Datasets used to train deep learning models in industrial settings often exhibit skewed distributions with some samples repeated a large number of times. This paper presents a simple yet effective solution to reduce the increased burden of repeated computation on redundant datasets. Our approach eliminates duplicates at the batch level, without altering the data distribution observed by the model, making it model-agnostic and easy to implement as a plug-and-play module. We also provide a mathematical expression to estimate the reduction in training time that our approach provides. Through empirical evidence, we show that our approach significantly reduces training times on various models across datasets with varying redundancy factors, without impacting their performance on the Named Entity Recognition task, both on publicly available datasets and in real industrial settings. In the latter, the approach speeds training by up to 87%, and by 46% on average, with a drop in model performance of 0.2% relative at worst. We finally release a modular and reusable codebase to further advance research in this area.

Cross-lingual phenomena are quite common in informal contexts like social media, where users are likely to mix their native language with English or other languages. However, few studies have focused so far on analyzing cross-lingual interactions in voice-assistant data, which present peculiar features in terms of sentence length, named entities, and use of spoken language. Also, little attention has been posed to European countries, where English is frequently used as a second language. In this paper, we present a large-scale empirical analysis of cross-lingual phenomena (code-mixing, linguistic borrowing, foreign named entities) in the interactions with a large-scale voice assistant in European countries. To do this, we first introduce a general, highly-scalable technique to generate synthetic mixed training data annotated with token-level language labels and we train two neural network models to predict them. We evaluate the models both on the synthetic dataset and on a real dataset of code-switched utterances, showing that the best performance is obtained by a character convolution based model. The results of the analysis highlight different behaviors between countries, having Italy with the highest ratio of cross-lingual utterances and Spain with a marked preference in keeping Spanish words. Our research, paired to the increase of the cross-lingual phenomena in time, motivates further research in developing multilingual Natural Language Understanding (NLU) models, which can naturally deal with cross-lingual interactions.

The quality of Natural Language Understanding (NLU) models is typically evaluated using aggregated metrics on a large number of utterances. In a dialog system, though, the manual analysis of failures on specific utterances is a time-consuming and yet critical endeavor to guarantee a high-quality customer experience. A crucial question for this analysis is how to create a test set of utterances that covers a diversity of possible customer requests. In this paper, we introduce the task of generating a test set with high semantic diversity for NLU evaluation in dialog systems and we describe an approach to address it. The approach starts by extracting high-traffic utterance patterns. Then, for each pattern, it achieves high diversity selecting utterances from different regions of the utterance embedding space. We compare three selection strategies based on clustering of utterances in the embedding space, on solving the maximum distance optimization problem and on simple heuristics such as random uniform sampling and popularity. The evaluation shows that the highest semantic and lexicon diversity is obtained by a greedy maximum sum of distance solver in a comparable runtime with the clustering and the heuristics approaches.