ExplAIn Lab

We will be at @iwcs2021 today, presenting our work! So if you are around, please come and have a chat.

We had one paper accepted at ACL2021! 🎉🎊🎉🎊 Congrats to the authors @Mokanarangan @mrc_valentino @andrenfreitas

This is one of the outcomes of the collaboration with Peak AI (@Peak_HQ) funded by the @innovateuk KTP programme.

Last week @alex_bogatu presented the paper Cost–effective Variational Active Entity Resolution at ICDE. lnkd.in/d5QUxvH Thanks to Alex Bogatu, Norman Paton, Mark Douthwaite and Stuart Davie for collaborating on this project! Blog post: lnkd.in/dZFi6CJ

Types of ML / NLP Papers

(2)📄Switching Contexts: Transportability Measures for NLP 😎@Mokanarangan, @fuzaguy, Philip Osborne and @andrenfreitas (3)📄Encoding Abstractive Knowledge Transitivity for Zero-shot Science Question Answering 😎Zili Zhou, @mrc_valentino , Donal Landers and @andrenfreitas

We are happy to announce we had 3 papers accepted at IWCS2021! (1)🗒️Do Natural Language Explanations Represent Valid Logical Arguments? Verifying Entailment in Explainable NLI Gold Standards 😎@mrc_valentino, Ian Pratt-Hartmann and @andrenfreitas (1/2) 🧵

STAR: Cross-modal STAtement Representation for selecting relevant mathematical premises (bit.ly/3aoUeTx) @debs_mf @andrenfreitas 🗣️ We combine Mathematical Language and Natural Language using STAR! #EACL2021 3/3

Unification-based Reconstruction of Multi-hop Explanations for Science Questions (bit.ly/3eec0Ku) @mrc_valentino @Mokanarangan @andrenfreitas 🗣️We propose an explanation retrieval method for QA that leverages explanatory patterns in the form of unification #EACL2021 2/3

This week we will be presenting two long papers at EACL2021 @eaclmeeting Unification-based Reconstruction of Multi-hop Explanations for Science Questions and STAR: Cross-modal STAtement Representation for selecting relevant mathematical premises #EACL2021 🧵 1/3

And of course a video demo! :D youtu.be/lE3O_BENBxk

Some references that have guided this work so far: arxiv.org/pdf/1909.03368… (Hewitt et al) arxiv.org/pdf/2004.03061…, arxiv.org/pdf/2010.02180… (Pimentel et al) arxiv.org/pdf/2009.07364… (Zhu et al)

We've found ourselves doing a lot of probing recently! To simplify the process, we've developed Probe-Ably: An extendable probing framework that automates the training, complexity control and results processing for your own representations and auxiliary task labels.

🐎🐎 .@SheffieldNLP has 10 papers accepted at #NAACL2021 and #EACL2021! Congratulations to all @SheffieldNLP authors! #NLProc

Our lab head @andrenfreitas joins @idiap_ch, with the @explAIn_lab now becoming a cross-institution research group idiap.ch/en/allnews/fou…

@kchonyc Thanks for this! Deborah Ferreira @debs_mf and our lab @explAIn_lab are newcomers on the scene, so this recognition is especially appreciated. Always wonderful when interests coincide nevertheless!

it's a double-embarrassment day. NaturalProof in wellecks.github.io/naturalproofs/…, which i loved working on, turned out also to be a repetition of the work by Deborah Ferreira arxiv.org/abs/2004.14959.