Canada

Deploying large language models (LLMs) is challenging because they are memory inefficient and compute-intensive for practical applications. In reaction, researchers train smaller task-specific models by either finetuning with human labels or distilling using LLM-generated labels. However, finetuning and distillation require large amounts of training data to achieve comparable performance to LLMs. We introduce Distilling step-by-step, a new mechanism that (a) trains smaller models that outperform LLMs, and (b) achieves so by leveraging less training data needed by finetuning or distillation. Our method extracts LLM rationales as additional supervision for training small models within a multi-task framework. We present three findings across 4 NLP benchmarks: First, compared to both finetuning and distillation, our mechanism achieves better performance with much fewer labeled/unlabeled training examples. Second, compared to few-shot prompted LLMs, we achieve better performance using substantially smaller model sizes. Third, we reduce both the model size and the amount of data required to outperform LLMs; our finetuned 770M T5 model outperforms the few-shot prompted 540B PaLM model using only 80% of available data on a benchmark, whereas standard finetuning the same T5 model struggles to match even by using 100% of the dataset.

&lt;iframe src=&quot;https://app.sli.do/event/iHwnn5G9xYwcAmSbeKA4Au/embed/polls/e8e76e77-fd12-4d5c-a5bd-583fdbcc5036&quot; width=&quot;300&quot; height=&quot;400&quot;&gt;&lt;/iframe&gt;

ACL 2023

Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes

chain-of-thought reasonings

distilling step-by-step

knowledge distillation

### Welcome to ACL 2023, the 61st Annual Meeting of the Association for Computational Linguistics! 
<br>The conference will be held in Toronto, Canada, July 9-14, 2023.<br><br>
Following the succession of the recent conferences in our field, ACL 2023 will adopt a hybrid format.
While the impact of Covid has considerably diminished in terms of traveling, obtaining visas to Canada
entails a very long process. Moreover, the global economic conditions pose challenges for many individuals to travel to conferences. Recognizing these circumstances, we know many participants may not be
able to attend the conference in person. Therefore, we are committed to providing a great virtual platform
so everyone has the opportunity to interact with other participants and enjoy the conference. Based on the
current registered participants, approxiately 30% have chosen to attend the conference virtually. Whether
you join us in person or virtually, we sincerely hope everyone has a remarkable conference experience.<br>
This General Chair’s message is where I express my gratitude to the many individuals who have made
enormous contributions to the conference over the past year.

Read [**ACL 2023 General Chair's message**](https://docs.google.com/document/d/1WobYM7norbG4dI48s75HfJoD89qgX5a_F-6U8AteLSA/edit?usp=sharing/) in full.

##### **[Conference Handbook](https://2023.aclweb.org/downloads/acl2023-handbook.pdf)**
<br><br><br><br>

The Association for Computational Linguistics (ACL) is the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics or natural language processing.

Deploying large language models (LLMs) is challenging because they are memory inefficient and compute-intensive for practical applications. In reaction, researchers train smaller task-specific models by either finetuning with human labels or distilling using LLM-generated labels. However, finetuning and distillation require large amounts of training data to achieve comparable performance to LLMs. We introduce Distilling step-by-step, a new mechanism that (a) trains smaller models that outperform LLMs, and (b) achieves so by leveraging less training data needed by finetuning or distillation. Our method extracts LLM rationales as additional supervision for training small models within a multi-task framework. We present three findings across 4 NLP benchmarks: First, compared to both finetuning and distillation, our mechanism achieves better performance with much fewer labeled/unlabeled training examples. Second, compared to few-shot prompted LLMs, we achieve better performance using substantially smaller model sizes. Third, we reduce both the model size and the amount of data required to outperform LLMs; our finetuned 770M T5 model outperforms the few-shot prompted 540B PaLM model using only 80% of available data on a benchmark, whereas standard finetuning the same T5 model struggles to match even by using 100% of the dataset.

<iframe src="https://app.sli.do/event/iHwnn5G9xYwcAmSbeKA4Au/embed/polls/e8e76e77-fd12-4d5c-a5bd-583fdbcc5036" width="300" height="400"></iframe>

poster

The impression is crucial for the referring physicians to grasp key information since it is concluded from the findings and reasoning of radiologists. To alleviate the workload of radiologists and reduce repetitive human labor in impression writing, many researchers have focused on automatic impression generation. However, recent works on this task mainly summarize the corresponding findings and pay less attention to the radiology images. In clinical, radiographs can provide more detailed valuable observations to enhance radiologists' impression writing, especially for complicated cases. Besides, each sentence in findings usually focuses on single anatomy, such that they only need to be matched to corresponding anatomical regions instead of the whole image, which is beneficial for textual and visual features alignment. Therefore, we propose a novel anatomy-enhanced multimodal model to promote impression generation. In detail, we first construct a set of rules to extract anatomies and put these prompts into each sentence to highlight anatomy characteristics. Then, two separate encoders are applied to extract features from the radiograph and findings. Afterward, we utilize a contrastive learning module to align these two representations at the overall level and use a co-attention to fuse them at the sentence level with the help of anatomy-enhanced sentence representation. The experimental results on two benchmark datasets confirm the effectiveness of the proposed method, which achieves state-of-the-art results.

Improving Radiology Summarization with Radiograph and Anatomy Prompts

A major challenge in the field of Text Generation is evaluation: Human evaluations are cost-intensive, and automated metrics often display considerable disagreements with human judgments. In this paper, we propose to apply automated metrics for Text Generation in a preference-based evaluation protocol. The protocol features a statistical model that incorporates various levels of uncertainty to account for the error-proneness of the metrics. We show that existing metrics are generally over-confident in assigning significant differences between systems. As a remedy, the model allows to combine human ratings with automated ratings. We show that it can reduce the required amounts of human ratings to arrive at robust and statistically significant results by more than 50%, while yielding the same evaluation outcome as the pure human evaluation in 95% of cases. We showcase the benefits of the evaluation protocol for three text generation tasks: dialogue systems, machine translation, and text summarization.

<iframe src="https://app.sli.do/event/jQes3hFZo3TYjGXEHm6w95/embed/polls/4e77c7a4-edfe-49b6-ad88-ffaba6c4406e" width="300" height="400"></iframe>

Correction of Errors in Preference Ratings from Automated Metrics for Text Generation

Temporal Knowledge Graph Completion aims to complete missing entities or relations under temporal constraints. Previous tensor decomposition-based models for TKGC only independently consider the combination of one single relation with one single timestamp, ignoring the global nature of the embedding. We propose a Frequency Attention (FA) model to capture the global temporal dependencies between one relation and the entire timestamp. Specifically, we use Discrete Cosine Transform (DCT) to capture the frequency of the timestamp embedding and further compute the frequency attention weight to scale embedding. Meanwhile, the previous temporal tucker decomposition method uses a simple norm regularization to constrain the core tensor, which limits the optimization performance. Thus, we propose Orthogonal Regularization (OR) variants for the core tensor, which can limit the non-superdiagonal elements of the 3-rd core tensor. Experiments on three standard TKGC datasets demonstrate that our method outperforms the state-of-the-art results on several metrics. The results suggest that the direct-current component is not the best feature for TKG representation learning. Additional analysis shows the effectiveness of our FA and OR models, even with smaller embedding dimensions.

Tucker Decomposition with Frequency Attention for Temporal Knowledge Graph Completion

Information extraction (IE) systems aim to automatically extract structured information, such as named entities, relations between entities, and events, from unstructured texts. While most existing work addresses a particular IE task, universally modeling various IE tasks with one model has achieved great success recently. Despite their success, they employ a one-stage learning strategy, i.e., directly learning to extract the target structure given the input text, which contradicts the human learning process. In this paper, we propose a unified easy-to-hard learning framework consisting of three stages, i.e., the easy stage, the hard stage, and the main stage, for IE by mimicking the human learning process. By breaking down the learning process into multiple stages, our framework facilitates the model to acquire general IE task knowledge and improve its generalization ability. Extensive experiments across four IE tasks demonstrate the effectiveness of our framework. We achieve new state-of-the-art results on 13 out of 17 datasets.

Easy-to-Hard Learning for Information Extraction

Efficient utilisation of both intra- and extra-textual context remains one of the critical gaps between machine and human translation. Existing research has primarily focused on providing individual, well-defined types of context in translation, such as the surrounding text or discrete external variables like the speaker's gender. This work introduces MTCue, a novel neural machine translation (NMT) framework that interprets all context (including discrete variables) as text. MTCue learns an abstract representation of context, enabling transferability across different data settings and leveraging similar attributes in low-resource scenarios. With a focus on a dialogue domain with access to document and metadata context, we extensively evaluate MTCue in four language pairs in both translation directions. Our framework demonstrates significant improvements in translation quality over a parameter-matched non-contextual baseline, as measured by BLEU (+0.88) and Comet (+1.58). Moreover, MTCue significantly outperforms a "tagging" baseline at translating English text. Analysis reveals that the context encoder of MTCue learns a representation space that organises context based on specific attributes, such as formality, enabling effective zero-shot control. Pre-training on context embeddings also improves MTCue's few-shot performance compared to the "tagging" baseline. Finally, an ablation study conducted on model components and contextual variables further supports the robustness of MTCue for context-based NMT.

MTCue: Learning Zero-Shot Control of Extra-Textual Attributes by Leveraging Unstructured Context in Neural Machine Translation

In real-world systems, scaling has been critical for improving the translation quality in autoregressive translation (AT), which however has not been well studied for non-autoregressive translation (NAT). In this work, we bridge the gap by systematically studying the impact of scaling on NAT behaviors. Extensive experiments on six WMT benchmarks over two advanced NAT models show that scaling can alleviate the commonly-cited weaknesses of NAT models, resulting in better translation performance. To reduce the side-effect of scaling on decoding speed, we empirically investigate the impact of NAT encoder and decoder on the translation performance. Experimental results on the large-scale WMT20 En-De show that the asymmetric architecture (e.g. bigger encoder and smaller decoder) can achieve comparable performance with the scaling model, while maintaining the superiority of decoding speed with standard NAT models. To this end, we establish a new benchmark by validating scaled NAT models on the scaled dataset, which can be regarded as a strong baseline for future works. We release code and system outputs at https://github.com/DeepLearnXMU/Scaling4NAT.

Revisiting Non-Autoregressive Translation at Scale

Recent advances in QA pair generation (QAG) have raised interest in applying this technique to the educational field. However, the diversity of QA types remains a challenge despite its contributions to comprehensive learning and assessment of children. In this paper, we propose a QAG framework that enhances QA type diversity by producing different interrogative sentences and implicit/explicit answers. Our framework comprises a QFS-based answer generator, an iterative QA generator, and a relevancy-aware ranker. The two generators aim to expand the number of candidates while covering various types. The ranker trained on the in-context negative samples clarifies the top-N outputs based on the ranking score. Extensive evaluations and detailed analyses demonstrate that our approach outperforms previous state-of-the-art results by significant margins, achieving improved diversity and quality. Our task-oriented processes are consistent with real-world demand, which highlights our system's high applicability.

Towards Diverse and Effective Question-Answer Pair Generation from Children Storybooks

Active learning (AL) aims to reduce labeling costs by querying the examples most beneficial for model learning. While the effectiveness of AL for fine-tuning transformer-based pre-trained language models (PLMs) has been demonstrated, it is less clear to what extent the AL gains obtained with one model transfer to others. We consider the problem of transferability of actively acquired datasets in text classification and investigate whether AL gains persist when a dataset built using AL coupled with a specific PLM is used to train a different PLM. We link the AL dataset transferability to the similarity of instances queried by the different PLMs and show that AL methods with similar acquisition sequences produce highly transferable datasets regardless of the models used. Additionally, we show that the similarity of acquisition sequences is influenced more by the choice of the AL method than the choice of the model.

<iframe src="https://app.sli.do/event/jQes3hFZo3TYjGXEHm6w95/embed/polls/ab7c6b95-eb31-4b79-a186-b143fe374418" width="300" height="400"></iframe>

On Dataset Transferability in Active Learning for Transformers

Large language models are very resource intensive, both financially and environmentally, and require an amount of training data which is simply unobtainable for the majority of NLP practitioners. Previous work has researched the scaling laws of such models, but optimal ratios of model parameters, dataset size, and computation costs focused on the large scale. In contrast, we analyze the effect those variables have on the performance of language models in constrained settings, by building three lightweight BERT models (16M/51M/124M parameters) trained over a set of small corpora (5M/25M/125M words).We experiment on four languages of different linguistic characteristics (Basque, Spanish, Swahili and Finnish), and evaluate the models on MLM and several NLU tasks. We conclude that the power laws for parameters, data and compute for low-resource settings differ from the optimal scaling laws previously inferred, and data requirements should be higher. Our insights are consistent across all the languages we study, as well as across the MLM and downstream tasks. Furthermore, we experimentally establish when the cost of using a Transformer-based approach is worth taking, instead of favouring other computationally lighter solutions.

<iframe src="https://app.sli.do/event/jQes3hFZo3TYjGXEHm6w95/embed/polls/76da64a5-fde5-4cb8-840c-e429c8d31810" width="300" height="400"></iframe>

Scaling Laws for BERT in Low-Resource Settings

The task of textual geolocation retrieving the coordinates of a place based on a free-form language description calls for not only grounding but also natural language understanding and geospatial reasoning. Even though there are quite a few datasets in English used for geolocation, they are currently based on open-source data (Wikipedia and Twitter), where the location of the described place is mostly implicit, such that the location retrieval resolution is limited. Furthermore, there are no datasets available for addressing the problem of textual geolocation in morphologically rich and resource-poor languages, such as Hebrew. In this paper, we present the Hebrew Geo-Location (HeGeL) corpus, designed to collect literal place descriptions and analyze lingual geospatial reasoning. We crowdsourced 5,649 literal Hebrew place descriptions of various place types in three cities in Israel. Qualitative and empirical analysis show that the data exhibits abundant use of geospatial reasoning and requires a novel environmental representation.

Downloads

Next from ACL 2023

Improving Radiology Summarization with Radiograph and Anatomy Prompts

Similar lecture

Distilling Step-by-Step! Outperforming Larger Language Models with Less Training Data and Smaller Model Sizes