Ireland

Neural language models have attracted a lot of attention in the past few years. More and more researchers are getting intrigued by how language models encode commonsense, specifically what kind of commonsense they understand, and why they do. This paper analyzed neural language models&#39; understanding of commonsense pragmatics (i.e., implied meanings) through human behavioral and neurophysiological data. These psycholinguistic tests are designed to draw conclusions based on predictive responses in context, making them very well suited to test word-prediction models such as BERT in natural settings. They can provide the appropriate prompts and tasks to answer questions about linguistic mechanisms underlying predictive responses. This paper adopted psycholinguistic datasets to probe language models&#39; commonsense reasoning. Findings suggest that GPT-3&#39;s performance was mostly at chance in the psycholinguistic tasks. We also showed that DistillBERT had some understanding of the (implied) intent that&#39;s shared among most people. Such intent is implicitly reflected in the usage of conversational implicatures and presuppositions. Whether or not fine-tuning improved its performance to human-level depends on the type of commonsense reasoning.

ACL 2022

Psycholinguistic Diagnosis of Language Models’ Commonsense Reasoning

implicatures

commonsense pragmatics

implicit meaning

presuppositions

cognitive plausibility

language models

# Welcome everyone to ACL 2022!

The 60th Annual Meeting of the Association for Computational Linguistics is taking place May 22-27, 2022 as a hybrid event, in Dublin and online. We are happy to welcome all of you to this anniversary edition with an almost 50-50 in-person and virtual participation. 
The main conference program features oral presentations, in-person and virtual posters and demo sessions, a plenary session for our best paper presentations and awards, three amazing keynote events and two new initiatives of invited talks: Spotlight Talks for Young Rising Stars (STIRS) and The Next Big Idea Talks. Posters (including Findings of ACL 2022) and demos are grouped by areas for both the in-person and the virtual sessions. For the virtual component, the talks will be on Zoom and the posters and the demos will be in GatherTown. The Student Research Workshop will have an oral session and a poster session as part of Poster Session 1. The program also features eight Tutorials and 28 Workshops. 

 
We wish you a wonderful conference! 
[**The ACL 2022 Organizing Committee**](https://www.2022.aclweb.org/organisers)
 
[**Conference Handbook**](https://drive.google.com/file/d/1_BUCMfhMVrjG9E2e71aHdHeE28KSje0l/view?usp=sharing) 
[**Mini Handbook**](https://drive.google.com/file/d/1qlBKl0wzmlVF1oCeMQl3BahLd9nLP5Ce/view?usp=sharing) 
[**Posters and Demo guides**](https://drive.google.com/file/d/1UucMAoCNncIOaH1rMMDa0owuG9qgvJTG/view?usp=sharing)

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 (NLP). 

Neural language models have attracted a lot of attention in the past few years. More and more researchers are getting intrigued by how language models encode commonsense, specifically what kind of commonsense they understand, and why they do. This paper analyzed neural language models' understanding of commonsense pragmatics (i.e., implied meanings) through human behavioral and neurophysiological data. These psycholinguistic tests are designed to draw conclusions based on predictive responses in context, making them very well suited to test word-prediction models such as BERT in natural settings. They can provide the appropriate prompts and tasks to answer questions about linguistic mechanisms underlying predictive responses. This paper adopted psycholinguistic datasets to probe language models' commonsense reasoning. Findings suggest that GPT-3's performance was mostly at chance in the psycholinguistic tasks. We also showed that DistillBERT had some understanding of the (implied) intent that's shared among most people. Such intent is implicitly reflected in the usage of conversational implicatures and presuppositions. Whether or not fine-tuning improved its performance to human-level depends on the type of commonsense reasoning.

workshop paper

Large-scale visual-linguistic pre-training aims to capture the generic representations from multimodal features, which are essential for downstream vision-language tasks. Existing methods mostly focus on learning the seman- tic connections between visual objects and lin- guistic content, which tend to be recognition- level information and may not be sufficient for commonsensical reasoning tasks like VCR. In this paper, we propose a novel commonsen- sical vision-language pre-training framework to bridge the gap. We first augment the con- ventional image-caption pre-training datasets with commonsense inferences from a visual- linguistic GPT-2. To pre-train models on image, caption and commonsense inferences together, we propose two new tasks: masked common- sense modeling (MCM) and commonsense type prediction (CTP). To reduce the shortcut effect between captions and commonsense inferences, we further introduce the domain-wise adaptive masking that dynamically adjusts the mask- ing ratio. Experimental results on downstream tasks, VCR and VQA, show the improvement of our pre-training strategy over previous meth- ods. Human evaluation also validates the rel- evance, informativeness, and diversity of the generated commonsense inferences. Overall, we demonstrate the potential of incorporating commonsense knowledge into the conventional recognition-level visual-linguistic pre-training.

Bridging the Gap between Recognition-level Pre-training and Commonsensical Vision-language Tasks

Previous studies have shown the efficacy of knowledge augmentation methods in pretrained language models. However, these methods behave differently across domains and downstream tasks. In this work, we investigate the augmentation of pretrained language models with knowledge graph data in the cause-effect relation classification and commonsense causal reasoning tasks. After automatically verbalizing triples in ATOMIC2020, a wide coverage commonsense reasoning knowledge graph, we continually pretrain BERT and evaluate the resulting model on cause-effect pair classification and answering commonsense causal reasoning questions. Our results show that a continually pretrained language model augmented with commonsense reasoning knowledge outperforms our baselines on two commonsense causal reasoning benchmarks, COPA and BCOPA-CE, and a Temporal and Causal Reasoning (TCR) dataset, without additional improvement in model architecture or using quality-enhanced data for fine-tuning.

Knowledge-Augmented Language Models for Cause-Effect Relation Classification

Recently, models have been shown to predict the effects of unexpected situations, e.g., would cloudy skies help or hinder plant growth? Given a context, the goal of such situational reasoning is to elicit the consequences of a new situation (st) that arises in that context. We propose a method to iteratively build a graph of relevant consequences explicitly in a structured situational graph (st-graph) using natural language queries over a finetuned language model (M). Across multiple domains, CURIE generates st-graphs that humans find relevant and meaningful in eliciting the consequences of a new situation. We show that st-graphs generated by CURIE improve a situational reasoning end task (WIQA-QA) by 3 points on accuracy by simply augmenting their input with our generated situational graphs, especially for a hard subset that requires background knowledge and multi-hop reasoning.

CURIE: An Iterative Querying Approach for Reasoning About Situations

A major issue in open-domain dialogue generation is the agent's tendency to generate repetitive and generic responses. The lack in response diversity has been addressed in recent years via the use of latent variable models, such as the Conditional Variational Auto-Encoder (CVAE), which typically involve learning a latent Gaussian distribution over potential response intents. However, due to latent variable collapse, training latent variable dialogue models are notoriously complex, requiring substantial modification to the standard training process and loss function. Other approaches proposed to improve response diversity also largely entail a significant increase in training complexity. Hence, this paper proposes a Randomized Link (RL) Transformer as an alternative to the latent variable models. The RL Transformer does not require any additional enhancements to the training process or loss function. Empirical results show that, when it comes to response diversity, the RL Transformer achieved comparable performance compared to latent variable models.

A Randomized Link Transformer for Diverse Open-Domain Dialogue Generation

Retailing combines complicated communication skills and strategies to reach an agreement between buyer and seller with identical or different goals. In each transaction a good seller finds an optimal solution by considering his/her own profits while simultaneously considering whether the buyer's needs have been met. In this paper, we manage the retailing problem by mixing cooperation and competition. We present a rich dataset of buyer-seller bargaining in a simulated marketplace in which each agent values goods and utility separately. Various attributes (preference, quality, and profit) are initially hidden from one agent with respect to its role; during the conversation, both sides may reveal, fake, or retain the information uncovered to come to a final decision through natural language. Using this dataset, we leverage transfer learning techniques on a pretrained, end-to-end model and enhance its decision-making ability toward the best choice in terms of utility by means of multi-agent reinforcement learning. An automatic evaluation shows that our approach results in more optimal transactions than human does. We also show that our framework controls the falsehoods generated by seller agents.

Conversational AI for Positive-sum Retailing under Falsehood Control

Existing research studies on cross-sentence relation extraction in long-form multi-party conversations aim to improve relation extraction without considering the explainability of such methods.
This work addresses that gap by focusing on extracting explanations that indicate that a relation exists while using only partially labeled explanations.
We propose our model-agnostic framework, D-REX, a policy-guided semi-supervised algorithm that optimizes for explanation quality and relation extraction simultaneously. We frame relation extraction as a re-ranking task and include relation- and entity-specific explanations as an intermediate step of the inference process. We find that human annotators are 4.2 times more likely to prefer D-REX's explanations over a joint relation extraction and explanation model. 
Finally, our evaluations show that \drex\;is simple yet effective and improves relation extraction performance of strong baseline models by 1.2-4.7%.

D-REX: Dialogue Relation Extraction with Explanations

Data augmentation is a widely employed technique to alleviate the problem of data scarcity. In this work, we propose a prompting-based approach to generate labelled training data for intent classification with off-the-shelf language models (LMs) such as GPT-3. An advantage of this method is that no task-specific LM-fine-tuning for data generation is required; hence the method requires no hyper-parameter tuning and is applicable even when the available training data is very scarce. We evaluate the proposed method in a few-shot setting on four diverse intent classification tasks. We find that GPT-generated data significantly boosts the performance of intent classifiers when intents in consideration are sufficiently distinct from each other. In tasks with semantically close intents, we observe that the generated data is less helpful. Our analysis shows that this is because GPT often generates utterances that belong to a closely-related intent instead of the desired one. We present preliminary evidence that a prompting-based GPT classifier could be helpful in filtering the generated data to enhance its quality.

Data Augmentation for Intent Classification with Off-the-shelf Large Language Models

Personal attributes represent structured information about a person, such as their hobbies, pets, family, likes and dislikes. We introduce the tasks of extracting and inferring personal attributes from human-human dialogue, and analyze the linguistic demands of these tasks. To meet these challenges, we introduce a simple and extensible model that combines an autoregressive language model utilizing constrained attribute generation with a discriminative reranker. Our model outperforms strong baselines on extracting personal attributes as well as inferring personal attributes that are not contained verbatim in utterances and instead requires commonsense reasoning and lexical inferences, which occur frequently in everyday conversation. Finally, we demonstrate the benefit of incorporating personal attributes in social chit-chat and task-oriented dialogue settings.

Extracting and Inferring Personal Attributes from Dialogue

In conversational QA, models have to lever- age information in previous turns to answer up- coming questions. Current approaches, such as Question Rewriting, struggle to extract rele- vant information as the conversation unwinds. We introduce the Common Ground (CG), an approach to accumulate conversational infor- mation as it emerges and select the relevant information at every turn. We show that CG of- fers a more efficient and human-like way to ex- ploit conversational information compared to existing approaches, leading to improvements on Open Domain Conversational QA.

From Rewriting to Remembering: Common Ground for Conversational QA Models

At the heart of improving conversational AI is the open problem of how to evaluate conversations. Issues with automatic metrics are well known (Liu et al., 2016), with human evaluations still considered the gold standard. Unfortunately, how to perform human evaluations is also an open problem: differing data collection methods have varying levels of human agreement and statistical sensitivity, resulting in differing amounts of human annotation hours and labor costs. In this work we compare five different crowdworker-based human evaluation methods and find that different methods are best depending on the types of models compared, with no clear winner across the board. While this highlights the open problems in the area, our analysis leads to advice of when to use which one, and possible future directions.

Psycholinguistic Diagnosis of Language Models’ Commonsense Reasoning

Downloads

Next from ACL 2022

Bridging the Gap between Recognition-level Pre-training and Commonsensical Vision-language Tasks

Similar lecture

AraT5: Text-to-Text Transformers for Arabic Language Generation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES