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.

ACL 2023

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

natural language rationales

chain-of-thought reasoning

large language models

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.

workshop paper

Argumentation is an important means of communication. For describing especially arguments about consequences, the notion of effect relations has been introduced recently. We propose a method to extract effect relations from large text resources and apply it on encyclopedic and argumentative texts. By connecting the extracted relations, we generate a knowledge graph which we call effect graph. For evaluating the effect graph, we perform crowd and expert annotations and create a novel dataset. We demonstrate a possible use case of the effect graph by proposing a method for explaining arguments from consequences.

Effect Graph: Effect Relation Extraction for Explanation Generation

We approach competitive-level programming problem-solving as a compos- ite task of reasoning and code generation. We propose a novel method to automatically anno- tate natural language explanations to <problem, solution> pairs. We show that despite poor per- formance in solving competitive-level program- ming problems, state-of-the-art LLMs exhibit a strong capacity in describing and explaining solutions. Our explanation generation method- ology can generate a structured solution expla- nation for the problem containing descriptions and analysis. To evaluate the quality of the annotated explanations, we examine their effec- tiveness in two aspects: 1) satisfying the human programming expert who authored the oracle solution, and 2) aiding LLMs in solving prob- lems more effectively. The experimental results on the CodeContests dataset demonstrate that while LLM GPT3.5’s and GPT-4’s abilities in describing the solution are comparable, GPT-4 shows a better understanding of the key idea behind the solution.

Explaining Competitive-Level Programming Solutions using LLMs

Current natural language systems designed for multi-step claim validation typically operate in two phases: retrieve a set of relevant premise statements using heuristics (planning), then generate novel conclusions from those statements using a large language model (deduction). The planning step often requires expensive Transformer operations and does not scale to arbitrary numbers of premise statements. In this paper, we investigate whether an efficient planning heuristic is possible via embedding spaces compatible with deductive reasoning. Specifically, we evaluate whether embedding spaces exhibit a property we call deductive additivity: the sum of premise statement embeddings should be close to embeddings of conclusions based on those premises. We explore multiple sources of off-the-shelf dense embeddings in addition to fine-tuned embeddings from GPT3 and sparse embeddings from BM25. We study embedding models both intrinsically, evaluating whether the property of deductive additivity holds, and extrinsically, using them to assist planning in natural language proof generation. Lastly, we create a dataset, Single-Step Reasoning Contrast (SSRC), to further probe performance on various reasoning types. Our findings suggest that while standard embedding methods frequently embed conclusions near the sums of their premises, they fall short of being effective heuristics and lack the ability to model certain categories of reasoning.

Deductive Additivity for Planning of Natural Language Proofs

The ease and speed of spreading misinformation and propaganda on the Web motivate the need to develop trustworthy technology for detecting fallacies in natural language arguments. However, state-of-the-art language modeling methods exhibit a lack of robustness on tasks like logical fallacy classification that require complex reasoning. In this paper, we propose a Case-Based Reasoning method that classifies new cases of logical fallacy by language-modeling-driven retrieval and adaptation of historical cases. We design four complementary strategies to enrich input representation for our model, based on external information about goals, explanations, counterarguments, and argument structure. Our experiments in in-domain and out-of-domain settings indicate that Case-Based Reasoning improves the accuracy and generalizability of language models. Our ablation studies suggest that representations of similar cases have a strong impact on the model performance, that models perform well with fewer retrieved cases, and that the size of the case database has a negligible effect on the performance. Finally, we dive deeper into the relationship between the properties of the retrieved cases and the model performance.

Case-Based Reasoning with Language Models for Classification of Logical Fallacies

It has been suggested in literature that large pre-trained language models (PLMs) are able to suppress human-level performance for natural language inference (NLI) tasks. However, the failure of learning the underlying generalizations and the inconsistency to small textual perturbations rise doubt about whether models rely on adopting shallow heuristics to guess the correct label. To mitigate this issue, we propose a neural-symbolic contrastive learning framework inspired by Inductive Logic Programming (ILP) to better capture logical relationships from data. Unlike the usual methods for NLI tasks, our approach represents data as logic programs, sets of logic rules. We aim to learn an embedding space in which – the examples share as various as possible textual information with as similar as possible underlying logical meanings that are close together, and vice versa. Experimental results affirm this approach's ability to enhance the model's transferability performance.

Neural-symbolic Contrastive Learning for Cross-domain Inference

We introduce a synthetic dataset called Sentences Involving Complex Compositional Knowledge (SICCK) and a novel analysis that investigates the performance of Natural Language Inference (NLI) models to understand compositionality in logic. We produce 1,304 sentence pairs by modifying 15 examples from the SICK dataset (Marelli et al., 2014). To this end, we modify the original texts using a set of phrases – modifiers that correspond to universal quantifiers, existential quantifiers, negation, and other concept modifiers in Natural Logic (NL) (MacCartney, 2009). We use these phrases to modify the subject, verb, and object parts of the premise and hypothesis. Lastly, we annotate these modified texts with the corresponding entailment labels following NL rules. We conduct a preliminary verification of how well the change in the structural and semantic composition is captured by neural NLI models, in both zero-shot and fine-tuned scenarios. We found that the performance of NLI models under the zero-shot setting is poor, especially for modified sentences with negation and existential quantifiers. After fine-tuning this dataset, we observe that models continue to perform poorly over negation, existential and universal modifiers.

Synthetic Dataset for Evaluating Complex Compositional Knowledge for Natural Language Inference

We introduce STREET, a unified multi-task and multi-domain natural language reasoning and explanation benchmark. Unlike most existing question-answering (QA) datasets, we expect models to not only answer questions, but also produce step-by-step structured explanations describing how premises in the question are used to produce intermediate conclusions that can prove the correctness of a certain answer. We perform extensive evaluation with popular language models such as few-shot prompting GPT-3 and fine-tuned T5. We find that these models still lag behind human performance when producing such structured reasoning steps. We believe this work will provide a way for the community to better train and test systems on multi-step reasoning and explanations in natural language.

STREET: A Multi-Task Structured Reasoning and Explanation Benchmark

Early exiting can reduce the average latency of pre-trained language models (PLMs) via its adaptive inference mechanism and work with other inference speed-up methods like model pruning, thus drawing much attention from the industry. In this work, we propose a novel framework, BADGE, which consists of two off-the-shelf methods for improving PLMs' early exiting. We first address the issues of training a multi-exit PLM, the backbone model for early exiting. We propose the novel architecture of block-wise bypasses, which can alleviate the conflicts in jointly training multiple intermediate classifiers and thus improve the overall performances of multi-exit PLM while introducing negligible additional flops to the model. Second, we propose a novel divergence-based early exiting (DGE) mechanism, which obtains early exiting signals by comparing the predicted distributions of two adjacent layers' exits. Extensive experiments on three proprietary datasets and three GLUE benchmark tasks demonstrate that our method can obtain a better speedup-performance trade-off than the existing baseline methods.\footnote{Code will be made publicly available to the research community upon acceptance.}

BADGE: Speeding Up BERT Inference after Deployment via Block-wise Bypasses and Divergence-based Early Exiting

Evidence plays a crucial role in automated fact-checking. When verifying real-world claims, existing fact-checking systems either assume the evidence sentences are given or use the search snippets returned by the search engine. Such methods ignore the challenges of collecting evidence and may not provide sufficient information to verify real-world claims. Aiming at building a better fact-checking system, we propose to incorporate full text from source documents as evidence and introduce two enriched datasets. The first one is a multilingual dataset, while the second one is monolingual (English). We further develop a latent variable model to jointly extract evidence sentences from documents and perform claim verification. Experiments indicate that including source documents can provide sufficient contextual clues even when gold evidence sentences are not annotated. The proposed system is able to achieve significant improvements upon best-reported models under different settings.

<iframe src="https://app.sli.do/event/pk54NpNnxuhWHUa3vSMXH2/embed/polls/abc5af62-a831-4ab0-84a5-d93c1ab2b9ce" width="300" height="400"></iframe>

Give Me More Details: Improving Fact-Checking with Latent Retrieval

Existing sentence textual similarity benchmark datasets only use a single number to summarize how similar the sentence encoder’s decision is to humans’. However, it is unclear what kind of sentence pairs a sentence encoder (SE) would consider similar. Moreover, existing SE benchmarks mainly consider sentence pairs with low lexical overlap, so it is unclear how the SEs behave when two sentences have high lexical overlap. We introduce a high-quality SE diagnostic dataset, HEROS. HEROS is constructed by transforming an original sentence into a new sentence based on certain rules to form a minimal pair, and the minimal pair has high lexical overlaps. The rules include replacing a word with a synonym, an antonym, a typo, a random word, and converting the original sentence into its negation. Different rules yield different subsets of HEROS. By systematically comparing the performance of over 60 supervised and unsupervised SEs on HEROS, we reveal that most unsupervised sentence encoders are insensitive to negation. We find the datasets used to train the SE are the main determinants of what kind of sentence pairs an SE considers similar. We also show that even if two SEs have similar performance on STS benchmarks, they can have very different behavior on HEROS. Our result reveals the blind spot of traditional STS benchmarks when evaluating SEs.

Next from ACL 2023

Effect Graph: Effect Relation Extraction for Explanation Generation

Similar lecture

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