mDAPT: Multilingual Domain Adaptive Pretraining in a Single Model

Domain adaptive pretraining, i.e. the continued unsupervised pretraining of a language model on domain-specific text, improves the modelling of text for downstream tasks within the domain. Numerous real-world applications are based on domain-specific text, e.g. working with financial or biomedical documents, and these applications often need to support multiple languages. However, large-scale domain-specific multilingual pretraining data for such scenarios can be difficult to obtain, due to regulations, legislation, or simply a lack of language- and domain-specific text. One solution is to train a single multilingual model, taking advantage of the data available in as many languages as possible. In this work, we explore the benefits of domain adaptive pretraining with a focus on adapting to multiple languages within a specific domain. We propose different techniques to compose pretraining corpora that enable a language model to both become domain-specific and multilingual. Evaluation on nine domain-specific datasets---for biomedical named entity recognition and financial sentence classification---covering seven different languages show that a single multilingual domain-specific model can outperform the general multilingual model, and performs close to its monolingual counterpart. This finding holds across two different pretraining methods, adapter-based pretraining and full model pretraining.

A Thorough Evaluation of Task-Specific Pretraining for Summarization

Task-agnostic pretraining objectives like masked language models or corrupted span prediction are applicable to a wide range of NLP downstream tasks (Raffel et al.,2019), but are outperformed by task-specific pretraining objectives like predicting extracted gap sentences on summarization (Zhang et al.,2020). We compare three summarization specific pretraining objectives with the task agnostic corrupted span prediction pretraining in controlled study. We also extend our study to a low resource and zero shot setup, to understand how many training examples are needed in order to ablate the task-specific pretraining without quality loss. Our results show that task-agnostic pretraining is sufficient for most cases which hopefully reduces the need for costly task-specific pretraining. We also report new state-of-the-art number for two summarization task using a T5 model with 11 billion parameters and an optimal beam search length penalty.

How much pretraining data do language models need to learn syntax?

Transformers-based pretrained language models achieve outstanding results in many well-known NLU benchmarks. However, while pretraining methods are very convenient, they are expensive in terms of time and resources. This calls for a study of the impact of pretraining data size on the knowledge of the models. We explore this impact on the syntactic capabilities of RoBERTa, using models trained on incremental sizes of raw text data. First, we use syntactic structural probes to determine whether models pretrained on more data encode a higher amount of syntactic information. Second, we perform a targeted syntactic evaluation to analyze the impact of pretraining data size on the syntactic generalization performance of the models. Third, we compare the performance of the different models on three downstream applications: part-of-speech tagging, dependency parsing and paraphrase identification. We complement our study with an analysis of the cost-benefit trade-off of training such models. Our experiments show that while models pretrained on more data encode more syntactic knowledge and perform better on downstream applications, they do not always offer a better performance across the different syntactic phenomena and come at a higher financial and environmental cost.

NICT Kyoto Submission for the WMT'21 Quality Estimation Task: Multimetric Multilingual Pretraining for Critical Error Detection

This paper presents the NICT Kyoto submission for the WMT'21 Quality Estimation (QE) Critical Error Detection shared task (Task 3). Our approach relies mainly on QE model pretraining for which we used 11 language pairs, three sentence-level and three word-level translation quality metrics. Starting from an XLM-R checkpoint, we perform continued training by modifying the learning objective, switching from masked language modeling to QE oriented signals, before finetuning and ensembling the models. Results obtained on the test set in terms of correlation coefficient and F-score show that automatic metrics and synthetic data perform well for pretraining, with our submissions ranked first for two out of four language pairs. A deeper look at the impact of each metric on the downstream task indicates higher performance for token oriented metrics, while an ablation study emphasizes the usefulness of conducting both self-supervised and QE pretraining.

ECONET: Effective Continual Pretraining of Language Models for Event Temporal Reasoning

While pre-trained language models (PTLMs) have achieved noticeable success on many NLP tasks, they still struggle for tasks that require event temporal reasoning, which is essential for event-centric applications. We present a continual pre-training approach that equips PTLMs with targeted knowledge about event temporal relations. We design self-supervised learning objectives to recover masked-out event and temporal indicators and to discriminate sentences from their corrupted counterparts (where event or temporal indicators got replaced). By further pre-training a PTLM with these objectives jointly, we reinforce its attention to event and temporal information, yielding enhanced capability on event temporal reasoning. This **E**ffective **CON**tinual pre-training framework for **E**vent **T**emporal reasoning (ECONET) improves the PTLMs' fine-tuning performances across five relation extraction and question answering tasks and achieves new or on-par state-of-the-art performances in most of our downstream tasks.

Investigating Post-pretraining Representation Alignment for Cross-Lingual Question Answering

Human knowledge is collectively encoded in the roughly 6500 languages spoken around the world, but it is not distributed equally across languages. Hence, for information-seeking question answering (QA) systems to adequately serve speakers of all languages, they need to operate cross-lingually. In this work we investigate the capabilities of multilingually pretrained language models on cross-lingual QA. We find that explicitly aligning the representations across languages with a post-hoc finetuning step generally leads to improved performance. We additionally investigate the effect of data size as well as the language choice in this fine-tuning step, also releasing a dataset for evaluating cross-lingual QA systems.

Language Model Pretraining and Transfer Learning for Very Low Resource Languages

This paper describes our submission for the shared task on Unsupervised MT and Very Low Resource Supervised MT at WMT 2021. We submitted systems for two language pairs: German ↔ Upper Sorbian (de ↔ hsb) and German-Lower Sorbian (de ↔ dsb). For de ↔ hsb, we pretrain our system using MASS (Masked Sequence to Sequence) objective and then finetune using iterative back-translation. Final finetunng is performed using the parallel data provided for translation objective. For de ↔ dsb, no parallel data is provided in the task, we use final de ↔ hsb model as initialization of the de ↔ dsb model and train it further using iterative back-translation, using the same vocabulary as used in the de ↔ hsb model.

Does Vision-and-Language Pretraining Improve Lexical Grounding?

Linguistic representations derived from text alone have been criticized for their lack of grounding, i.e., connecting words to their meanings in the physical world. Vision-and- Language (VL) models, trained jointly on text and image or video data, have been offered as a response to such criticisms. However, while VL pretraining has shown success on multimodal tasks such as visual question answering, it is not yet known how the internal linguistic representations themselves compare to their text-only counterparts. This paper compares the semantic representations learned via VL vs. text-only pretraining for two recent VL models using a suite of analyses (clustering, probing, and performance on a commonsense question answering task) in a language-only setting. We find that the multimodal models fail to significantly outperform the text-only variants, suggesting that future work is required if multimodal pretraining is to be pursued as a means of improving NLP in general.

Revisiting Pretraining with Adapters

Pretrained language models have served as the backbone for many state-of-the-art NLP results. These models are large and expensive to train. Recent work suggests that continued pretraining on task-specific data is worth the effort as pretraining leads to improved performance on downstream tasks. We explore alternatives to full-scale task-specific pretraining of language models through the use of adapter modules, a parameter-efficient approach to transfer learning. We find that adapter-based pretraining is able to achieve comparable results to task-specific pretraining while using a fraction of the overall trainable parameters. We further explore direct use of adapters without pretraining and find that the direct fine-tuning performs mostly on par with pretrained adapter models, contradicting previously proposed benefits of continual pretraining in full pretraining fine-tuning strategies. Lastly, we perform an ablation study on task-adaptive pretraining to investigate how different hyperparameter settings can change the effectiveness of the pretraining.

SCRIPT: Self-Critic PreTraining of Transformers

We introduce Self-CRItic Pretraining Transformers (SCRIPT) for representation learning of text. The popular masked language modeling (MLM) pretraining methods like BERT replace some tokens with [MASK] and an encoder is trained to recover them, while ELECTRA trains a discriminator to detect replaced tokens proposed by a generator. In contrast, we train a language model as in MLM and further derive a discriminator or critic on top of the encoder without using any additional parameters. That is, the model itself is a critic. SCRIPT combines MLM training and discriminative training for learning rich representations and compute- and sample-efficiency. We demonstrate improved sample-efficiency in pretraining and enhanced representations evidenced by improved downstream task performance on GLUE and SQuAD over strong baselines. Also, the self-critic scores can be directly used as pseudo-log-likelihood for efficient scoring.