Vista.ue at SemEval-2019 Task 5: Single Multilingual Hate Speech Detection Model

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 524–528 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 524 Vista.ue at SemEval-2019 Task 5: Single Multilingual Hate Speech Detection Model Kashyap Raiyani, Teresa Gonc¸alves, Paulo Quaresma and Vitor Beires Nogueira Computer Science Department, University of ´Evora, Portugal (kshyp,tcg,pq,vbn)@uevora.pt Abstract This paper shares insight from participating in SemEval-2019 Task 5. The main propose of this system-description paper is to facili- tate the reader with replicability and to pro- vide insightful analysis of the developed sys- tem. Here in Vista.ue, we proposed a sin- gle multilingual hate speech detection model. This model was ranked 46/70 for English Task A and 31/43 for English Task B. Vista.ue was able to rank 38/41 for Spanish Task A and 22/25 for Spanish Task B. 1 Introduction According to the article (Bosco et al., 2017), nearly a quarter of a billion people, throughout the world, currently live in a country other than their place of birth. This is an increase of 41% from 2000 to 2015. This figure includes more than 21 million refugees often vulnerable and dissat- isfied. Since 2015, Europe is facing an unprece- dented refugee crisis, the by-effect of the Syrian civil war and the terrible living conditions in equa- torial Africa. 1,300,000 people have generated this increased migration flow to Europe which can only but increase, putting European stable soci- eties, so far, under pressure. Therefore, the implications for the European so- ciety and the way we behave towards immigra- tion, immigrant integration and social inclusion for newcomers and their children, are becoming more decisive and must be addressed either at a local or global level, considering a political and social perspective. While this phenomenon stim- ulates the generation and diffusion of hate speech and hate crimes, at the same time several initia- tives are promoted, but they should be further im- proved to increase the awareness and empathy of receiving populations while avoiding polarization against immigrants. Hate speech analysis and hate maps allow both a greater understanding of social phenomena linked to the integration of migrants, that more targeted actions to improve it. The integration of mi- grants is strongly linked to the new cultural con- text where they try to rebuild their lives. The pro- cess of acculturation depends on personal and so- cial variables of the migrant, in large part in turn dependent on the cultural context of his/her ori- gin, on the characteristics of the context of reset- tlement and on events occurring during this life period. The different migrants strategies firstly af- fect the different outcomes achieved. In particular, he can decide whether or not to maintain the cul- tural identity of origin and whether or not to es- tablish and maintain new relationships within the new contest. This gives rise to four possible dif- ferent outcomes: integration, assimilation, separa- tion/segregation, marginalization (Berry, 1997). 1.1 Motivation Data released by European Community about pop- ulation change (Union, 2015) show that from the 1990s …

Content truncated to 500 words