The rapid advancement of natural language technologies presents both opportunities and challenges for ensuring that these tools genuinely benefit society. This talk explores how we can align multilingual and responsible language technologies with real-world social good applications, designed in close interdisciplinary collaboration. I will discuss approaches for mitigating online harms such as toxic and hate speech, while highlighting the importance of extending these solutions beyond popular languages. Central to this effort is the question of cross-lingual transfer—how far existing methods can generalize across languages and cultures, and how we might design approaches that make transfer more efficient under limited-resource conditions. Together, these perspectives highlight the potential of multilingual, cultural-aware, and socially responsible NLP to drive meaningful impact.