EMNLP 2023 Papers

Adapting general-purpose language models has proven to be effective in tackling downstream tasks within specific domains. In this paper, we address the task of extracting entities from the economics literature on impact evaluation. To this end, we release EconBERTa, a large language model pretrained on scientific publications in economics, and ECON-IE, a new expert-annotated dataset of economics abstracts for Named Entity Recognition (NER). We find that EconBERTa reaches state-of-the-art performance on our downstream NER task. Additionally, we extensively analyze the model's generalization capacities, finding that most errors correspond to detecting only a subspan of an entity or failure to extrapolate to longer sequences. This limitation is primarily due to an inability to detect part-of-speech sequences unseen during training, and this effect diminishes when the number of unique instances in the training set increases. Examining the generalization abilities of domain-specific language models paves the way towards improving the robustness of NER models for causal knowledge extraction.

In a spoken dialogue system, an NLU model is preceded by a speech recognition system that can deteriorate the performance of natural language understanding. This paper proposes a method for investigating the impact of speech recognition errors on the performance of natural language understanding models. The proposed method combines the back transcription procedure with a fine-grained technique for categorizing the errors that affect the performance of NLU models. The method relies on the usage of synthesized speech for NLU evaluation. We show that the use of synthesized speech in place of audio recording does not change the outcomes of the presented technique in a significant way.

In natural language processing (NLP), the representation of text plays a crucial role in various tasks such as language modeling, sentiment analysis, and machine translation. The standard approach is to represent text in the same way as we, as humans, read and write. In this paper, we propose a novel approach to represent text with only consonants which presents a compact representation of English text that offers improved efficiency without sacrificing performance. We exploit the fact that consonants are more discriminative than vowels and by representing text using consonants, we can significantly reduce the overall memory and compute footprint required for storing and processing textual data. We present two alternative representations: 'consonants-only', where we completely remove the vowels from the text, and 'masked-vowels', where we mask all the vowels into one special symbol. To evaluate our approaches, we conducted experiments on various NLP tasks, including text classification, part-of-speech (POS) tagging, named-entity recognition (NER), and neural machine translation (NMT), in addition to language modeling. Our results demonstrate that the proposed consonant-based representation achieves comparable performance compared to the standard text representation while requiring significantly fewer computational resources. Furthermore, we show that our representation can be seamlessly integrated with existing NLP models and frameworks, providing a practical solution for efficient text processing. Last but not the least, we present a technique to retrieve the vowel information from our processed text representation keeping in mind the need to reproduce text in human readable form in some NLP applications.

For knowledge intensive NLP tasks, it has been widely accepted that accessing more information is a contributing factor to improvements in the model's end-to-end performance. However, counter-intuitively, too much context can have a negative impact on the model when evaluated on common question answering (QA) datasets. In this paper, we analyze how passages can have a detrimental effect on retrieve-then-read architectures used in question answering. Our empirical evidence indicates that the current read architecture does not fully leverage the retrieved passages and significantly degrades its performance when using the whole passages compared to utilizing subsets of them. Our findings demonstrate that model accuracy can be improved by 10\% on two popular QA datasets by filtering out detrimental passages. Additionally, these outcomes are attained by utilizing existing retrieval methods without further training or data. We further highlight the challenges associated with identifying the detrimental passages. First, even with the correct context, the model can make an incorrect prediction, posing a challenge in determining which passages are most influential. Second, evaluation typically considers lexical matching, which is not robust to variations of correct answers. Despite these limitations, our experimental results underscore the pivotal role of identifying and removing these detrimental passages for the context-efficient retrieve-then-read pipeline.

This paper introduces PMIndiaSum, a multilingual and massively parallel summarization corpus focused on languages in India. Our corpus provides a training and testing ground for four language families, 14 languages, and the largest to date with 196 language pairs. We detail our construction workflow including data acquisition, processing, and quality assurance. Furthermore, we publish benchmarks for monolingual, cross-lingual, and multilingual summarization by fine-tuning, prompting, as well as translate-and-summarize. Experimental results confirm the crucial role of our data in aiding summarization between Indian languages. Our dataset is publicly available and can be freely modified and re-distributed.

The theory of Conceptual Spaces is an influential cognitive-linguistic framework for representing the meaning of concepts. Conceptual spaces are constructed from a set of quality dimensions, which essentially correspond to primitive perceptual features (e.g. hue or size). These quality dimensions are usually learned from human judgements, which means that applications of conceptual spaces tend to be limited to narrow domains (e.g. modelling colour or taste). Encouraged by recent findings about the ability of Large Language Models (LLMs) to learn perceptually grounded representations, we explore the potential of such models for learning conceptual spaces. Our experiments show that LLMs can indeed be used for learning meaningful representations to some extent. However, we also find that fine-tuned models of the BERT family are able to match or even outperform the largest GPT-3 model, despite being 2 to 3 orders of magnitude smaller.

Real-world domain experts (e.g., doctors) rarely annotate only a decision label in their day-to-day workflow without providing explanations. Yet, existing low-resource learning techniques, such as Active Learning (AL), that aim to support human annotators mostly focus on the label while neglecting the natural language explanation of a data point. This work proposes a novel AL architecture to support experts' real-world need for label and explanation annotations in low-resource scenarios. Our AL architecture leverages an explanation-generation model to produce explanations guided by human explanations, a prediction model that utilizes generated explanations toward prediction faithfully, and a novel data diversity-based AL sampling strategy that benefits from the explanation annotations. Automated and human evaluations demonstrate the effectiveness of incorporating explanations into AL sampling and the improved human annotation efficiency and trustworthiness with our AL architecture. Additional ablation studies illustrate the potential of our AL architecture for transfer learning, generalizability, and integration with large language models (LLMs). While LLMs exhibit exceptional explanation-generation capabilities for relatively simple tasks, their effectiveness in complex real-world tasks warrants further in-depth study.

Language models (LMs) gradually become general-purpose interfaces in the interactive and embodied world, where the understanding of physical concepts is an essential prerequisite. However, it is unclear whether LMs can understand physical concepts in the human world. To investigate this, we design a benchmark VEC that covers the tasks of (i) Visual concepts, such as the shape and material of objects, and (ii) Embodied Concepts, learned from the interaction with the world such as the temperature of objects. Our zero (few)-shot prompting results show that the understanding of certain visual concepts emerges as scaling up LMs, but there are still basic concepts to which the scaling law does not apply. For example, OPT-175B performs close to humans with a zero-shot accuracy of $85$\% on the material concept, yet behaves like random guessing on the mass concept. Instead, vision-augmented LMs such as CLIP and BLIP achieve a human-level understanding of embodied concepts. Analysis indicates that the rich semantics in visual representation can serve as a valuable source of embodied knowledge. Inspired by this, we propose a distillation method to transfer embodied knowledge from VLMs to LMs, achieving performance gain comparable with that by scaling up parameters of LMs $134\times$. Our dataset is available at https://github.com/TobiasLee/VEC.

This paper investigates the problem-solving capabilities of Large Language Models (LLMs) by evaluating their performance on stumpers, unique single-step intuition problems that pose challenges for human solvers but are easily verifiable. We compare the performance of four state-of-the-art LLMs (Davinci-2, Davinci-3, GPT-3.5-Turbo, GPT-4) to human participants. Our findings reveal that the new-generation LLMs excel in solving stumpers and surpass human performance. However, humans exhibit superior skills in verifying solutions to the same problems. This research enhances our understanding of LLMs' cognitive abilities and provides insights for enhancing their problem-solving potential across various domains.

Prompt tuning prepends a soft prompt to the input embeddings or hidden states and only optimizes the prompt to adapt pretrained models (PTMs) to downstream tasks. The previous work manually selects prompt layers which are far from optimal and failed to exploit the potential of prompt tuning. In this work, we propose a novel framework, \underline{S}elective \underline{P}rompt \underline{T}uning (SPT), that learns to select the proper prompt layers by inserting a prompt controlled by a learnable probabilistic gate at each intermediate layer. We further propose a novel bi-level optimization framework, SPT-DARTS, that can better optimize the learnable gates and improve the final prompt tuning performances of the learned prompt layer settings. We conduct extensive experiments with ten benchmark datasets under the full-data and few-shot scenarios. The results demonstrate that our SPT framework can perform better than the previous state-of-the-art PETuning baselines with comparable or fewer tunable parameters. \footnote{Codes will be publicly available upon acceptance. }

Emotion Recognition in Conversation (ERC) has been widely studied due to its importance in developing emotion-aware empathetic machines. The rise of pre-trained language models (PLMs) has further pushed the limit of ERC performance. However, most recent works on ERC using PLMs are heavily data-driven, and requires fine-tuning the entire PLMs. To improve both sample and computational efficiency, we propose a derivative-free optimization method called Cross-Task Prompt Tuning (CTPT) for few-shot conversational emotion recognition. Unlike existing methods that learn independent knowledge from individual tasks, CTPT leverages sharable cross-task knowledge by exploiting external knowledge from other source tasks to improve learning performance under the few-shot setting. Moreover, CTPT only needs to optimize a vector under the low intrinsic dimensionality without gradient, which is highly parameter-efficient compared with existing approaches. Experiments on five different contextual conversation datasets demonstrate that our CTPT method has superior results on both few-shot scenarios and zero-shot transfers.

Our physical world is constantly evolving over time, rendering challenges for pre-trained language models to understand and reason over the temporal contexts of texts. Existing work focuses on strengthening the direct association between a piece of text and its time-stamp. However, the knowledge-time association is usually insufficient for the downstream tasks that require reasoning over temporal dependencies between knowledge. In this work, we make use of the underlying nature of time, all temporally-scoped sentences are strung together through a one-dimensional time axis, and suggest creating a graph structure based on the relative placements of events along the time axis. Inspired by the graph view, we propose \textsc{RemeMo} ($\underline{Re}lative Ti\underline{me} \underline{Mo}deling$), which explicitly connects all temporally-scoped facts by modeling the time relations between any two sentences. Experimental results show that \textsc{RemeMo} outperforms the baseline T5 on multiple temporal question answering datasets under various settings. Further analysis suggests that \textsc{RemeMo} is especially good at modeling long-range complex temporal dependencies.

Expository documents are vital resources for conveying complex information to readers. Despite their usefulness, writing expository text by hand is a challenging process that requires careful content planning, obtaining facts from multiple sources, and the ability to clearly synthesize these facts. To ease these burdens, we propose the task of expository text generation, which seeks to automatically generate an accurate and stylistically consistent expository text for a topic by intelligently searching a knowledge source. We solve our task by developing IRP, a framework that overcomes the limitations of retrieval-augmented models and iteratively performs content planning, fact retrieval, and rephrasing. Through experiments on three diverse, newly-collected datasets, we show that IRP produces factual and organized expository texts that accurately inform readers.

Wasserstein distance is a powerful tool for comparing probability distributions and is widely used for document classification and retrieval tasks in NLP. In particular, it is known as the word mover's distance (WMD) in the NLP community. WMD exhibits excellent performance for various NLP tasks; however, one of its limitations is its computational cost and thus is not useful for large-scale distribution comparisons. In this study, we propose a simple and effective nearest neighbor search based on the Wasserstein distance. Specifically, we employ the L1 embedding method based on the tree-based Wasserstein approximation and subsequently used the nearest neighbor search to efficiently find the $k$-nearest neighbors. Through benchmark experiments, we demonstrate that the proposed approximation has comparable performance to the vanilla Wasserstein distance and can be computed three orders of magnitude faster than the vanilla Wasserstein distance.

Textbooks are one of the main mediums for delivering high-quality education to students. In particular, explanatory and illustrative visuals play a key role in retention, comprehension and general transfer of knowledge. However, many textbooks lack these interesting visuals to support student learning. In this paper, we investigate the effectiveness of vision-language models to automatically enhance textbooks with images from the web. We collect a dataset of e-textbooks in the math, science, social science and business domains. We then set up a text-image matching task that involves retrieving and appropriately assigning web images to textbooks, which we frame as a matching optimization problem. Through a crowd-sourced evaluation, we verify that (1) while the original textbook images are rated higher, automatically assigned ones are not far behind, and (2) the precise formulation of the optimization problem matters. We release the dataset of textbooks with an associated image bank to inspire further research in this intersectional area of computer vision and NLP for education.

Given the high-stakes nature of healthcare decision-making, we aim to improve the efficiency of human annotators rather than replacing them with fully automated solutions. We introduce a new comprehensive resource, SYMPTOMIFY, a dataset of annotated vaccine adverse reaction reports detailing individual vaccine reactions. The dataset, consisting of over 800k reports, surpasses previous datasets in size. Notably, it features reasoning-based explanations alongside background knowledge obtained via language model knowledge harvesting. We evaluate performance across various methods and learning paradigms, paving the way for future comparisons and benchmarking.

The great success of Language Models (LMs) for various Natural Language Processing (NLP) tasks is accompanied by computational challenges during both pre-training and fine-tuning. Pre-training has attracted significant attention due to its huge computational footprint. We focus on the fine-tuning of pre-trained LMs, which is expected to be performed much more frequently as the pre-trained models are adapted to downstream tasks. During fine-tuning, the presence of variable-length input sequences necessitates the use of padding tokens when batching sequences. These padding tokens lead to ineffectual computations, adversely impacting the efficiency of fine-tuning. We also observe that LMs memorize the limited task-specific training data despite the use of known regularization methods. Based on these insights, we present TokenDrop + BucketSampler, a framework that simultaneously improves efficiency and accuracy of LM fine-tuning. BucketSampler generates batches of samples with lower variance in sequence lengths to reduce the number of padding tokens, but does so without the accompanying accuracy drop seen in previous approaches. TokenDrop is a new regularizer that prunes a random subset of insignificant tokens from each input sequence in every epoch to prevent overfitting. TokenDrop drops more tokens from the longer sequences in each batch to further reduce variance in input lengths and the need for padding. TokenDrop + BucketSampler accelerates fine-tuning on diverse downstream tasks by up to 10.61X, while also producing models that are up to 1.17% more accurate compared to conventional fine-tuning. Code is available at https://github.com/amrnag/TokenDrop-BucketSampler. .

Accurate processing of non-compositional language relies on generating good representations for such expressions. In this work, we study the representation of language non-compositionality by proposing a language model, PIER+, that builds on BART and can create semantically meaningful and contextually appropriate representations for English potentially idiomatic expressions (PIEs). PIEs are characterized by their non-compositionality and contextual ambiguity in their literal and idiomatic interpretations. Via intrinsic evaluation on embedding quality and extrinsic evaluation on PIE processing and NLU tasks, we show that representations generated by PIER+ result in 33\% higher homogeneity score for embedding clustering than BART, whereas 3.12\% and 3.29\% gains in accuracy and sequence accuracy for PIE sense classification and span detection compared to the state-of-the-art IE representation model, GIEA. These gains are achieved without sacrificing PIER+'s performance on NLU tasks (+/- 1\% accuracy) compared to BART.

We study continual event extraction, which aims to extract incessantly emerging event information while avoiding forgetting. We observe that the semantic confusion on event types stems from the annotations of the same text being updated over time. The imbalance between event types even aggravates this issue. This paper proposes a novel continual event extraction model with semantic confusion rectification. We mark pseudo labels for each sentence to alleviate semantic confusion. We transfer pivotal knowledge between current and previous models to enhance the understanding of event types. Moreover, we encourage the model to focus on the semantics of long-tailed event types by leveraging other associated types. Experimental results show that our model outperforms state-of-the-art baselines and is proficient in imbalanced datasets.