Leveraging representations from pre-trained transformer-based encoders achieves state-of-the-art performance on numerous NLP tasks. Larger encoders can improve accuracy for spoken language understanding (SLU) but are challenging to use given the inference latency constraints of online systems (especially on CPU machines).We evaluate using a larger 170M parameter BERT encoder that shares representations across languages, domains and tasks for SLU compared to using smaller 17M parameter BERT encoders with language-, domain- and task-decoupled finetuning.Running inference with a larger shared encoder on GPU is latency neutral and reduces infrastructure cost compared to running inference for decoupled smaller encoders on CPU machines. The larger shared encoder reduces semantic error rates by 4.62% for test sets representing user requests to voice-controlled devices and 5.79% on the tail of the test sets on average across four languages.

Teacher-student knowledge distillation is a popular technique for compressing today’s prevailing large language models into manageable sizes that fit low-latency downstream applications. Both the teacher and the choice of transfer set used for distillation are crucial ingredients in creating a high quality student. Yet, the generic corpora used to pretrain the teacher and the corpora associated with the downstream target domain are often significantly different, which raises a natural question: should the student be distilled over the generic corpora, so as to learn from high-quality teacher predictions, or over the downstream task corpora to align with finetuning? Our study investigates this trade-off using Domain Classification (DC) and Intent Classification/Named Entity Recognition (ICNER) as downstream tasks. We distill several multilingual students from a larger multilingual LM with varying proportions of generic and task-specific datasets, and report their performance after finetuning on DC and ICNER. We observe significant improvements across tasks and test sets when only task-specific corpora is used. We also report on how the impact of adding task-specific data to the transfer set correlates with the similarity between generic and task-specific data. Our results clearly indicate that, while distillation from a generic LM benefits downstream tasks, students learn better using target domain data even if it comes at the price of noisier teacher predictions. In other words, target domain data still trumps teacher knowledge.

Scaling conversational personal assistants to a multitude of languages puts high demands on collecting and labelling data, a setting in which cross-lingual learning techniques can help to reconcile the need for well-performing Natural Language Understanding (NLU) with a desideratum to support many languages without incurring unacceptable cost. In this work, we show that automatically annotating unlabeled utterances using Machine Translation in an offline fashion and adding them to the training data can improve performance for existing NLU features for low-resource languages, where a straightforward translate-test approach as considered in existing literature would fail the latency requirements of a live environment. We demonstrate the effectiveness of our method with intrinsic and extrinsic evaluation using a real-world commercial dialog system in German. Beyond an intrinsic evaluation, where 56% of the resulting automatically labeled utterances had a perfect match with ground-truth labels, we see significant performance improvements in an extrinsic evaluation settings when manual labeled data is available in small quantities.