability and interpretability. Specifically, most of the well-performed metrics are required to train on evaluation datasets of specific NLG tasks and evaluation dimensions, which may cause over-fitting to task-specific datasets. Furthermore, existing metrics only provide an evaluation score for each dimension without revealing the evidence to interpret how this score is obtained. To deal with these challenges, we propose a simple yet effective metric called DecompEval. This metric formulates NLG evaluation as an instruction-style question answering task and utilizes instruction-tuned pre-trained language models (PLMs) without training on evaluation datasets, aiming to enhance the generalization ability. To make the evaluation process more interpretable, we decompose our devised instruction-style question about the quality of generated texts into the subquestions that measure the quality of each sentence. The subquestions with their answers generated by PLMs are then recomposed as evidence to obtain the evaluation result. Experimental results show that DecompEval achieves state-of-the-art performance in untrained metrics for evaluating text summarization and dialogue generation, which also exhibits strong dimension-level / task-level generalization ability and interpretability.
User engagement is a critical metric for evaluating the quality of open-domain dialogue systems. Prior work has focused on conversation-level engagement by using heuristically constructed features such as the number of turns and the total time of the conversation. In this paper, we investigate the possibility and efficacy of estimating utterance-level engagement and define a novel metric, {\em predictive engagement}, for automatic evaluation of open-domain dialogue systems. Our experiments demonstrate that (1) human annotators have high agreement on assessing utterance-level engagement scores; (2) conversation-level engagement scores can be predicted from properly aggregated utterance-level engagement scores. Furthermore, we show that the utterance-level engagement scores can be learned from data. These scores can improve automatic evaluation metrics for open-domain dialogue systems, as shown by correlation with human judgements. This suggests that predictive engagement can be used as a real-time feedback for training better dialogue models.
We introduce the first system towards the novel task of answering complex multisentence recommendation questions in the tourism domain. Our solution uses a pipeline of two modules: question understanding and answering. For question understanding, we define an SQL-like query language that captures the semantic intent of a question; it supports operators like subset, negation, preference and similarity, which are often found in recommendation questions. We train and compare traditional CRFs as well as bidirectional LSTM-based models for converting a question to its semantic representation. We extend these models to a semisupervised setting with partially labeled sequences gathered through crowdsourcing. We find that our best model performs semi-supervised training of BiDiLSTM+CRF with hand-designed features and CCM(Chang et al., 2007) constraints. Finally, in an end to end QA system, our answering component converts our question representation into queries fired on underlying knowledge sources. Our experiments on two different answer corpora demonstrate that our system can significantly outperform baselines with up to 20 pt higher accuracy and 17 pt higher recall.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191