United States

Transformer models yield impressive results on many NLP and sequence modeling tasks. Remarkably, Transformers can handle long sequences, which allows them to produce long coherent outputs: entire paragraphs produced by GPT-3 or well-structured images produced by DALL-E. These large language models are impressive but also very inefficient and costly, which limits their applications and accessibility. We postulate that having an explicit hierarchical architecture is the key to Transformers that efficiently handle long sequences. To verify this claim, we first study different ways to downsample and upsample activations in Transformers so as to make them hierarchical. We use the best performing upsampling and downsampling layers to create Hourglass - a hierarchical Transformer language model. Hourglass improves upon the Transformer baseline given the same amount of computation and can yield the same results as Transformers more efficiently. In particular, Hourglass sets new state-of-the-art for Transformer models on the ImageNet32 generation task and improves language modeling efficiency on the widely studied enwik8 benchmark.

NAACL 2022

Hierarchical Transformers Are More Efficient Language Models

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible. The experience for on-site participants would closely approximate a normal pre-COVID *ACL conference, with 5-6 thematically organized parallel sessions and live Q/A and interactive discussion immediately after the talks. Presentations by virtual participants will be equitably interleaved with those of on-site participants, projected on the auditorium screens as if on-site, and also followed immediately by live Q/A and interactive discussion at a time during reasonable waking hours for the virtual presenter. For all participants, on-site and virtual, who are unable to attend a session due to either time-zone issues or because they are participating in another session live, talk recordings and slides will be available online at a minimum after the live presentation (and in many cases before as well), and questions may be submitted in advance on session-specific discussion boards and answered live in session with the usual visual aids if desired.

<iframe style="width:700px;height:400px" src="https://online.fliphtml5.com/ebtyf/ceby/" seamless="seamless" scrolling="no" frameborder="0" allowtransparency="true" allowfullscreen="true" ></iframe>

Please Note: The EMNLP registration system is not currently connected to the underline site as we are still in the process of building out EMNLP 2021. You will receive access instructions from underline the week of November 1st. 

Access is given only to EMNLP upon registration, if you have not registered please do so [here](https://2021.emnlp.org/registration).

Registered attendees will receive access the week of November 1st.

EMNLP 2021

EMNLP 2021 is planned to be a hybrid event in Punta Cana, Dominican Republic, with both on-site and fully virtual participation possible.

poster

## Welcome to NAACL 2022!
The Annual Meeting of the North American Chapter of the Association for Computational Linguistics (NAACL) is taking place July 10-15, 2022 as a hybrid event, in Seattle, WA and online. We are happy to welcome all of you to conference! 
 
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, keynote presentations, a plenary panel on the place of linguistics and symbolic structures in NLP, and an Industrial Track panel on Careers in NLP.
 
Posters (including Findings of NAACL 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 and will have an oral session and a poster session.
 
The program also features 6 Tutorials and 26 Workshops.
 
 We wish you a wonderful conference!
 
[The NAACL 2022 Organizing Committee](https://2022.naacl.org/committees/organization/)

2022 Annual Conference of the North American Chapter of the Association for Computational Linguistics

Pretrained language models based on the transformer architecture have shown great success in NLP.
Textual training data often comes from the web and is thus tagged with time-specific information, but most language models ignore this information.
They are trained on the textual data alone, limiting their ability to generalize temporally.
In this work, we extend the key component of the transformer architecture, i.e., the self-attention mechanism, and propose temporal attention - a time-aware self-attention mechanism.
Temporal attention can be applied to any transformer model and requires the input texts to be accompanied with their relevant time points. 
This mechanism allows the transformer to capture this temporal information and create time-specific contextualized word representations.
We leverage these representations for the task of semantic change detection; we apply our proposed mechanism to BERT and experiment on three datasets in different languages (English, German, and Latin) that also vary in time, size, and genre.
Our proposed model achieves state-of-the-art results on all the datasets.

Temporal Attention for Language Models

In this work, we explore how to train task-specific language models aimed towards learning rich representation of keyphrases from text documents. We experiment with different masking strategies for pre-training transformer language models (LMs) in discriminative as well as generative settings. In the discriminative setting, we introduce a new pre-training objective - Keyphrase Boundary Infilling with Replacement (KBIR), showing large gains in performance (upto 9.26 points in F1) over SOTA, when the LM pre-trained using KBIR is fine-tuned for the task of keyphrase extraction. In the generative setting, we introduce a new pre-training setup for BART - KeyBART, that reproduces the keyphrases related to the input text in the CatSeq format, instead of the denoised original input. This also led to gains in performance (upto 4.33 points in F1@M) over SOTA for keyphrase generation. Additionally, we also fine-tune the pre-trained language models on named entity recognition (NER), question answering (QA), relation extraction (RE), abstractive summarization and achieve comparable performance with that of the SOTA, showing that learning rich representation of keyphrases is indeed beneficial for many other fundamental NLP tasks.

Learning Rich Representation of Keyphrases from Text

Recent work has shown that either (1) increasing the input length or (2) increasing model size can improve the performance of Transformer-based neural models. In this paper, we present LongT5, a new model that explores the effects of scaling both the input length and model size at the same time. 
Specifically, we integrate attention ideas from long-input transformers (ETC), and adopt pre-training strategies from summarization pre-training (PEGASUS) into the scalable T5 architecture. The result is a new attention mechanism we call Transient Global (TGlobal), which mimics ETC's local/global attention mechanism, but without requiring additional side-inputs. We are able to achieve state-of-the-art results on several summarization and question answering tasks, as well as outperform the original T5 models on these tasks. We have open  sourced our architecture and training code, as well as our pre-trained model checkpoints.

LongT5: Efficient Text-To-Text Transformer for Long Sequences

Language models (LMs) are typically trained once on a large-scale corpus and used for years without being updated. However, in a dynamic world, new entities constantly arise. We propose a framework to analyze what LMs can infer about new entities that did not exist when the LMs were pretrained. We derive a dataset of entities indexed by their origination date and paired with their English Wikipedia articles, from which we can find sentences about each entity. We evaluate LMs' perplexity on masked spans within these sentences. We show that models more informed about the entities, such as those with access to a textual definition of them, achieve lower perplexity on this benchmark. Our experimental results demonstrate that making inferences about new entities remains difficult for LMs. Given its wide coverage on entity knowledge and temporal indexing, our dataset can be used to evaluate LMs and techniques designed to modify or extend their knowledge. Our automatic data collection pipeline can be easily used to continually update our benchmark.

Entity Cloze By Date: What LMs Know About Unseen Entities

We present a self-supervised pre-training approach for learning rich visual language representations for both handwritten and printed historical document transcription. After supervised fine-tuning of our pre-trained encoder representations for low-resource document transcription on two languages, (1) a heterogeneous set of handwritten Islamicate manuscript images and (2) early modern English printed documents, we show a meaningful improvement in recognition accuracy over the same supervised model trained from scratch with as few as 30 line image transcriptions for training. Our masked language model-style pre-training strategy, where the model is trained to be able to identify the true masked visual representation from distractors sampled from within the same line, encourages learning robust contextualized language representations invariant to scribal writing style and printing noise present across documents.

Lacuna Reconstruction: Self-Supervised Pre-Training for Low-Resource Historical Document Transcription

We introduce MTG, a new benchmark suite for training and evaluating multilingual text generation. It is the first-proposed multilingual multiway text generation dataset with the largest human-annotated data (400k). It includes four generation tasks (story generation, question generation, title generation and text summarization) across five languages (English, German, French, Spanish and Chinese). The multiway setup enables testing knowledge transfer capabilities for a model across languages and tasks. Using MTG, we train and analyze several popular multilingual generation models from different aspects. Our benchmark suite fosters model performance enhancement with more human-annotated parallel data. It provides comprehensive evaluations with diverse generation scenarios. Code and data are available at \urlhttps://github.com/zide05/MTG.

MTG: A Benchmark Suite for Multilingual Text Generation

The current state-of-the-art for few-shot cross-lingual transfer learning first trains on abundant labeled data in the source language and then fine-tunes with a few examples on the target language, termed target-adapting. Though this has been demonstrated to work on a variety of tasks, in this paper we show some deficiencies of this approach and propose a one-step mixed training method that trains on both source and target data with stochastic gradient surgery, a novel gradient-level optimization. Unlike the previous studies that focus on one language at a time when target-adapting, we use one model to handle all target languages simultaneously to avoid excessively language-specific models. Moreover, we discuss the unreality of utilizing large target development sets for model selection in previous literature. We further show that our method is both development-free for target languages, and is also able to escape from overfitting issues. We conduct a large-scale experiment on 4 diverse NLP tasks across up to 48 languages. Our proposed method achieves state-of-the-art performance on all tasks and outperforms target-adapting by a large margin, especially for languages that are linguistically distant from the source language, e.g., 7.36% F1 absolute gain on average for the NER task, up to 17.60% on Punjabi.

Por Qué Não Utiliser Alla Språk? Mixed Training with Gradient Optimization in Few-Shot Cross-Lingual Transfer

In this paper, we introduce DOCmT5, a multilingual sequence-to-sequence language model pretrained with large-scale parallel documents. While previous approaches have focused on leveraging sentence-level parallel data, we try to build a general-purpose pretrained model that can understand and generate long documents.  We propose a simple and effective pretraining objective - Document reordering Machine Translation (DrMT), in which the input documents that are shuffled and masked need to be translated. DrMT brings consistent improvements over strong baselines on a variety of document-level generation tasks, including over 12 BLEU points for seen-language pair document-level MT, over 7 BLEU points for unseen-language-pair document-level MT and over 3 ROUGE-1 points for seen-language pair cross-lingual summarization. We achieve state-of-the-art (SOTA) on WMT20 De-En and IWSLT15 Zh-En document translation tasks. We also conduct extensive analysis on various factors for document pretraining, including (1) the effects of pretraining data quality and (2) The effects of combining mono-lingual and cross-lingual pretraining. We plan to make our model checkpoints publicly available.

DOCmT5: Document-Level Pretraining of Multilingual Language Models

Translate-train or few-shot cross-lingual transfer can be used to improve the zero-shot performance of multilingual pretrained language models.  Few-shot utilizes high-quality low-quantity samples (often manually translated from the English corpus ). Translate-train employs a machine translation of the English corpus, resulting in samples with lower quality that could be scaled to high quantity. Given the lower cost and higher availability of machine translation compared to manual professional translation, it is important to systematically compare few-shot and translate-train, understand when each has an advantage,  and investigate how to choose the shots to translate in order to increase the few-shot gain.  This work aims to fill this gap: we compare and quantify the performance gain of few-shot vs. translate-train using three different base models and a varying number of samples for three tasks/datasets (XNLI, PAWS-X, XQuAD) spanning 17 languages. We show that scaling up the training data using machine translation gives a larger gain compared to using the small-scale (higher-quality) few-shot data. When few-shot is beneficial, we show that there are random sets of samples that perform better across languages and that the performance on English and on the machine-translation of the samples can both be used to choose the shots to manually translate for an increased few-shot gain.

How to Translate Your Samples and Choose Your Shots? Analyzing Translate-train & Few-shot Cross-lingual Transfer

We propose a novel open-domain question-answering dataset based on the Common Crawl project. With a previously unseen number of around 130 million multilingual question-answer pairs (including about 60 million English data-points), we use our large-scale, natural, diverse and high-quality corpus to in-domain pre-train popular language models for the task of question-answering. In our experiments, we find that our Common Crawl Question Answering dataset (CCQA) achieves promising results in zero-shot, low resource and fine-tuned settings across multiple tasks, models and benchmarks.

Hierarchical Transformers Are More Efficient Language Models

Downloads

Next from NAACL 2022

Temporal Attention for Language Models

Similar lecture

IndoNLG: Benchmark and Resources for Evaluating Indonesian Natural Language Generation