United Arab Emirates

We propose EntityCS, a method that focuses on Entity-level Code-Switching to capture fine-grained cross-lingual semantics without corrupting syntax. We construct and release an EntityCS corpus with 93 languages based on English Wikipedia and Wikidata, and design novel masking strategies for entity prediction. XLM trained on the constructed EntityCS corpus with the proposed masking strategies shows consistent improvement on entity-centric downstream tasks.

EMNLP 2022

EntityCS: Improving Zero-Shot Cross-lingual Transfer with Entity-Centric Code Switching

entity knowledge

crossingual transfer

code switching

workshop paper

## Welcome to EMNLP 2022!
I am delighted to welcome you to EMNLP 2022! I believe this conference has been complicated beyond any precedent. Over the past year, it’s been thrilling to see the organization team approach each new puzzle with creativity and enthusiasm. We hope that those participating in Abu Dhabi as well as those joining remotely will leave the conference feeling newly inspired by the program and newly connected to our ever-growing community. Following EMNLP 2021 and major NLP conferences since, EMNLP 2022 is “hybrid,” serving both virtual and in-person participants.

Our key innovations for EMNLP 2022 include:

* EMNLP 2022 is “hybrid” in a second sense, as well: we allowed both direct and rolling review paper submissions, building on the pilot experiment of EMNLP 2021, which considered a small number of ARR submissions. 
* Familiar from NAACL but new to EMNLP, we’ve added an industry track.
* During the conference, “portals” will link virtual poster sessions to in-person conference participants during poster sessions each day.
* The first ACL-family conference in the United Arab Emirates.

 *Message from Noah A. Smith, University of Washington and Allen Institute for AI, Seattle, Washington, USA* 
***EMNLP 2022 General Chair***
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/9eec7d4a287ee18c278b08229290aa83.png)](https://drive.google.com/file/d/1OlPv6QBeo62VVTughj2jkiLeyHd1WnUt/view)
 
[![](https://assets.underline.io/uploads/markdown_image/1/image/a3db7a768409f05192210d98601edb25.png)](https://emnlp2022.rocket.chat/)

To access this site you need to register. Please register [here](https://2022.emnlp.org/registration/).

Register here

Welcome!
EMNLP 2022 will take place in Abu Dhabi from December 7th to December 11th, 2022. And it will be held in hybrid mode, both online and offline.

Automatic question generation (AQG) is the task of generating a question from a given passage and an answer. Most existing AQG methods aim at encoding the passage and the answer to generate the question. However, limited work has focused on modeling the correlation between the target answer and the generated question. Moreover, unseen or rare word generation has not been studied in previous works. In this paper, we propose a novel approach which incorporates question generation with its dual problem, question answering, into a unified primal-dual framework. Specifically, the question generation component consists of an encoder that jointly encodes the answer with the passage, and a decoder that produces the question. The question answering component then re-asks the generated question on the passage to ensure that the target answer is obtained. We further introduce a knowledge distillation module to improve the model generalization ability. We conduct an extensive set of experiments on SQuAD and HotpotQA benchmarks. Experimental results demonstrate the superior performance of the proposed approach over several state-of-the-art methods.

Learning to Generate Question by Asking Question: A Primal-Dual Approach with Uncommon Word Generation

Prompting has shown impressive success in enabling large pre-trained language models (LMs) to perform diverse NLP tasks, especially with only few downstream data. Automatically finding the optimal prompt for each task, however, is challenging. Most existing work resorts to tuning *soft* prompts (e.g., embeddings) which fall short of interpretability, reusability across LMs, and applicability when gradients are not accessible. *Discrete* prompts, on the other hand, are difficult to optimize, and are often created by "enumeration (e.g., paraphrasing)-then-selection" heuristics that do not explore the prompt space systematically. This paper proposes RLPrompt, an efficient discrete prompt optimization approach with reinforcement learning (RL). RLPrompt formulates a parameter-efficient policy network that generates the optimized discrete prompt after training with reward. To harness the complex and stochastic reward signals from the large LM environment, we incorporate effective reward stabilization that substantially enhances training efficiency. RLPrompt is flexibly applicable to different types of LMs, such as masked (e.g., BERT) and left-to-right models (e.g., GPTs), for both classification and generation tasks. Experiments on few-shot classification and unsupervised text style transfer show superior performance over a wide range of existing fine-tuning or prompting methods. Interestingly, the resulting optimized prompts are often ungrammatical gibberish text; and surprisingly, those gibberish prompts are transferrable between different LMs to retain significant performance, indicating that LM prompting may not follow human language patterns.

RLPrompt: Optimizing Discrete Text Prompts with Reinforcement Learning

Given an input sequence (or prefix), modern language models often assign high probabilities to output sequences that are repetitive, incoherent, or irrelevant to the prefix; as such, model-generated text also contains such artifacts. To address these issues we present RankGen, a 1.2B parameter encoder model for English that scores model generations given a prefix. RankGen can be flexibly incorporated as a scoring function in beam search and used to decode from any pretrained language model. We train RankGen using large-scale contrastive learning to map a prefix close to the ground-truth sequence that follows it and far away from two types of negatives: (1) random sequences from the same document as the prefix, and (2) sequences generated from a large language model conditioned on the prefix. Experiments across four different language models (345M-11B parameters) and two domains show that RankGen significantly outperforms decoding algorithms like nucleus, top-k, and typical sampling on both automatic metrics (85.0 vs 77.3 MAUVE) as well as human evaluations with English writers (74.5% human preference over nucleus sampling). Analysis reveals that RankGen outputs are more relevant to the prefix and improve continuity and coherence compared to baselines. We release our model checkpoints, code, and human preference data with explanations to facilitate future research.

RankGen: Improving Text Generation with Large Ranking Models

Generative models of code, pretrained on large corpora of programs, have shown great success in translating natural language to code (Chen et al., 2021; Austin et al., 2021; Li et al., 2022, inter alia). While these models do not explicitly incorporate program semantics (i.e., execution results) during training, they are able to generate correct solutions for many problems. However, choosing a single correct program from a generated set for each problem remains challenging. In this work, we introduce execution result--based minimum Bayes risk decoding (MBR-EXEC) for program selection and show that it improves the few-shot performance of pretrained code models on natural-language-to-code tasks. We select output programs from a generated candidate set by marginalizing over program implementations that share the same semantics. Because exact equivalence is intractable, we execute each program on a small number of test inputs to approximate semantic equivalence. Across datasets, execution or simulated execution significantly outperforms the methods that do not involve program semantics. We find that MBR-EXEC consistently improves over all execution-unaware selection methods, suggesting it as an effective approach for natural language to code translation.

Natural Language to Code Translation with Execution

Social media has increasingly played a key role in emergency response: first responders can use public posts to better react to ongoing crisis events and deploy the necessary resources where they are most needed. Timeline extraction and abstractive summarization are critical technical tasks to leverage large numbers of social media posts about events. Unfortunately, there are few datasets for benchmarking technical approaches for those tasks. This paper presents \datasetname{}, the largest dataset of local crisis event timelines available to date. \datasetname{} contains 1,000 crisis event timelines across four domains: wildfires, local fires, traffic, and storms. We built \datasetname{} using a semi-automated cluster-then-refine approach to collect data from the public Twitter stream. 
Our initial experiments indicate a significant gap between the performance of strong baselines compared to the human performance on both tasks.
Our dataset, code, and models are publicly available (https://github.com/CrisisLTLSum/CrisisTimelines).

CrisisLTLSum: A Benchmark for Local Crisis Event Timeline Extraction and Summarization

A deployed question answering (QA) model can easily fail when the test data has a distribution shift compared to the training data. Robustness tuning (RT) methods have been widely studied to enhance model robustness against distribution shifts before model deployment. However, can we improve a model after deployment? To answer this question, we evaluate test-time adaptation (TTA) to improve a model after deployment. We first introduce ColdQA, a unified evaluation benchmark for robust QA against text corruption and changes in language and domain. We then evaluate previous TTA methods on ColdQA and compare them to RT methods. We also propose a novel TTA method called online imitation learning (OIL). Through extensive experiments, we find that TTA is comparable to RT methods, and applying TTA after RT can significantly boost the performance on ColdQA. Our proposed OIL improves TTA to be more robust to variation in hyper-parameters and test distributions over time.

Robust Question Answering against Distribution Shifts with Test-Time Adaption: An Empirical Study

Contrary to the notions of language as an amodal system, natural human communication is multimodal and combines speech, gestures, writing, and pictures. To account for this, recent work has proposed that our vocal, bodily, and graphic modalities persist in parallel in a multimodal language faculty, and both unimodal and multimodal expressions arise out of emergent states of a shared architecture. Such a model carries different expectations for the ways in which modalities may be similar or different from each other, and how they may interact. I will highlight these properties specifically for our graphic modality, which I argue can manifest in full visual languages when displaying both a systematic lexicon and complex grammar. I will use analysis of a corpus of several hundred annotated comics to show distinctive patterns that suggest they are drawn in different visual languages. Yet, I will also show that consistent “universal” linguistic principles persist across this structural diversity. Finally, I will argue that a multimodal language faculty requires us to change our conception of linguistic relativity, and I will show how subtle structures of spoken languages permeate across to visual languages. Altogether, this work argues for a multimodal basis of linguistic structure, and heralds a reconsideration of what constitutes the language system.

The multimodal language faculty and the visual languages of comics

Causal inference is becoming an increasingly important topic in deep learning, with the potential to help with critical deep learning problems such as model robustness, interpretability, and fairness. In addition, causality is naturally widely used in various disciplines of science, to discover causal relationships among variables and estimate causal effects of interest. In this tutorial, we introduce the fundamentals of causal discovery and causal effect estimation to the natural language processing (NLP) audience, provide an overview of causal perspectives to NLP problems, and aim to inspire novel approaches to NLP further. This tutorial is inclusive to a variety of audiences and is expected to facilitate the community’s developments in formulating and addressing new, important NLP problems in light of emerging causal principles and methodologies.

Tutorial on Causal Inference for Natural Language Processing

State-of-the-art language models in NLP perform best when fine-tuned even on small datasets, but due to their increasing size, fine-tuning and downstream usage have become extremely compute-intensive. Being able to efficiently and effectively fine-tune the largest pre-trained models is thus key in order to reap the benefits of the latest advances in NLP. In this tutorial, we provide a comprehensive overview of parameter-efficient fine-tuning methods. We highlight their similarities and differences by presenting them in a unified view. We explore the benefits and usage scenarios of a neglected property of such parameter-efficient models—modularity—such as composition of modules to deal with previously unseen data conditions. We finally highlight how both properties—parameter efficiency and modularity—can be useful in the real-world setting of adapting pre-trained models to under-represented languages and domains with scarce annotated data for several downstream applications. 

Modular and Parameter-Efficient Fine-Tuning for NLP Models

The autoregressive mechanism makes it convenient to train sequence generation models, but the generation is slow since autoregressive models can only generate the target sequence word-by-word. Non-autoregressive (NAR) models generate all tokens in parallel by removing the sequential dependencies within the target sequence. While NAR models have received much attention in sequence generation tasks, it is non-trivial for NAR models to obtain satisfactory performance, where the major challenge is to capture the multi-modal data distribution. In this tutorial, we will provide a comprehensive introduction to non-autoregressive sequence generation in four sections. First, we will briefly introduce the background of sequence generation, give the motivation of NAR generation and the multi-modality challenge faced by NAR models. Second, we will introduce efforts to improve NAR generation. We focus on non-autoregressive translation in this paper, which includes many representative methods from different aspects: training objectives, data distillation, latent modeling, upsampling, improved decoding approaches, etc. Third, we will introduce the application of NAR models on a wide range of sequence generation tasks, including language modeling, speech recognition, text-to-speech, etc. Finally, we will conclude this tutorial by summarizing the strengths and challenges of NAR models, and discussing current concerns and future directions of NAR generation. 


EntityCS: Improving Zero-Shot Cross-lingual Transfer with Entity-Centric Code Switching

Downloads

Next from EMNLP 2022

Learning to Generate Question by Asking Question: A Primal-Dual Approach with Uncommon Word Generation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

EntityCS: Improving Zero-Shot Cross-lingual Transfer with Entity-Centric Code Switching

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from EMNLP 2022

Learning to Generate Question by Asking Question: A Primal-Dual Approach with Uncommon Word Generation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads