🚀 Exciting news! Two of our papers have been accepted to Findings of ACL @naaclmeeting! 🎉 If you're into speech processing, diarization, or linguistic analysis with LLMs, check these out. A quick thread 🧵👇.

🎉 Thanks to our collaborators at MELD Lab led by Sarah Moeller @UFLinguistics (Rahul Porwal, Alice Rozet, Pryce Houck, Jotsna Gowda) and the @SLaMLab_HHU led by @tang_kevin @HHU_de (Lian Remme)!. If you're at #NAACL2024, let's chat! 🚀 #NLP #SpeechProcessing #AAE #Diarization.

Join millions who have switched to Grok.

🔹 AAE’s grammatical complexity challenges NLP models, highlighting the need for better training data & architectural adjustments. 📖 Full paper:

📌 Key insights:.🔹 We test models on Habitual Be & Multiple Negation, two core AAE features.🔹 LLMs outperform traditional baselines but still struggle with bias (e.g., recency effects, formality mismatches).

2️⃣ LLMs as Grammatical Feature Taggers for African American English (AAE) 🏷️🗣️. Can LLM models accurately analyze African American English (AAE)? We evaluate rule-based models, transformers, and LLMs on key AAE grammatical features.

🔹Two diarization models both show higher confusion rates on TTRPGs.🔹wespeaker significantly underestimates the number of speakers, making it clear that TTRPGs push diarization to its limits. 📖Full paper:

📌 Key insights:.🔹TTRPGs create a unique diarization challenge due to frequent speaker turns & voice conversions.🔹Traditional diarization models struggle to distinguish real speakers from their fictional characters.🔹We introduce a new dataset & compare it to AMI/ICSI corpora.

1️⃣ Playing with Voices: Tabletop Role-Playing Game Recordings as a Diarization Challenge 🎭🎙️. Tabletop RPGs (TTRPGs) are full of dynamic conversations where players frequently alter their voices to role-play characters. But what does this mean for speaker diarization?.

9/End This is joint work by:.Alexis Davis, Joshua Martin, @EricCooks2, @vilaromel, Danyell Wilson-Howard, @tang_kevin, @DrJaniceKrieger .Affiliations: @MayoClinic, @HHU_de, @UFLinguistics,.@UFHealthCancer, @UF_IFAS, @bethunecookman. Please repost!.

8/ Intrigued? 📑 Our full article breaks down: . ✅ The iterative design process .✅ Lessons learned .✅ Opportunities for process improvement . We hope this inspires more inclusive approaches to health communication and bridge the gap between "English" and "Englishes.".

7/ 🧩 This isn't just about AAE—it's a call to action for inclusive health communication. No "one size fits all" for English speakers. Let’s adapt our strategies to reflect the richness of linguistic diversity. 🌟 #HealthCommunication.

6/ 💬 Message designers: Let’s rethink how we approach language in interventions. Beyond translations or "simplification," linguistic responsiveness means understanding the cultural and social nuances of how people speak and connect. 🌎.#LinguisticResponsiveness.

5/ 🚧 Challenges? Oh, there were plenty! . Social stereotypes about AAE were tough to navigate. They slowed us down, but also deepened our understanding. Our takeaway: A diverse team was essential to address these implicit barriers. 🤝.

4/ 💡 Key insight: The process matters as much as the product. We collaborated with: .👉 Communication scientists .👉 Linguists .👉 Community advisory boards .👉 Professional voice talents . Each 🗨️voice helped us navigate challenges (and biases!). 🔍.

3/ 🗣️ Our focus: Speakers of African American English (AAE). We developed a culturally & linguistically grounded cancer prevention intervention, tackling not just the what of health messaging but how it's said. #CancerPrevention.

2/ 🌟 Why does linguistic accommodation matter in health interventions? 🌟 . Language shapes how we connect, understand, and respond. But what happens when health messages are designed for "standard English" but audiences speak diverse "Englishes"? 🤔.#PublicHealth.

1/🚨Exciting news!🚨 Our new article is out in the Journal of Participatory Medicine @jmirpub! 🎉 "From English to 'Englishes': A Process Perspective on Enhancing the Linguistic Responsiveness of Culturally Tailored Cancer Prevention Interventions"✨👇See🧵

This is 🤝 #InterdisciplinaryResearch between linguists, neurologists, and AI researchers.Read the paper: .💬 #MedTwitter #Parkinsons #SpeechResearch.Affiliations of the authors: @HHU_de @UFLinguistics @UFPHHP @UFCISE. Please retweet.

5️⃣ Why does this matter? Lenition patterns may help distinguish PD from APD early on. This could offer a non-invasive way to support diagnosis, though further research is needed to validate its accuracy in broader contexts. 🤔 #Neurology.

4️⃣Key Findings:.🟢 PD patients: More stable articulation, preserving voiced/voiceless contrasts. 🔴 APD patients: Greater lenition, especially in voiceless stops, with more articulatory variability, reflecting broader motor deficits. #ParkinsonsResearch.