ACL 2023 Papers

Data drift is the change in model input data that is one of the key factors leading to machine learning models performance degradation over time. Monitoring drift helps detecting these issues and preventing their harmful consequences. Meaningful drift interpretation is a fundamental step towards effective re-training of the model. In this study we propose an end-to-end framework for reliable model-agnostic change-point detection and interpretation in large task-oriented dialog systems, proven effective in multiple customer deployments. We evaluate our approach and demonstrate its benefits with a novel variant of intent classification training dataset, simulating customer requests to a dialog system. We make the data publicly available.

We apply reinforcement learning techniques to topic modeling by replacing the variational autoencoder in ProdLDA with a continuous action space reinforcement learning policy. We train the system with a policy gradient algorithm REINFORCE. Additionally, we introduced several modifications: modernize the neural network architecture, weight the ELBO loss, use contextual embeddings, and monitor the learning process via computing topic diversity and coherence for each training step. Experiments areperformed on 11 data sets. Our unsupervised model outperforms all other unsupervised models and performs on par with or better than most models using supervised labeling. Our model is outperformed on certain data sets by a model using supervised labeling and contrastive learning. We have also conducted an ablation study to provide empirical evidence of performance improvements from changes we made to ProdLDA and found that the reinforcement learning formulation boosts performance. We open-source our code implementation.

Text classification datasets from specialised or technical domains are in high demand, especially in industrial applications. However, due to the high cost of annotation such datasets are usually expensive to create. While Active Learning (AL) can reduce the labeling cost, required AL strategies are often only tested on general knowledge domains and tend to use information sources that are not consistent across tasks. We propose Reinforced Active Learning (RAL) to train a Reinforcement Learning policy that utilizes many different aspects of the data and the task in order to select the most informative unlabeled subset dynamically over the course of the AL procedure. We demonstrate the superior performance of the proposed RAL framework compared to strong AL baselines across four intricate multi-class, multi-label text classification datasets taken from specialised domains. In addition, we experiment with a unique data augmentation approach to further reduce the number of samples RAL needs to annotate.

We present a reality check on large language models and inspect the promise of retrieval-augmented language models in comparison. Such language models are semi-parametric, where models integrate model parameters and knowledge from external data sources to make their predictions, as opposed to the parametric nature of vanilla large language models. We give initial experimental findings that semi-parametric architectures can be enhanced with views, a query analyzer/planner, and provenance to make a significantly more powerful system for question answering in terms of accuracy and efficiency, and potentially for other NLP tasks.

We study the problem of defying catastrophic forgetting when learning a series of language processing tasks. Compared with previous methods, we emphasize the importance of not caching history tasks’ data, which makes the problem more challenging. Our proposed method applies the parameter isolation strategy. For each task, it allocates a small portion of private parameters and learns them with a shared pre-trained model. To load correct parameters at testing time, we introduce a simple yet effective non-parametric method. Experiments on continual language learning benchmarks show that our method is significantly better than all existing no-data-cache methods, and is comparable (or even better) than those using historical data.

In real-world systems, an important requirement for model updates is to avoid regressions in user experience caused by flips of previously correct classifications to incorrect ones. Multiple techniques for that have been proposed in the recent literature. In this paper, we apply one such technique, focal distillation, to model updates in a goal-oriented dialog system and assess its usefulness in practice. In particular, we evaluate its effectiveness for key language understanding tasks, including sentence classification and sequence labeling tasks, we further assess its effect when applied to repeated model updates over time, and test its compatibility with mislabeled data. Our experiments on a public benchmark and data from a deployed dialog system demonstrate that focal distillation can substantially reduce regressions, at only minor drops in accuracy, and that it further outperforms naive supervised training in challenging mislabeled data and label expansion settings.

Voice assistants help users make phone calls, send messages, create events, navigate and do a lot more. However assistants have limited capacity to understand their users’ context. In this work, we aim to take a step in this direction. Our work dives into a new experience for users to refer to phone numbers, addresses, email addresses, urls, and dates on their phone screens. We focus on reference understanding, which is particularly interesting when, similar to visual grounding, there are multiple similar texts on screen. We collect a dataset and propose a lightweight general purpose model for this novel experience. Since consuming pixels directly is expensive, our system is designed to rely only on text extracted from the UI. Our model is modular, offering flexibility, better interpretability and efficient run time memory.

Factuality is important to dialogue summarization. Factual error correction (FEC) of model-generated summaries is one way to improve factuality. Current FEC evaluation that relies on factuality metrics is not reliable and detailed enough. To address this problem, we are the first to manually annotate a FEC dataset for dialogue summarization containing 4000 items and propose FERRANTI, a fine-grained evaluation framework based on reference correction that automatically evaluates the performance of FEC models on different error categories. Using this evaluation framework, we conduct sufficient experiments with FEC approaches under a variety of settings and find the best training modes and significant differences in the performance of the existing approaches on different factual error categories.

Bias in machine learning models can be an issue when the models are trained on particular types of data that do not generalize well, causing under performance in certain groups of users. In this work, we focus on reducing the bias related to new customers in a digital voice assistant system. It is observed that natural language understanding models often have lower performance when dealing with requests coming from new users rather than experienced users. To mitigate this problem, we propose a framework that consists of two phases (1) a fixing phase with four active learning strategies used to identify important samples coming from new users, and (2) a self training phase where a teacher model trained from the first phase is used to annotate semi-supervised samples to expand the training data with relevant cohort utterances. We explain practical strategies that involve an identification of representative cohort-based samples through density clustering as well as employing implicit customer feedbacks to improve new customers’ experience. We demonstrate the effectiveness of our approach in a real world large scale voice assistant system for two languages, German and French through both offline experiments as well as A/B testings.

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.

Continual pre-training is the paradigm where pre-trained language models (PLMs) continually acquire fresh knowledge from growing data and gradually get upgraded. Before an upgraded PLM is released, we may have tuned the original PLM for various tasks and stored the adapted weights. However, when tuning the upgraded PLM, these outdated adapted weights will typically be ignored and discarded, causing a potential waste of resources. We bring this issue to the forefront and contend that proper algorithms for recycling outdated adapted weights should be developed. To this end, we formulate the task of recyclable tuning for continual pre-training. In pilot studies, we find that after continual pre-training, the upgraded PLM remains compatible with the outdated adapted weights to some extent. Motivated by this finding, we analyze the connection between continually pre-trained PLMs from two novel aspects, i.e., mode connectivity, and functional similarity. Based on the corresponding findings, we propose both an initialization-based method and a distillation-based method for our task. We demonstrate their feasibility in improving the convergence and performance for tuning the upgraded PLM. We also show that both methods can be combined to achieve better performance.

Generating intermediate steps, or Chain of Thought (CoT), is an effective way to significantly improve language models’ (LM) multi-step reasoning capability. However, the CoT lengths can grow rapidly with the problem complexity, easily exceeding the maximum context size. Instead of increasing the context limit, which has already been heavily investigated, we explore an orthogonal direction: making LMs divide a problem into multiple contexts. We propose a new inference framework, called Recursion of Thought (RoT), which introduces several special tokens that the models can output to trigger context-related operations. Extensive experiments with multiple architectures including GPT-3 show that RoT dramatically improves LMs’ inference capability to solve problems, whose solution consists of hundreds of thousands of tokens.

Self-attention-based models have achieved remarkable progress in short-text mining. However, the quadratic computational complexities restrict their application in long text processing. Prior works have adopted the chunking strategy to divide long documents into chunks and stack a self-attention backbone with the recurrent structure to extract semantic representation. Such an approach disables parallelization of the attention mechanism, significantly increasing the training cost and raising hardware requirements. Revisiting the self-attention mechanism and the recurrent structure, this paper proposes a novel long-document encoding model, Recurrent Attention Network (RAN), to enable the recurrent operation of self-attention. Combining the advantages from both sides, the well-designed RAN is capable of extracting global semantics in both token-level and document-level representations, making it inherently compatible with both sequential and classification tasks, respectively. Furthermore, RAN is computationally scalable as it supports parallelization on long document processing. Extensive experiments demonstrate the long-text encoding ability of the proposed RAN model on both classification and sequential tasks, showing its potential for a wide range of applications.

We propose reconstruction probing, a new analysis method for contextualized representations based on reconstruction probabilities in masked language models (MLMs). This method relies on comparing the reconstruction probabilities of tokens in a given sequence when conditioned on the representation of a single token that has been fully contextualized and when conditioned on only the decontextualized lexical prior of the model. This comparison can be understood as quantifying the contribution of contextualization towards reconstruction—the difference in the reconstruction probabilities can only be attributed to the representational change of the single token induced by contextualization. We apply this analysis to three MLMs and find that contextualization boosts reconstructability of tokens that are close to the token being reconstructed in terms of linear and syntactic distance. Furthermore, we extend our analysis to finer-grained decomposition of contextualized representations, and we find that these boosts are largely attributable to static and positional embeddings at the input layer.

Pre-trained encoder-only and sequence-to-sequence (seq2seq) models each have advantages, however training both model types from scratch is computationally expensive. We explore recipes to improve pre-training efficiency by initializing one model from the other. (1) Extracting the encoder from a seq2seq model, we show it under-performs a Masked Language Modeling (MLM) encoder, particularly on sequence labeling tasks. Variations of masking during seq2seq training, reducing the decoder size, and continuing with a small amount of MLM training do not close the gap. (2) Conversely, using an encoder to warm-start seq2seq training, we show that by unfreezing the encoder partway through training, we can match task performance of a from-scratch seq2seq model. Overall, this two-stage approach is an efficient recipe to obtain both a multilingual encoder and a seq2seq model, matching the performance of training each model from scratch while reducing the total compute cost by 27%.

Ultra-fine entity typing (UFET) predicts extremely free-formed types (e.g., president, politician) of a given entity mention (e.g., Joe Biden) in context. State-of-the-art (SOTA) methods use the cross-encoder (CE) based architecture. CE concatenates a mention (and its context) with each type and feeds the pair into a pretrained language model (PLM) to score their relevance. It brings deeper interaction between the mention and the type to reach better performance but has to perform N (the type set size) forward passes to infer all the types of a single mention. CE is therefore very slow in inference when the type set is large (e.g., N=10k for UFET). % Cross-encoder also ignores the correlation between different types.To this end, we propose to perform entity typing in a recall-expand-filter manner. The recall and expansion stages prune the large type set and generate K (typically much smaller than N) most relevant type candidates for each mention. At the filter stage, we use a novel model called {pasted macro ‘NAME’} to concurrently encode and score all these K candidates in only one forward pass to obtain the final type prediction. We investigate different model options for each stage and conduct extensive experiments to compare each option, experiments show that our method reaches SOTA performance on UFET and is thousands of times faster than the CE-based architecture. We also found our method is very effective in fine-grained (130 types) and coarse-grained (9 types) entity typing. Our code is available at {pasted macro ‘CODE’}.

Reasoning, as an essential ability for complex problem-solving, can provide back-end support for various real-world applications, such as medical diagnosis, negotiation, etc. This paper provides a comprehensive survey of cutting-edge research on reasoning with language model prompting. We introduce research works with comparisons and summaries and provide systematic resources to help beginners. We also discuss the potential reasons for emerging such reasoning abilities and highlight future research directions. Resources are available at https://github.com/zjunlp/Prompt4ReasoningPapers (updated periodically).

Explainable question answering (XQA) aims to answer a given question and provide an explanation why the answer is selected. Existing XQA methods focus on reasoning on a single knowledge source, e.g., structured knowledge bases, unstructured corpora, etc. However, integrating information from heterogeneous knowledge sources is essential to answer complex questions. In this paper, we propose to leverage question decomposing for heterogeneous knowledge integration, by breaking down a complex question into simpler ones, and selecting the appropriate knowledge source for each sub-question. To facilitate reasoning, we propose a novel two-stage XQA framework, Reasoning over Hierarchical Question Decomposition Tree (RoHT). First, we build the Hierarchical Question Decomposition Tree (HQDT) to understand the semantics of a complex question; then, we conduct probabilistic reasoning over HQDT from root to leaves recursively, to aggregate heterogeneous knowledge at different tree levels and search for a best solution considering the decomposing and answering probabilities. The experiments on complex QA datasets KQA Pro and Musique show that our framework outperforms SOTA methods significantly, demonstrating the effectiveness of leveraging question decomposing for knowledge integration and our RoHT framework.

Large language models (LLMs) have revolutionized NLP by solving downstream tasks with little to no labeled data. Despite their versatile abilities, the larger question of their ability to reason remains ill-understood. This paper addresses reasoning in math word problems (MWPs) by studying symbolic versions of the numeric problems, since a symbolic expression is a “concise explanation” of the numeric answer. We create and use a symbolic version of the SVAMP dataset and find that GPT-3’s davinci-002 model also has good zero-shot accuracy on symbolic MWPs. To evaluate the faithfulness of the model’s reasoning, we go beyond accuracy and additionally evaluate the alignment between the final answer and the outputted reasoning, which correspond to numeric and symbolic answers respectively for MWPs. We explore a self-prompting approach to encourage the symbolic reasoning to align with the numeric answer, thus equipping the LLM with the ability to provide a concise and verifiable reasoning and making it more interpretable. Surprisingly, self-prompting also improves the symbolic accuracy to be higher than both the numeric and symbolic accuracies, thus providing an ensembling effect. The SVAMP-Sym dataset will be released for future research on symbolic math problems.