Meteo Trentino and FBK started a collaboration in 2009 with the goal of developing a translation system for weather forecast bulletins. On a daily basis, Meteo Trentino compiles and makes available through the WEB site a weather bulletin related to the Trentino area written in Italian. An equivalent or shortened version of that bulletin is also published in English and German after a manual translation made by forecasters. FBK has been asked to provide forecasters with a system for the automatic translation of Italian bulletins into those two languages. In the course of 2009, FBK has developed the translation system and a simple interface (shown in the figure below) which have been delivered to Meteo Trentino, whose forecasters are currently using. The project has been extended to 2010 with the goal of improving the quality of the automatic translation by exploiting feedback from real users.
Our #PickOfTheWeek 📚: “Training Dynamics Impact Post-Training Quantization Robustness” by @actatjer, @n_ajroldi, @jonasgeiping
Pick of the week @fbk_mt
Cool paper on how robustness to quantization in LLMs is also impacted by training conditions, instead of just the number of training tokens!
https://arxiv.org/abs/2510.06213
Our #PickOfTheWeek 📚: “High-Fidelity Simultaneous Speech-To-Speech Translation” by @t0m1ab, @lmazare, @EXGRV, @ptrkprz, @honualx and @neilzegh
🎙️➡️🌍 This work shows how a full-duplex SpeechLLM architecture can be adapted for simultaneous S2ST and S2TT.
Can LLM Agents Learn how to behave According to your preferences?
PersonaAgent is a framework that combines personalized memory and tool use to help LLM agents adapt their actions to individual users without fine-tuning the underlying model.
Our #PickOfTheWeek, selected by @mgaido91: "Alignment Is All You Need: Instruction-Free Training for General Audio-Language Models" by Xuanru Zhou, Yiwen Shao, Jiahong Li, and Dong Yu 🎧
Can instruction-free training power the next generation of audio-language models?
Cool to see we are not the only ones working on building SpeechLLMs with instruction-following capabilities without requiring LLM adaptation and instruction-following data that is expensive to collect!
https://arxiv.org/abs/2608.18132
@fbk_mt