ACL 2023 论文列表

There is a multitude of textual data relevant to the financial markets, spanning genres such as financial news, earnings conference calls, and social media posts. Earnings conference calls are one of the most important to information flow as they reflect a direct communication between company executives, financial analysts, and large shareholders. Since these calls contain content that is forward-looking in nature, they can be used to forecast the future performance of the company relative to market expectations. However, they typically contain over 5,000 words of text and large amounts of industry jargon. This length and domain-specific language present problems for many generic pretrained language models. In this work, we introduce a novel task of predicting earnings surprises from earnings call transcripts and contribute a new long document dataset that tests financial understanding with complex signals. We explore a variety of approaches for this long document classification task and establish some strong baselines. Furthermore, we demonstrate that it is possible to predict companies’ future earnings surprises from solely the text of their conference calls with reasonable accuracy. Finally, we probe the models through different interpretability methods and reveal some intuitive explanations of the linguistic features captured that go beyond traditional sentiment analysis.

Among the problems with leaderboard culture in NLP has been the widespread lack of confidence estimation in reported results. In this work, we present a framework and simulator for estimating p-values for comparisons between the results of two systems, in order to understand the confidence that one is actually better (i.e. ranked higher) than the other. What has made this difficult in the past is that each system must itself be evaluated by comparison to a gold standard. We define a null hypothesis that each system’s metric scores are drawn from the same distribution, using variance found naturally (though rarely reported) in test set items and individual labels on an item (responses) to produce the metric distributions. We create a test set that evenly mixes the responses of the two systems under the assumption the null hypothesis is true. Exploring how to best estimate the true p-value from a single test set under different metrics, tests, and sampling methods, we find that the presence of response variance (from multiple raters or multiple model versions) has a profound impact on p-value estimates for model comparison, and that choice of metric and sampling method is critical to providing statistical guarantees on model comparisons.

In open-ended natural-language generation, existing text decoding methods typically struggle to produce text which is both diverse and high-quality. Greedy and beam search are known to suffer from text degeneration and linguistic diversity issues, while temperature, top-k, and nucleus sampling yield diverse but often lower-quality outputs. In this work, we build upon Minimum Bayes Risk Decoding (MBRD), a family of decoding methods based on Bayesian risk minimization, to address this diversity-quality trade-off. Inspired by the principle of the wisdom of the crowd, MBRD seeks to select a candidate from a pool of candidates that has the least expected risk under a generative model according to a given utility function. The crowd of candidates serves as an approximation for the distribution over human-generated references. We show that MBRD generalizes numerous decoding methods, including majority voting, and can be used as a drop-in replacement for existing sampling methods. Across a wide range of tasks—such as summarization, data-to-text, translation, and textual style transfer—MBRD yields 3-7 ROUGE and BLEU point improvements, including state-of-the-art results on WebNLG and WMT’16.

Understanding users’ intentions in e-commerce platforms requires commonsense knowledge. In this paper, we present FolkScope, an intention knowledge graph construction framework, to reveal the structure of humans’ minds about purchasing items. As commonsense knowledge is usually ineffable and not expressed explicitly, it is challenging to perform information extraction. Thus, we propose a new approach that leverages the generation power of large language models (LLMs) and human-in-the-loop annotation to semi-automatically construct the knowledge graph. LLMs first generate intention assertions via e-commerce specific prompts to explain shopping behaviors, where the intention can be an open reason or a predicate falling into one of 18 categories aligning with ConceptNet, e.g., IsA, MadeOf, UsedFor, etc. Then we annotate plausibility and typicality labels of sampled intentions as training data in order to populate human judgments to all automatic generations. Last, to structurize the assertions, we propose pattern mining and conceptualization to form more condensed and abstract knowledge. Extensive evaluations and study demonstrate that our constructed knowledge graph can well model e-commerce knowledge and have many potential applications.

Few-shot named entity recognition (NER), identifying named entities with a small number of labeled data, has attracted much attention. Frequently, entities are nested within each other. However, most of the existing work on few-shot NER addresses flat entities instead of nested entities. To tackle nested NER in a few-shot setting, it is crucial to utilize the limited labeled data to mine unique features of nested entities, such as the relationship between inner and outer entities and contextual position information. Therefore, in this work, we propose a novel method based on focusing, bridging and prompting for few-shot nested NER without using source domain data. Both focusing and bridging components provide accurate candidate spans for the prompting component. The prompting component leverages the unique features of nested entities to classify spans based on soft prompts and contrastive learning. Experimental results show that the proposed approach achieves state-of-the-art performance consistently on the four benchmark datasets (ACE2004, ACE2005, GENIA and KBP2017) and outperforms several competing baseline models on F1-score by 9.33% on ACE2004, 6.17% on ACE2005, 9.40% on GENIA and 5.12% on KBP2017 on the 5-shot setting.

In a controllable text generation dataset, there exist unannotated attributes that could provide irrelevant learning signals to models that use it for training and thus degrade their performance. We propose focused prefix tuning (FPT) to mitigate the problem and to enable the control to focus on the desired attribute. Experimental results show that FPT can achieve better control accuracy and text fluency than baseline models in single-attribute control tasks. In multi-attribute control tasks, FPT achieves comparable control accuracy with the state-of-the-art approach while keeping the flexibility to control new attributes without retraining existing models.

Inquiry conversation is a common form of conversation that aims to complete the investigation (e.g., court hearing, medical consultation and police interrogation) during which a series of focus shifts occurs. While many models have been proposed to generate a smooth response to a given conversation history, neglecting the focus can limit performance in inquiry conversation where the order of the focuses plays there a key role. In this paper, we investigate the problem of response generation in inquiry conversation by taking the focus into consideration. We propose a novel Focus-aware Response Generation (FRG) method by jointly optimizing a multi-level encoder and a set of focal decoders to generate several candidate responses that correspond to different focuses. Additionally, a focus ranking module is proposed to predict the next focus and rank the candidate responses. Experiments on two orthogonal inquiry conversation datasets (judicial, medical domain) demonstrate that our method generates results significantly better in automatic metrics and human evaluation compared to the state-of-the-art approaches.

In this paper, we investigate how to improve tagging-based Grammatical Error Correction models. We address two issues of current tagging-based approaches, label imbalance issue, and tagging entanglement issue. Then we propose to down-weight the loss of well-classified labels using Focal Loss and decouple the error detection layer from the label tagging layer through an extra self-attention-based matching module. Experiments over three latest Chinese Grammatical Error Correction datasets show that our proposed methods are effective. We further analyze choices of hyper-parameters for Focal Loss and inference tweaking.

Stutter removal is an essential scenario in the field of speech editing. However, when the speech recording contains stutters, the existing text-based speech editing approaches still suffer from: 1) the over-smoothing problem in the edited speech; 2) lack of robustness due to the noise introduced by stutter; 3) to remove the stutters, users are required to determine the edited region manually. To tackle the challenges in stutter removal, we propose FluentSpeech, a stutter-oriented automatic speech editing model. Specifically, 1) we propose a context-aware diffusion model that iteratively refines the modified mel-spectrogram with the guidance of context features; 2) we introduce a stutter predictor module to inject the stutter information into the hidden sequence; 3) we also propose a stutter-oriented automatic speech editing (SASE) dataset that contains spontaneous speech recordings with time-aligned stutter labels to train the automatic stutter localization model. Experimental results on VCTK and LibriTTS datasets demonstrate that our model achieves state-of-the-art performance on speech editing. Further experiments on our SASE dataset show that FluentSpeech can effectively improve the fluency of stuttering speech in terms of objective and subjective metrics. Code and audio samples can be found at https://github.com/Zain-Jiang/Speech-Editing-Toolkit.

Sparsely gated Mixture of Experts (MoE) models have been shown to be a compute-efficient method to scale model capacity for multilingual machine translation. However, for low-resource tasks, MoE models severely over-fit. We show effective regularization strategies, namely dropout techniques for MoE layers in EOM and FOM, Conditional MoE Routing and Curriculum Learning methods that prevent over-fitting and improve the performance of MoE models on low-resource tasks without adversely affecting high-resource tasks. On a massively multilingual machine translation benchmark, our strategies result in about +1 chrF++ improvement in very low resource language pairs. We perform an extensive analysis of the learned MoE routing to better understand the impact of our regularization methods and how we can improve them.

Recent works have shown that attaching prompts to the input is effective at conditioning Language Models (LM) to perform specific tasks. However, prompts are always included in the input text during inference, even when they are fixed, thus incurring substantial computational and memory overhead. Also, there is currently no straightforward method of utilizing prompts that are longer than the maximum input length of the LMs without incurring additional costs during inference. We formally define Fixed Input Parameterization (FIP) problem that focuses on injecting the fixed prompt into the parameters of an LM to be an efficient alternative to attaching fixed prompts to the input. We show that in scenarios with long fixed prompts, FIP can be up to 280 times more efficient in terms of total FLOPs than previous approaches. We further explore methodologies for FIP and show promising results in persona-dependent conversation, semantic parsing, and zero-shot learning with task instructions. Through these explorations, we show that FIP can be a promising direction for conditioning language models, in scenarios with long and fixed prompts.

Pre-trained language models (PLMs) are known to be overly parameterized and have significant redundancy, indicating a small degree of freedom of the PLMs. Motivated by the observation, in this paper, we study the problem of re-parameterizing and fine-tuning PLMs from a new perspective: Discovery of intrinsic task-specific subspace. Specifically, by exploiting the dynamics of the fine-tuning process for a given task, the parameter optimization trajectory is learned to uncover its intrinsic task-specific subspace. A key finding is that PLMs can be effectively fine-tuned in the subspace with a small number of free parameters. Beyond, we observe some outlier dimensions emerging during fine-tuning in the subspace. Disabling these dimensions degrades the model performance significantly. This suggests that these dimensions are crucial to induce task-specific knowledge to downstream tasks.

The Wav2Vec and its variants have achieved unprecedented success in computational auditory and speech processing. Meanwhile, neural encoding studies that integrate the superb representation capability of Wav2Vec and link those representations to brain activities have provided novel insights into a fundamental question of how auditory and speech processing unfold in the human brain. Without an explicit definition, most existing studies treat each transformer encoding layer in Wav2Vec as a single artificial neuron (AN). That is, the layer-level embeddings are used to predict neural responses. However, the comprehensive layer-level embedding aggregates multiple types of contextual attention captured by multi-head self-attention (MSA) modules. Thus, the layer-level ANs lack fine-granularity for neural encoding. To address this limitation, we define the elementary units, i.e., each hidden dimension, as neuron-level ANs in Wav2Vec2.0, quantify their temporal responses, and couple those ANs with their biological-neuron (BN) counterparts in the human brain. Our experimental results demonstrated that: 1) The proposed neuron-level ANs carry meaningful neurolinguistic information; 2) Those ANs anchor to their BN signatures; 3) The AN-BN anchoring patterns are interpretable from a neurolinguistic perspective. More importantly, our results suggest an intermediate stage in both the computational representation in Wav2Vec2.0 and the cortical representation in the brain. Our study validates the fine-grained ANs in Wav2Vec2.0, which may serve as a novel and general strategy to link transformer-based deep learning models to neural responses for probing the sensory processing in the brain.

Adaptive inference is a simple method for reducing inference costs. The method works by maintaining multiple classifiers of different capacities, and allocating resources to each test instance according to its difficulty. In this work, we compare the two main approaches for adaptive inference, Early-Exit and Multi-Model, when training data is limited. First, we observe that for models with the same architecture and size, individual Multi-Model classifiers outperform their Early-Exit counterparts by an average of 2.3%. We show that this gap is caused by Early-Exit classifiers sharing model parameters during training, resulting in conflicting gradient updates of model weights. We find that despite this gap, Early-Exit still provides a better speed-accuracy trade-off due to the overhead of the Multi-Model approach. To address these issues, we propose SWEET (Separating Weights for Early-Exit Transformers) an Early-Exit fine-tuning method that assigns each classifier its own set of unique model weights, not updated by other classifiers. We compare SWEET’s speed-accuracy curve to standard Early-Exit and Multi-Model baselines and find that it outperforms both methods at fast speeds while maintaining comparable scores to Early- Exit at slow speeds. Moreover, SWEET individual classifiers outperform Early-Exit ones by 1.1% on average. SWEET enjoys the benefits of both methods, paving the way for further reduction of inference costs in NLP.

Recent studies have revealed some issues of Multi-Head Attention (MHA), e.g., redundancy and over-parameterization. Specifically, the heads of MHA were originally designed to attend to information from different representation subspaces, whereas prior studies found that some attention heads likely learn similar features and can be pruned without harming performance. Inspired by the minimum-redundancy feature selection, we assume that focusing on the most representative and distinctive features with minimum resources can mitigate the above issues and lead to more effective and efficient MHAs. In particular, we propose Grouped Head Attention, trained with a self-supervised group constraint that group attention heads, where each group focuses on an essential but distinctive feature subset. We additionally propose a Voting-to-Stay procedure to remove redundant heads, thus achieving a transformer with lighter weights. Extensive experiments are consistent with our hypothesis. Moreover, our method achieves significant performance gains on three well-established tasks while considerably compressing parameters.

The U.S. Securities and Exchange Commission (SEC) mandates all public companies to file periodic financial statements that should contain numerals annotated with a particular label from a taxonomy. In this paper, we formulate the task of automating the assignment of a label to a particular numeral span in a sentence from an extremely large label set. Towards this task, we release a dataset, Financial Numeric Extreme Labelling (FNXL), annotated with 2,794 labels. We benchmark the performance of the FNXL dataset by formulating the task as (a) a sequence labelling problem and (b) a pipeline with span extraction followed by Extreme Classification. Although the two approaches perform comparably, the pipeline solution provides a slight edge for the least frequent labels.

Recent years have witnessed a growing interest in investigating what Transformer-based language models (TLMs) actually learn from the training data. This is especially relevant for complex tasks such as the understanding of non-literal meaning. In this work, we probe the performance of three black-box TLMs and two intrinsically transparent white-box models on figurative language classification of sarcasm, similes, idioms, and metaphors. We conduct two studies on the classification results to provide insights into the inner workings of such models. With our first analysis on feature importance, we identify crucial differences in model behavior. With our second analysis using an online experiment with human participants, we inspect different linguistic characteristics of the four figurative language types.

NLP models often rely on superficial cues known as dataset biases to achieve impressive performance, and can fail on examples where these biases do not hold. Recent work sought to develop robust, unbiased models by filtering biased examples from training sets. In this work, we argue that such filtering can obscure the true capabilities of models to overcome biases, which might never be removed in full from the dataset. We suggest that in order to drive the development of models robust to subtle biases, dataset biases should be amplified in the training set. We introduce an evaluation framework defined by a bias-amplified training set and an anti-biased test set, both automatically extracted from existing datasets. Experiments across three notions of bias, four datasets and two models show that our framework is substantially more challenging for models than the original data splits, and even more challenging than hand-crafted challenge sets. Our evaluation framework can use any existing dataset, even those considered obsolete, to test model robustness. We hope our work will guide the development of robust models that do not rely on superficial biases and correlations. To this end, we publicly release our code and data.

Fusion-in-Decoder (FiD) is a powerful retrieval-augmented language model that sets the state-of-the-art on many knowledge-intensive NLP tasks. However, the architecture used for FiD was chosen by making minimal modifications to a standard T5 model, which our analysis shows to be highly suboptimal for a retrieval-augmented model. In particular, FiD allocates the bulk of FLOPs to the encoder, while the majority of inference time results from memory bandwidth constraints in the decoder. We propose two simple changes to the FiD architecture to alleviate memory bandwidth constraints, and speed up inference by 7x. This allows us to use a much larger decoder at modest cost. We denote FiD with the above modifications as FiDO, and show that it strongly improves performance over existing FiD models for a wide range of inference budgets. For example, FiDO-Large-XXL performs faster inference than FiD-Base and achieves better performance than FiD-Large.