This paper explores cross-lingual transfer learning in natural language understanding (NLU), with the focus on bootstrapping Arabic from high-resource English and French languages for domain classification, intent classification, and named entity recognition tasks. We adopt a BERT-based architecture and pretrain three models using open-source Wikipedia data and large-scale commercial datasets: monolingual:Arabic, bilingual:Arabic-English, and trilingual:Arabic-English-French models. Additionally, we use off-the-shelf machine translator to translate internal data from source English language to the target Arabic language, in an effort to enhance transfer learning through translation. We conduct experiments that finetune the three models for NLU tasks and evaluate them on a large internal dataset. Despite the morphological, orthographical, and grammatical differences between Arabic and the source languages, transfer learning performance gains from source languages and through machine translation are achieved on a real-world Arabic test dataset in both a zero-shot setting and in a setting when the models are further finetuned on labeled data from the target language.

In the area of data augmentation research, the main focus to date has been on the improvement of the generation models, while the examination and improvements to synthetic data evaluation methods remains less explored. In our work, we explore a number of sentence similarity measures in the context of data generation filtering, and evaluate their impact on the performance of the targeted Natural Language Understanding problem on the example of the intent classification and named entity recognition tasks. Our experiments on ATIS dataset show that the right choice of filtering technique can bring up to 33% in sentence accuracy improvement for targeted underrepresented intents.

Current voice assistants typically use the best hypothesis yielded by their Automatic Speech Recognition (ASR) module as input to their Natural Language Understanding (NLU) module, thereby losing helpful information that might be stored in lower-ranked ASR hypotheses. We explore the change in performance of NLU associated tasks when utilizing five-best ASR hypotheses when compared to status quo for two language datasets, German and Portuguese. To harvest information from the ASR five-best, we leverage extractive summarization and joint extractive-abstractive summarization models for Domain Classification (DC) experiments while using a sequence-to-sequence model with a pointer generator network for Intent Classification (IC) and Named Entity Recognition (NER) multi-task experiments. For the DC full test set, we observe significant improvements of up to 7.2% and 15.5% in micro-averaged F1 scores, for German and Portuguese, respectively. In cases where the best ASR hypothesis was not an exact match to the transcribed utterance (mismatched test set), we see improvements of up to 6.7% and 8.8% micro-averaged F1 scores, for German and Portuguese, respectively. For IC and NER multi-task experiments, when evaluating on the mismatched test set, we see improvements across all domains in German and in 17 out of 19 domains in Portuguese (improvements based on change in SeMER scores). Our results suggest that the use of multiple ASR hypotheses, as opposed to one, can lead to significant performance improvements in the DC task for these non-English datasets. In addition, it could lead to significant improvement in the performance of IC and NER tasks in cases where the ASR model makes mistakes.