
PhD candidate in NLP @UniUtrecht | Measuring language variation with ML/NLP | now mainly on 🦋 via https://nitter.cf/t.co/gpk3bBPSrd
Joined December 2019
- Tweets83
- Following237
- Followers219
- Likes254
Pinned Tweet
Interested in whether people👂 each other in a conversation?
🚨New paper accepted at #EMNLP2024 with @tyskevdb and @dongng about detecting paraphrases between speakers
🤖 Detect? huggingface.co/AnnaWegmann/H…
📊 Analyze? huggingface.co/datasets/Anna…
📄 Read? arxiv.org/pdf/2404.06670
Anna Wegmann retweeted
Podcasts are a popular medium, but data for computational research is limited! We introduce the Structured Podcast Research Corpus (SPoRC - huggingface.co/datasets/blit…), a large, multimodal dataset of English podcasts 🧵
arxiv.org/abs/2411.07892
Come talk to me and @dongng on Wednesday, Poster Session E from 4.00-5.30PM about paraphrases in dialog.
See you at #EMNLP2024!
Interested in whether people👂 each other in a conversation?
🚨New paper accepted at #EMNLP2024 with @tyskevdb and @dongng about detecting paraphrases between speakers
🤖 Detect? huggingface.co/AnnaWegmann/H…
📊 Analyze? huggingface.co/datasets/Anna…
📄 Read? arxiv.org/pdf/2404.06670
2
9
727
Anna Wegmann retweeted
#Tokenization is undeniably a key player in the success story of #LLMs but we poorly understand why.
I want to highlight progress we made in understanding the role of tokenization, developing the core incidents and mitigating its problems. 🧵👇
Anna Wegmann retweeted
Curious about using LLMs to simulate conversations? Check out this big collaborative project we did @umsi !
#NLProc
👩🏼💻 Real or Robotic? 🤖 Can LLMs accurately simulate qualities of human responses in dialogue?
Human conversations with LLMs are great for assessing the capabilities of LLMs. But having lots of folks chat with LLMs is challenging (💰⏳🕵️). Could we have another LLM *simulate* being a human talking to an LLM as a substitute?
In our new preprint, we test whether models can roleplay as the human in human-LLM conversations. Using the WildChat dataset and 100K+ simulations we test how well these LLM responses actually mimic with human ones. Our study spans 🇬🇧 English, 🇨🇳 Chinese, and 🇷🇺 Russian, using 21 linguistic metrics like lexical, semantic, syntactic, and stylistic features.
Anna Wegmann retweeted
The 34th edition of Computational Linguistics in The Netherlands (the Dutch-Belgian #NLProc conference) will be held @UniLeiden on August 30
The list of accepted abstracts is on the website and registration is open for everyone interested 💬
#clin34
clin34.leidenuniv.nl/
Anna Wegmann retweeted
🔎What values and opinions do we see when we use 6 LLMs to generate 156,000 responses to 62 political propositions?
Our paper "Revealing Fine-Grained Values and Opinions in Large Language Models" answers this.
📰 arxiv.org/abs/2406.19238v1
#NLProc #LLMs
Anna Wegmann retweeted
📢Is current “human-AI alignment” research clarified and comprehensive? 🤔 We systematically reviewed 400+ papers across HCI, NLP, and ML to develop a framework for 👫<>🤖"Bidirectional Human-AI Alignment", encompassing the dual paths of “Aligning AI to Human” and “Aligning Human to AI”. We also clarified core questions 🎯 of 'what is the alignment goal?', 'with whom to align?', and 'what are the human values?’
Further, we share 👩💻 our findings on values and interaction techniques for alignment. Check out the three challenges and potential solutions we envision for future research🌟! #HumanAIAlignment
💎arxiv.org/abs/2406.09264. 🧵1/
Huge thanks to our amazing team 💗 @tknearem, @reshmigh, Kenan Alkiek, @kundan_official, @YachuanLiu, @ziqiao_ma, @savvas_petridis, @yolohao, Li Qiwei, Sushrita Rakshit, @ChengleiSi, @yutxie, and our fabulous advisors @jeffbigham, @bentley79, Joyce Chai, @zacharylipton, @meiqzh, @radamihalcea, Michael Terry, @Diyi_Yang, @merrierm, @presnick, @david__jurgens! 🙏 Many thanks for all your great effort🤗!
Anna Wegmann retweeted
[1/13] LLMs are increasingly skilled at mimicking human agents in social settings, but have they truly developed a consistent personality? Check out our work accepted to #NAACL2024 where we question the reliability of persona tests applied to LLMs.
Arxiv: arxiv.org/abs/2311.09718
Anna Wegmann retweeted
📰New preprint! w/ @christian_igel @raghavian📰
BMRS: Bayesian Model Reduction for Structured Pruning
Structured pruning makes neural nets efficient by removing full structures (e.g. neurons).
But how do we know what to prune?
Here's our approach: arxiv.org/pdf/2406.01345
Anna Wegmann retweeted
📢 JOBS📢
Come work with us @MilaNLProc!
Looking for 2 POSTDOCS (two-year positions w/extension) to work on personalized and subjective approaches to #NLProc.
Deadline: May 30 2024
Start date: from Sep 2024
Link: jobmarket.unibocconi.eu/?id=…
Anna Wegmann retweeted
Next up in the DiLCo Lecture Series 2024: Christoph Purschke @questoph presenting his multi-method approach to "Monitoring the public debate on multilingualism in Luxembourg". Thursday, April 25 at 4 pm CEST, open access. Registration: dilco.uni-hamburg.de/.../reg… @unihh
Anna Wegmann retweeted
Thread on our new paper!
I'm excited to share that the journal version of our paper, "An archival perspective on pretraining data", is now available (open access) from Patterns!
This project was led by @MeeraDesai18, along with @IrenePasquetto, @az_jacobs, and myself
1/n
Anna Wegmann retweeted
Replying to @natfriedman
Some colleagues and I have been studying the impact of ChatGPT on SO using data on posts, not views: arxiv.org/abs/2307.07367
Besides a big decrease after ChatGPT, we observe a completely flat 2022, and earlier a big bump in activity during early Covid.
Anna Wegmann retweeted
Copenhagen is beautiful & #ic2s2 is amazing, but do you know what’s neither? 🚫 unintended bias towards marginalized people in hate speech detection systems.
Presenting our poster (w/ @hide_yourself @clauwa @IAugenstein) today about how data augmentation can lead to such biases!
Anna Wegmann retweeted
🚨 New working paper!
Are Large Language Models a threat to digital public goods?
@RMaria_drc N. Laurentsyeva and I find a 16% decrease in activity on @StackOverflow since release of #ChatGPT. Decrease is language dependent & reaches 25% by June: arxiv.org/abs/2307.07367
Thread⬇️
Anna Wegmann retweeted
How does annotator identity influence their judgments for NLP tasks? Collaborating with @Prolific, @david__jurgens and I created POPQUORN: a dataset with 45000 annotations on 4 NLP tasks by 1484 annotators with rich demographic information.
Paper: arxiv.org/abs/2306.06826
🧵 1/11
Anna Wegmann retweeted
It takes 360.000 gallons of water/day to cool a data centre! Exploitation of workers, workplace automation, & mass discrimination of marginalised groups, these are REAL existential risks, not this latest PR stunt, my interview independent.co.uk/tech/rishi… @Independent @oiioxford @BKCHarvard





























