Andrei Avram, Verginica Barbu Mititelu, Dumitru-Clementin Cercel. Proceedings of the 19th Workshop on Multiword Expressions (MWE 2023). 2023.

Correctly identifying multiword expressions (MWEs) is an important task for most natural language processing systems since their misidentification can result in ambiguity and misunderstanding of the underlying text. In this work, we evaluate the performance of the mBERT model for MWE identification in a multilingual context by training it on all 14 languages available in version 1.2 of the PARSEME corpus. We also incorporate lateral inhibition and language adversarial training into our methodology to create language-independent embeddings and improve its capabilities in identifying multiword expressions. The evaluation of our models shows that the approach employed in this work achieves better results compared to the best system of the PARSEME 1.2 competition, MTLB-STRUCT, on 11 out of 14 languages for global MWE identification and on 12 out of 14 languages for unseen MWE identification. Additionally, averaged across all languages, our best approach outperforms the MTLB-STRUCT system by 1.23% on global MWE identification and by 4.73% on unseen global MWE identification.

Publicly available speech datasets for Romanian are still scarce, being far from enough for obtaining state-of-the-art performance with modern deep neural networks. As a response to this issue, during the development of an internal automatic speech recognition system for the national emergency call center, we have created the Romanian Speech Alert Corpus, a new Romanian speech corpus that was obtained by crowd-sourcing the reading of sentences in our institution. This paper describes the data acquisition process, several statistics about the resulted corpus and the algorithm we employed for removing the inadequate recordings.

Network Intrusion Detection Systems (NIDS) are a critical component in any security system and their role is to monitor traffic on a network, looking for suspicious activity (e.g. attacks, unauthorized activities). Neural networks have become an increasingly popular solution for implementing network intrusion detection. Their ability to identify complex patterns and behaviours have made them a suitable solution for differentiating between normal and malicious traffic. One of the major problems encountered in implementing such neural networks is that they are computational and memory expensive and they can cause massive overhead in small embedded devices. This paper proposes a deep autoencoded dense neural network for detecting network attacks, trained and evaluated on NSL-KDD dataset that can satisfy the memory and real time constraints of such systems. The hyperparameters of the model are tuned, and three variants with different sizes and accuracies are proposed in order to produce flexibility in choosing the correct architecture for the desired platform.