We study the power of cross-attention in the Transformer architecture within the context of transfer learning for machine translation, and extend the findings of studies into cross-attention when training from scratch. We conduct a series of experime
nts through fine-tuning a translation model on data where either the source or target language has changed. These experiments reveal that fine-tuning only the cross-attention parameters is nearly as effective as fine-tuning all parameters (i.e., the entire translation model). We provide insights into why this is the case and observe that limiting fine-tuning in this manner yields cross-lingually aligned embeddings. The implications of this finding for researchers and practitioners include a mitigation of catastrophic forgetting, the potential for zero-shot translation, and the ability to extend machine translation models to several new language pairs with reduced parameter storage overhead.
During the fine-tuning phase of transfer learning, the pretrained vocabulary remains unchanged, while model parameters are updated. The vocabulary generated based on the pretrained data is suboptimal for downstream data when domain discrepancy exists
. We propose to consider the vocabulary as an optimizable parameter, allowing us to update the vocabulary by expanding it with domain specific vocabulary based on a tokenization statistic. Furthermore, we preserve the embeddings of the added words from overfitting to downstream data by utilizing knowledge learned from a pretrained language model with a regularization term. Our method achieved consistent performance improvements on diverse domains (i.e., biomedical, computer science, news, and reviews).
In this paper, we describe our MiSS system that participated in the WMT21 news translation task. We mainly participated in the evaluation of the three translation directions of English-Chinese and Japanese-English translation tasks. In the systems su
bmitted, we primarily considered wider networks, deeper networks, relative positional encoding, and dynamic convolutional networks in terms of model structure, while in terms of training, we investigated contrastive learning-reinforced domain adaptation, self-supervised training, and optimization objective switching training methods. According to the final evaluation results, a deeper, wider, and stronger network can improve translation performance in general, yet our data domain adaption method can improve performance even more. In addition, we found that switching to the use of our proposed objective during the finetune phase using relatively small domain-related data can effectively improve the stability of the model's convergence and achieve better optimal performance.
We propose a novel framework to train models to classify acceptability of responses generated by natural language generation (NLG) models, improving upon existing sentence transformation and model-based approaches. An NLG response is considered accep
table if it is both semantically correct and grammatical. We don't make use of any human references making the classifiers suitable for runtime deployment. Training data for the classifiers is obtained using a 2-stage approach of first generating synthetic data using a combination of existing and new model-based approaches followed by a novel validation framework to filter and sort the synthetic data into acceptable and unacceptable classes. Our 2-stage approach adapts to a wide range of data representations and does not require additional data beyond what the NLG models are trained on. It is also independent of the underlying NLG model architecture, and is able to generate more realistic samples close to the distribution of the NLG model-generated responses. We present results on 5 datasets (WebNLG, Cleaned E2E, ViGGO, Alarm, and Weather) with varying data representations. We compare our framework with existing techniques that involve synthetic data generation using simple sentence transformations and/or model-based techniques, and show that building acceptability classifiers using data that resembles the generation model outputs followed by a validation framework outperforms the existing techniques, achieving state-of-the-art results. We also show that our techniques can be used in few-shot settings using self-training.
We develop a novel approach for confidently accelerating inference in the large and expensive multilayer Transformers that are now ubiquitous in natural language processing (NLP). Amortized or approximate computational methods increase efficiency, bu
t can come with unpredictable performance costs. In this work, we present CATs -- Confident Adaptive Transformers -- in which we simultaneously increase computational efficiency, while guaranteeing a specifiable degree of consistency with the original model with high confidence. Our method trains additional prediction heads on top of intermediate layers, and dynamically decides when to stop allocating computational effort to each input using a meta consistency classifier. To calibrate our early prediction stopping rule, we formulate a unique extension of conformal prediction. We demonstrate the effectiveness of this approach on four classification and regression tasks.
We propose to control paraphrase generation through carefully chosen target syntactic structures to generate more proper and higher quality paraphrases. Our model, AESOP, leverages a pretrained language model and adds deliberately chosen syntactical
control via a retrieval-based selection module to generate fluent paraphrases. Experiments show that AESOP achieves state-of-the-art performances on semantic preservation and syntactic conformation on two benchmark datasets with ground-truth syntactic control from human-annotated exemplars. Moreover, with the retrieval-based target syntax selection module, AESOP generates paraphrases with even better qualities than the current best model using human-annotated target syntactic parses according to human evaluation. We further demonstrate the effectiveness of AESOP to improve classification models' robustness to syntactic perturbation by data augmentation on two GLUE tasks.
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.
Adapter layers are lightweight, learnable units inserted between transformer layers. Recent work explores using such layers for neural machine translation (NMT), to adapt pre-trained models to new domains or language pairs, training only a small set
of parameters for each new setting (language pair or domain). In this work we study the compositionality of language and domain adapters in the context of Machine Translation. We aim to study, 1) parameter-efficient adaptation to multiple domains and languages simultaneously (full-resource scenario) and 2) cross-lingual transfer in domains where parallel data is unavailable for certain language pairs (partial-resource scenario). We find that in the partial resource scenario a naive combination of domain-specific and language-specific adapters often results in catastrophic forgetting' of the missing languages. We study other ways to combine the adapters to alleviate this issue and maximize cross-lingual transfer. With our best adapter combinations, we obtain improvements of 3-4 BLEU on average for source languages that do not have in-domain data. For target languages without in-domain data, we achieve a similar improvement by combining adapters with back-translation. Supplementary material is available at https://tinyurl.com/r66stbxj.
Recent task-oriented dialogue systems learn a model from annotated dialogues, and such dialogues are in turn collected and annotated so that they are consistent with certain domain knowledge. However, in real scenarios, domain knowledge is subject to
frequent changes, and initial training dialogues may soon become obsolete, resulting in a significant decrease in the model performance. In this paper, we investigate the relationship between training dialogues and domain knowledge, and propose Dialogue Domain Adaptation, a methodology aiming at adapting initial training dialogues to changes intervened in the domain knowledge. We focus on slot-value changes (e.g., when new slot values are available to describe domain entities) and define an experimental setting for dialogue domain adaptation. First, we show that current state-of-the-art models for dialogue state tracking are still poorly robust to slot-value changes of the domain knowledge. Then, we compare different domain adaptation strategies, showing that simple techniques are effective to reduce the gap between training dialogues and domain knowledge.
Multimodal Neural Machine Translation (MNMT) is an interesting task in natural language processing (NLP) where we use visual modalities along with a source sentence to aid the source to target translation process. Recently, there has been a lot of wo
rks in MNMT frameworks to boost the performance of standalone Machine Translation tasks. Most of the prior works in MNMT tried to perform translation between two widely known languages (e.g. English-to-German, English-to-French ). In this paper, We explore the effectiveness of different state-of-the-art MNMT methods, which use various data oriented techniques including multimodal pre-training, for low resource languages. Although the existing methods works well on high resource languages, usability of those methods on low-resource languages is unknown. In this paper, we evaluate the existing methods on Hindi and report our findings.
