ACL 2023 论文列表

Continual Learning (CL) methods focus on accumulating knowledge over time while avoiding catastrophic forgetting. Recently, Wortsman et al. (2020) proposed a CL method, SupSup, which uses a randomly initialized, fixed base network (model) and finds a supermask for each new task that selectively keeps or removes each weight to produce a subnetwork. They prevent forgetting as the network weights are not being updated. Although there is no forgetting, the performance of SupSup is sub-optimal because fixed weights restrict its representational power. Furthermore, there is no accumulation or transfer of knowledge inside the model when new tasks are learned. Hence, we propose ExSSNeT (Exclusive Supermask SubNetwork Training), that performs exclusive and non-overlapping subnetwork weight training. This avoids conflicting updates to the shared weights by subsequent tasks to improve performance while still preventing forgetting. Furthermore, we propose a novel KNN-based Knowledge Transfer (KKT) module that utilizes previously acquired knowledge to learn new tasks better and faster. We demonstrate that ExSSNeT outperforms strong previous methods on both NLP and Vision domains while preventing forgetting. Moreover, ExSSNeT is particularly advantageous for sparse masks that activate 2-10% of the model parameters, resulting in an average improvement of 8.3% over SupSup. Furthermore, ExSSNeT scales to a large number of tasks (100).

This work proposes a method named EvolveMT for the efficient combination of multiple machine translation (MT) engines. The method selects the output from one engine for each segment, using online learning techniques to predict the most appropriate system for each translation request. A neural quality estimation metric supervises the method without requiring reference translations. The method’s online learning capability enables it to adapt to changes in the domain or MT engines dynamically, eliminating the requirement for retraining. The method selects a subset of translation engines to be called based on the source sentence features. The degree of exploration is configurable according to the desired quality-cost trade-off. Results from custom datasets demonstrate that EvolveMT achieves similar translation accuracy at a lower cost than selecting the best translation of each segment from all translations using an MT quality estimator. To the best of our knowledge, EvolveMT is the first MT system that adapts itself after deployment to incoming translation requests from the production environment without needing costly retraining on human feedback.

Event ontology provides a shared and formal specification about what happens in the real world and can benefit many natural language understanding tasks. However, the independent development of event ontologies often results in heterogeneous representations that raise the need for establishing alignments between semantically related events. There exists a series of works about ontology alignment (OA), but they only focus on the entity-based OA, and neglect the event-based OA. To fill the gap, we construct an Event Ontology Alignment (EventOA) dataset based on FrameNet and Wikidata, which consists of 900+ event type alignments and 8,000+ event argument alignments. Furthermore, we propose a multi-view event ontology alignment (MEOA) method, which utilizes description information (i.e., name, alias and definition) and neighbor information (i.e., subclass and superclass) to obtain richer representation of the event ontologies. Extensive experiments show that our MEOA outperforms the existing entity-based OA methods and can serve as a strong baseline for EventOA research.

In search engines, query expansion (QE) is a crucial technique to improve search experience. Previous studies often rely on long-term search log mining, which leads to slow updates and is sub-optimal for time-sensitive news searches. In this work, we present Event-Centric Query Expansion (EQE), the QE system used in a famous Chinese search engine. EQE utilizes a novel event retrieval framework that consists of four stages, i.e., event collection, event reformulation, semantic retrieval and online ranking, which can select the best expansion from a significant amount of potential events rapidly and accurately. Specifically, we first collect and filter news headlines from websites. Then we propose a generation model that incorporates contrastive learning and prompt-tuning techniques to reformulate these headlines to concise candidates. Additionally, we fine-tune a dual-tower semantic model to serve as an encoder for event retrieval and explore a two-stage contrastive training approach to enhance the accuracy of event retrieval. Finally, we rank the retrieved events and select the optimal one as QE, which is then used to improve the retrieval of event-related documents. Through offline analysis and online A/B testing, we observed that the EQE system has significantly improved many indicators compared to the baseline. The system has been deployed in a real production environment and serves hundreds of millions of users.

Recent work on Event Extraction has reframed the task as Question Answering (QA), with promising results. The advantage of this approach is that it addresses the error propagation issue found in traditional token-based classification approaches by directly predicting event arguments without extracting candidates first. However, the questions are typically based on fixed templates and they rarely leverage contextual information such as relevant arguments. In addition, prior QA-based approaches have difficulty handling cases where there are multiple arguments for the same role. In this paper, we propose QGA-EE, which enables a Question Generation (QG) model to generate questions that incorporate rich contextual information instead of using fixed templates. We also propose dynamic templates to assist the training of QG model. Experiments show that QGA-EE outperforms all prior single-task-based models on the ACE05 English dataset.

Question generation (QG) is the task of generating a valid and fluent question based on a given context and the target answer. According to various purposes, even given the same context, instructors can ask questions about different concepts, and even the same concept can be written in different ways. However, the evaluation for QG usually depends on single reference-based similarity metrics, such as n-gram-based metric or learned metric, which is not sufficient to fully evaluate the potential of QG methods. To this end, we propose to paraphrase the reference question for a more robust QG evaluation. Using large language models such as GPT-3, we created semantically and syntactically diverse questions, then adopt the simple aggregation of the popular evaluation metrics as the final scores. Through our experiments, we found that using multiple (pseudo) references is more effective for QG evaluation while showing a higher correlation with human evaluations than evaluation with a single reference.

Evaluation is the central means for assessing, understanding, and communicating about NLP models. In this position paper, we argue evaluation should be more than that: it is a force for driving change, carrying a sociological and political character beyond its technical dimensions. As a force, evaluation’s power arises from its adoption: under our view, evaluation succeeds when it achieves the desired change in the field. Further, by framing evaluation as a force, we consider how it competes with other forces. Under our analysis, we conjecture that the current trajectory of NLP suggests evaluation’s power is waning, in spite of its potential for realizing more pluralistic ambitions in the field. We conclude by discussing the legitimacy of this power, who acquires this power and how it distributes. Ultimately, we hope the research community will more aggressively harness evaluation to drive change.

While large language models (LLMs) have proven to be effective on a large variety of tasks, they are also known to hallucinate information. To measure whether an LLM prefers factually consistent continuations of its input, we propose a new benchmark called FIB (Factual Inconsistency Benchmark) that focuses on the task of summarization. Specifically, our benchmark involves comparing the scores an LLM assigns to a factually consistent versus a factually inconsistent summary for an input news article. For factually consistent summaries, we use human-written reference summaries that we manually verify as factually consistent. To generate summaries that are factually inconsistent, we generate summaries from a suite of summarization models that we have manually annotated as factually inconsistent. A model’s factual consistency is then measured according to its accuracy, i.e. the proportion of documents where it assigns a higher score to the factually consistent summary. To validate the usefulness of {pasted macro ‘BENCHMARK’}, we evaluate 23 large language models ranging from 1B to 176B parameters from six different model families including BLOOM and OPT. We find that existing LLMs generally assign a higher score to factually consistent summaries than to factually inconsistent summaries. However, if the factually inconsistent summaries occur verbatim in the document, then LLMs assign a higher score to these factually inconsistent summaries than factually consistent summaries. We validate design choices in our benchmark including the scoring method and source of distractor summaries.

Evaluating grounded neural language model performance with respect to pragmatic qualities like the trade off between truthfulness, contrastivity and overinformativity of generated utterances remains a challenge in absence of data collected from humans. To enable such evaluation, we present a novel open source image-text dataset “Annotated 3D Shapes” (A3DS) comprising over nine million exhaustive natural language annotations and over 12 million variable-granularity captions for the 480,000 images provided by Burgess & Kim (2018).We showcase the evaluation of pragmatic abilities developed by a task-neutral image captioner fine-tuned in a multi-agent communication setting to produce contrastive captions. The evaluation is enabled by the dataset because the exhaustive annotations allow to quantify the presence of contrastive features in the model’s generations. We show that the model develops human-like patterns (informativity, brevity, over-informativity for specific features (e.g., shape, color biases)).

Zero-shot event extraction (EE) methods infer richly structured event records from text, based only on a minimal user specification and no training examples, which enables flexibility in exploring and developing applications. Most event extraction research uses the Automatic Content Extraction (ACE) annotated dataset to evaluate supervised EE methods, but can it be used to evaluate zero-shot and other low-supervision EE? We describe ACE’s event structures and identify significant ambiguities and issues in current evaluation practice, including (1) coreferent argument mentions, (2) conflicting argument head conventions, and (3) ignorance of modality and event class details. By sometimes mishandling these subtleties, current work may dramatically understate the actual performance of zero-shot and other low-supervision EE, considering up to 32% of correctly identified arguments and 25% of correctly ignored event mentions as false negatives. For each issue, we propose recommendations for future evaluations so the research community can better utilize ACE as an event evaluation resource.

We present PaRTE, a collection of 1,126 pairs of Recognizing Textual Entailment (RTE) examples to evaluate whether models are robust to paraphrasing. We posit that if RTE models understand language, their predictions should be consistent across inputs that share the same meaning. We use the evaluation set to determine if RTE models’ predictions change when examples are paraphrased. In our experiments, contemporary models change their predictions on 8-16% of paraphrased examples, indicating that there is still room for improvement.

Lexical matching remains the de facto evaluation method for open-domain question answering (QA). Unfortunately, lexical matching fails completely when a plausible candidate answer does not appear in the list of gold answers, which is increasingly the case as we shift from extractive to generative models. The recent success of large language models (LLMs) for QA aggravates lexical matching failures since candidate answers become longer, thereby making matching with the gold answers even more challenging. Without accurate evaluation, the true progress in open-domain QA remains unknown. In this paper, we conduct a thorough analysis of various open-domain QA models, including LLMs, by manually evaluating their answers on a subset of NQ-open, a popular benchmark. Our assessments reveal that while the true performance of all models is significantly underestimated, the performance of the InstructGPT (zero-shot) LLM increases by nearly +60%, making it on par with existing top models, and the InstructGPT (few-shot) model actually achieves a new state-of-the-art on NQ-open. We also find that more than 50% of lexical matching failures are attributed to semantically equivalent answers. We further demonstrate that regex matching ranks QA models consistent with human judgments, although still suffering from unnecessary strictness. Finally, we demonstrate that automated evaluation models are a reasonable surrogate for lexical matching in some circumstances, but not for long-form answers generated by LLMs. The automated models struggle in detecting hallucinations in LLM answers and are thus unable to evaluate LLMs. At this time, there appears to be no substitute for human evaluation.

The long-standing one-to-many issue of the open-domain dialogues poses significant challenges for automatic evaluation methods, i.e., there may be multiple suitable responses which differ in semantics for a given conversational context. To tackle this challenge, we propose a novel learning-based automatic evaluation metric (CMN), which can robustly evaluate open-domain dialogues by augmenting Conditional Variational Autoencoders (CVAEs) with a Next Sentence Prediction (NSP) objective and employing Mutual Information (MI) to model the semantic similarity of text in the latent space. Experimental results on two open-domain dialogue datasets demonstrate the superiority of our method compared with a wide range of baselines, especially in handling responses which are distant to the “golden” reference responses in semantics.

Cross-lingual summarization aims to help people efficiently grasp the core idea of the document written in a foreign language. Modern text summarization models generate highly fluent but often factually inconsistent outputs, which has received heightened attention in recent research. However, the factual consistency of cross-lingual summarization has not been investigated yet. In this paper, we propose a cross-lingual factuality dataset by collecting human annotations of reference summaries as well as generated summaries from models at both summary level and sentence level. Furthermore, we perform the fine-grained analysis and observe that over 50% of generated summaries and over 27% of reference summaries contain factual errors with characteristics different from monolingual summarization. Existing evaluation metrics for monolingual summarization require translation to evaluate the factuality of cross-lingual summarization and perform differently at different tasks and levels. Finally, we adapt the monolingual factuality metrics as an initial step towards the automatic evaluation of summarization factuality in cross-lingual settings. Our dataset and code are available at https://github.com/kite99520/Fact_CLS.

The ever-increasing size of language models curtails their widespread access to the community, thereby galvanizing many companies and startups into offering access to large language models through APIs. One particular API, suitable for dense retrieval, is the semantic embedding API that builds vector representations of a given text. With a growing number of APIs at our disposal, in this paper, our goal is to analyze semantic embedding APIs in realistic retrieval scenarios in order to assist practitioners and researchers in finding suitable services according to their needs. Specifically, we wish to investigate the capabilities of existing APIs on domain generalization and multilingual retrieval. For this purpose, we evaluate the embedding APIs on two standard benchmarks, BEIR, and MIRACL. We find that re-ranking BM25 results using the APIs is a budget-friendly approach and is most effective on English, in contrast to the standard practice, i.e., employing them as first-stage retrievers. For non-English retrieval, re-ranking still improves the results, but a hybrid model with BM25 works best albeit at a higher cost. We hope our work lays the groundwork for thoroughly evaluating APIs that are critical in search and more broadly, in information retrieval.

In work on AMR (Abstract Meaning Representation), similarity metrics are crucial as they are used to evaluate AMR systems such as AMR parsers. Current AMR metrics are all based on nodes or triples matching without considering the entire structures of AMR graphs. To address this problem, and inspired by learned similarity evaluation on plain text, we propose AMRSim, an automatic AMR graph similarity evaluation metric. To overcome the high cost of collecting human-annotated data, AMRSim automatically generates silver AMR graphs and utilizes self-supervised learning methods. We evaluated AMRSim on various datasets and found that AMRSim significantly improves the correlations with human semantic scores and remains robust under diverse challenges. We also discuss how AMRSim can be extended to multilingual cases.

In recent years machine translation has become very successful for high-resource language pairs. This has also sparked new interest in research on the automatic translation of low-resource languages, including Indigenous languages. However, the latter are deeply related to the ethnic and cultural groups that speak (or used to speak) them. The data collection, modeling and deploying machine translation systems thus result in new ethical questions that must be addressed. Motivated by this, we first survey the existing literature on ethical considerations for the documentation, translation, and general natural language processing for Indigenous languages. Afterward, we conduct and analyze an interview study to shed light on the positions of community leaders, teachers, and language activists regarding ethical concerns for the automatic translation of their languages. Our results show that the inclusion, at different degrees, of native speakers and community members is vital to performing better and more ethical research on Indigenous languages.

In automatic emotion recognition (AER), labels assigned by different human annotators to the same utterance are often inconsistent due to the inherent complexity of emotion and the subjectivity of perception. Though deterministic labels generated by averaging or voting are often used as the ground truth, it ignores the intrinsic uncertainty revealed by the inconsistent labels. This paper proposes a Bayesian approach, deep evidential emotion regression (DEER), to estimate the uncertainty in emotion attributes. Treating the emotion attribute labels of an utterance as samples drawn from an unknown Gaussian distribution, DEER places an utterance-specific normal-inverse gamma prior over the Gaussian likelihood and predicts its hyper-parameters using a deep neural network model. It enables a joint estimation of emotion attributes along with the aleatoric and epistemic uncertainties. AER experiments on the widely used MSP-Podcast and IEMOCAP datasets showed DEER produced state-of-the-art results for both the mean values and the distribution of emotion attributes.

In this paper, we define a widely neglected property in dialogue text, duality, which is a hierarchical property that is reflected in human behaviours in daily conversations: Based on the logic in a conversation (or a sentence), people can infer follow-up utterances (or tokens) based on the previous text, and vice versa. We propose a hierarchical duality learning for dialogue (HDLD) to simulate this human cognitive ability, for generating high quality responses that connect both previous and follow-up dialogues. HDLD utilizes hierarchical dualities at token hierarchy and utterance hierarchy. HDLD maximizes the mutual information between past and future utterances. Thus, even if future text is invisible during inference, HDLD is capable of estimating future information implicitly based on dialogue history and generates both coherent and informative responses. In contrast to previous approaches that solely utilize future text as auxiliary information to encode during training, HDLD leverages duality to enable interaction between dialogue history and the future. This enhances the utilization of dialogue data, leading to the improvement in both automatic and human evaluation.