#MachineTranslation Research Unit @FBK_research. #nlproc #deeplearning #ai
Trento, Trentino-South Tyrol
Joined October 2015
- Tweets3.3K
- Following452
- Followers1.4K
- Likes3K
🧩 The authors derive word-level “contextual alignments” from an MT model’s perplexity and enforce them by inserting silences 🤫 or resynthesizing targets with an alignment-aware TTS.
📝 The model also predicts aligned text tokens as scaffolding for speech, and 🗣️ conditional training plus classifier-free guidance improve voice transfer.
📄 Paper: arxiv.org/abs/2502.03382
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.
In this case, personalization goes beyond remembering user information: agents can use past interactions to adapt their reasoning and tool usage. This solution outperforms RAG-based personalization and general-purpose agents across four personalized decision-making tasks.
🔗 Paper link: PersonaAgent: Bridging Memory and Action for Personalized LLM Agents: aclanthology.org/2026.findin…
#PickOfTheWeek #Personalization #LLM #AIAgents #MTUnit #FBK
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!
arxiv.org/abs/2608.18132
@fbk_mt
This paper shows how alignment alone can streamline training and improve real-world evaluation beyond lab benchmarks.
#AudioAI #VoiceAI #MachineLearning #LLM #AIResearch
Our #PickOfTheWeek, selected by @BeatriceSavoldi:
"The AI Observatory: A Public Measure of Real-World AI Use" by @ShayneRedford, @AnkaReuel, @zoeykii et al.
A crucial framework tracking how AI models are deployed, adopted, and evaluated beyond lab benchmarks.
Pick of the week @fbk_mt
📚"The AI Observatory: A Public Measure of Real-World AI Use"
A study and taxonomy to explore real conversations across multiple datasets.
dataprovenance.org/ai_observ…
Can algorithmic gender prediction ever be valid?
Check out this week's top pick by @lina_conti: "Algorithmic Gender Prediction Is Illegitimate, But Gender Imputation Can Yield Valid Measurements" by @evandongyx & @ang3linawang.
Pick of the week by @evandongyx & @ang3linawang:
arxiv.org/pdf/2608.13444
Predicting gender from images or names can reveal discrimination. But the practice itself harms trans people. This paper works through when that tradeoff might be justified and how to do it responsibly.
Our pick of the week by
@dhairya_su47605
: "Task-Circuit Quantization: Leveraging Knowledge Localization and Interpretability for Compression" by @hanqi_xiao, @yilin_sung, @EliasEskin and @mohitban47
#Quantization #Interpretibility
#PickoftheWeek @fbk_mt
Super cool paper on leavaraging Interpretability for Compression!
arxiv.org/pdf/2504.07389
Our pick of the week:
"Large Language Diffusion Model" by Shen Nie, Fengqi Zhu, @ZebinYou, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, @LiChongxuan
It is very cool to see how the researcher combine diffusion model and transformer blocks to train
a large model that can handle reversal reasoning tasks better than open source LLMs and achieve competent scores on most NLP tasks
📄: arxiv.org/pdf/2502.09992
#DeepGenerativeModel #Diffusion
Our pick of the week by
@mgaido91
: "FlexiSLM: A Dynamic and Controllable Frame Rate Spoken Language Model" by Jiaqi Li, Chaoren Wang, Xiaohai Tian, Mingjie Chen, Xinyu Liang, Xu Li, Yufan Lin, Junwen Qiu, Jun Zhang, Lu Lu, Haizhou Li and @drwuz
#SLM #EfficientInference
Cool to see a work that adaptively chooses at inference how much to compress the input speech sequence, to control inference costs and quality based on the input, without enforcing a global trade-off to each segment: arxiv.org/pdf/2606.31247
@fbk_mt
Our pick of the week by
@FBKZhihangXie : "Speech-XL: Towards Long-Form Speech Understanding in Large Speech Language Models" by Haoqin Sun, @Chenyang_Lyu, Shiwan Zhao, Xuanfan Ni, Xiangyu Kong, @wangly0229, Weihua Luo and Yong Qin
#SpeechLLM #LongFormSpeech #SLU
🚀 New paper: Speech-XL for long-form SpeechLLMs
📄 arxiv.org/abs/2602.05373
🧩 Uses Speech Summarization Tokens to compress local speech intervals into compact KV states efficiently.
✨ Improves long-form speech understanding while reducing memory and FLOPs on 10-minute audio.
Last week, we had a great talk for our MT Seminar Series!
@julius_gulius a PhD from @cambridgenlp presented a talk on "Effective uses of grammatical knowledge in extremely low-resource Machine Translation"
#MachineTranslation #LowResourceMT #NLProc #FBK
Our pick of the week by
@BeatriceSavoldi
: "Accuracy: Community Perspectives on Machine Translation" by Yujun Wang,
@EhudReiter
, Shimei Pan,
@egere14
and Wei Zhao #MachineTranslation #TranslationQuality #Evaluation
📖 #PickoftheWeek @fbk_mt "Accuracy: Community Perspectives on Machine Translation"
A cool analysis of the conflicting interests of different communities around MT(AI developers, LSPs, and users)
arxiv.org/pdf/2606.09655
#NLP #MachineTranslation #DiverseStakeholders
We are at the Alliance for Language Technologies - #ALTEDIC Week 2026!
@luisabentivogli and @negri_teo are attending two project meetings (ALT-EDIC4EU and #LLMs4EU), presenting the Evaluation of Spoken Language Translation in the context of IWSLT.
#LanguageTechnologies #FBK
