Milad Alshomary retweeted
New paper accepted to ๐๐๐๐ ๐ฎ๐ฌ๐ฎ๐ฒ: โ๐ถ๐๐๐ฅ๐ง: ๐๐ป๐๐ฒ๐ฟ๐ฝ๐ฟ๐ฒ๐๐ฎ๐ฏ๐น๐ฒ ๐๐บ๐ฏ๐ฒ๐ฑ๐ฑ๐ถ๐ป๐ด๐ ๐๐ถ๐ฎ ๐ฆ๐ฒ๐ป๐๐ฒ ๐๐ฒ๐ฐ๐ผ๐บ๐ฝ๐ผ๐๐ถ๐๐ถ๐ผ๐ปโ - a joint work with researchers at Columbia NLP - Milad Alshomary and Professor Kathleen McKeown.
๐๐ฒ๐บ๐ผ: iBERT.io
๐ฃ๐ฎ๐ฝ๐ฒ๐ฟ: arxiv.org/pdf/2510.09882
๐ช๐ต๐ฎ๐ ๐ถ๐ ๐ถ๐๐๐ฅ๐ง? iBERT is a general-purpose encoder that learns sparse, modular embeddings โ making meaning in language transparent and editable. Instead of treating representations as dense, opaque vectors, iBERT implicitly decomposes them into disentangled components that reveal structure in how meaning is encoded โ enabling precise interventions in embedding space. Whether for control, attribution, analysis, or editing, iBERT offers representations that are inherently interpretable and controllable. This built-in decomposition enables targeted edits, transparent probing, and fine-grained control โ all before any decoding or downstream use.
- ๐๐ป๐๐ฒ๐ฟ๐ฝ๐ฟ๐ฒ๐๐ฎ๐ฏ๐น๐ฒ ๐๐๐ฟ๐๐ฐ๐๐๐ฟ๐ฒ
- ๐๐ผ๐ป๐๐ฟ๐ผ๐น๐น๐ฎ๐ฏ๐น๐ฒ ๐ฟ๐ฒ๐ฝ๐ฟ๐ฒ๐๐ฒ๐ป๐๐ฎ๐๐ถ๐ผ๐ป๐
- ๐๐ผ๐บ๐ฝ๐ฒ๐๐ถ๐๐ถ๐๐ฒ ๐ฝ๐ฒ๐ฟ๐ณ๐ผ๐ฟ๐บ๐ฎ๐ป๐ฐ๐ฒ โ all in the same general-purpose encoder
#EACL2026 #iBERT #NLP #NLProc #InterpretableAI #RepresentationLearning #FoundationModels
Excited to share that our paper, "Generalizable Analysis of Authorial Style by Leveraging All Transformer Layers," is accepted at #EMNLP2025
Co-authors: Nikhil Varimalla, Vishal Anand, Smaranda Muresan, Kathleen McKeown
#AuthorshipAnalysis #interpretability
3
110
I spent some good time at @coling2025
Thanks to the organizers who did a very good job puttig it all together.
1
12
790
Our paper "Latent Space Interpretation for Stylistic Analysis and Explainable Authorship Attribution" has been accepted at @coling2025.
Co-authors: @narutatsuri , @MApidianaki, Ajay Patel, Smaranda Muresan, and Kathleen McKeown
Looking forward to discussing it at the conference
9
265
Milad Alshomary retweeted
Iโm hiring PhD students in computer science at Columbia!
Our lab will tackle core challenges in understanding and controlling neural models that interact with language.
for example,
- methods for LLM control
- discoveries of LLM properties
- pretraining for understanding
Milad Alshomary retweeted
Replying to @emnlpmeeting @mohitban47 @YunNungChen @zoeyliu_18 @jessyjli @feiliu_nlp @snigdhac25 @LucianaBenotti @malihealikhani @vinodkpg
Thanks to Demo Chairs @dirazuhf, @ManlingLi_, and @Hoper_Tom
Publication Chairs @miladalsh, @gozde_gul_sahin and Danilo Croce
and Handbook Chair Marco Polignano
A bit late to the party, but here is an accepted paper at #LRECCOLING2024: "Modeling the Quality of Dialogical Explanations" (arxiv.org/abs/2403.00662)
We created a corpus of Explanation dialogues from the ELI5 Subreddit and modeled the quality of these dialogues.
#NLProc
We studied the use of metaphors in news editorials and their effects on different ideologies. We believe that the corpus we released is of interest to the community
Our paper 'Analyzing the use of metaphors in news editorials for political framing' got accepted at the @naacl mains! Congratulations to all the co-authors @roxanneelbaff @miladalsh and Henning Wachsmuth. @AIHannover @trr_318 @webis_de #NLP #NLProc. See you in Mexico!
1
5
402
I successfully defended my thesis and graduated from Paderborn University.
Thanks a lot for the beautiful colleagues who supported me along this path.
The next step is Columbia University where I will have a postdoctoral position to work with Smaranda Muresan and Kathy McKeown.
Exciting shared tasks are waiting for you in Touchรฉ 2024.
If you are working on argumentation and like to solve challenging tasks, this definitely might interest you!
Working in Argumentation? Time to participate in Touchรฉ 2024!
Three shared tasks:
- Human Value Detection
- Ideology and Power Identification in Parliamentary Debates
- Image Retrieval/Generation for Arguments
Submission deadline is May 6th!
More info: touche.webis.de/
Milad Alshomary retweeted
Super exciting poster session of our paper with great conversations with the lovely people at #EMNLP2023. #NLP #NLProc @trr_318 @webis_de
Glad that I attended the #EMNLP23 conference for one last time as a member of the @AIHannover, with the amazing @stahlm, @MeghdutSengupta, and Henning Wachsmuth!
Closing of the #argmining23 workshop! I had the honor to co-organize this year's version with Chung-chi Chen, Smaranda Muresan, Joonsuk Park, and Julia Roomberg!
Milad Alshomary retweeted
The program is online [1]. We will host a panel session, a keynote talk, two shared tasks, and a collection of interesting papers.
Looking forward to seeing you at #EMNLP
#NLProc
argmining-org.github.io/2023โฆ
Milad Alshomary retweeted
Due to multiple requests, we are glad to announce that the submission deadline will be extended till the 7th of September.
Looking forward to receiving your awesome submissions: argmining-org.github.io/2023โฆ
Milad Alshomary retweeted
We are looking for a postdoc for our automated fact-checking project AVeriTeC! Join us (@MichaelSejr Zhenyun Deng, @MoyYuan @pietro_lesci and the rest of the gang @cambridgenlp) and help check more facts with evidence and run @FEVERworkshop jobs.cam.ac.uk/job/42366/ deadline 15/9
Indeed, we learned a lot in this talk about e-deliberation. Thanks, @GabriellaLapesa for the excellent talk.
1
4
399
Check out the new corpus and taxonomy for appropriateness in argumentation.
What makes an argument (in)appropriate?
Checkout our new #ACL2023NLP paper 'Modeling Appropriate Language in Argumentation'.
w/@ShahbazSyd, Felix Lange, @martinpotthast
and Henning Wachsmuth
Preprint: arxiv.org/pdf/2305.14935
Corpus & Code: github.com/timonziegenbein/aโฆ
Milad Alshomary retweeted
(1/3) Our workshop will host two exciting shared tasks, co-occurring with EMNLP-23: PragTag-2023 and ImageArg-23.
Make sure to participate!
#NLProc
@emnlpmeeting





























