These days' multi-intent utterances have become very important for the spoken language understanding (SLU). The multi-intent systems and algorithms add more complexity (Compare to the singleintent-based system) to the SLU. As, it requires an accurate system, which can identify intents and slots at fine-grain (i.e., word/token) level and also able to handle the relation between intents and slots locally at utterance level. In this case, intents may belong to multiple domains and multiple different classes. Similarly, slots may also belong to multiple different classes, and slots of the same class may be related to multiple different intent classes. Unfortunately, very few works have been done till now to address these issues at the fine-grain level. To solve this problem, we propose a smart stacking-ensemble strategy. The first stage of this system uses a combination of three different types of powerful multitasking NLP models, developed on top of pre-trained BERT, XLNet, and Elmo. Finally, a stacking ensemble layer learns to predict the best possible results. We have evaluated our model on four publicly available datasets. The evaluation results on the stateof-the-art public datasets show that our devised system outperforms the existing multi-intent-based systems at token-level and sentence-level.

In Spoken Language Understanding (SLU), the ability to detect out-of-domain (OOD) input dialog plays a very important role (e.g., voice assistance and chatbot systems). However, most of the existing OOD detection methods rely heavily on manually labeled OOD. Manual labeling of the OOD data for a dynamically changing and evolving area is time-consuming and not immediately possible. It limits the feasibility of these models in practical applications. So, to solve this problem, we are considering the scenario of having no OOD labeled data (i.e., zero-shot learning). To achieve this goal, we have used the intent focused semantic parsing, extracted with the help of Transformer-based techniques (e.g., BERT [26]). The two main components of the intent-focused semantic parsing are -(a) the sentence-level intents and (b) token-level intent classes, which show the relation of slot tokens with intent classes. Finally, we combine both information and use a One Class Neural Network (OC-NN) based zero-shot classifier. Our devised system has shown better results compared to the state-of-the-art on four publicly available datasets.INDEX TERMS Spoken Language Understanding, Out-of-domain (OOD) detection, zero-shot learning.