ParallelMeaningBank

Woke up this morning and found out that being an old-timer is pretty cool!

Ready to present my poster on Transparent Semantic Parsing with Universal Dependencies using Graph Transformations at #COLING2022

🎉10 Years of #BabelNet Workshop is now over: what a great experience!🎉 Thanks to everyone who participated both in person and online! @SapienzaNLP @babelscape @ERC_Research @AccademiaCrusca @RNavigli #workshop @WikiResearch #BabelNet10anniversary

Impressive work by Bruno Guillaume on graph rewriting for computational semantics, including a very cool search interface for Parallel Meaning Bank data: @4.0.0" target="_blank" rel="nofollow noopener">semantics.grew.fr/?corpus=en_gol…

@FiboKowalsky @andreasvc It actually is strictly standoff for the taggers and parsers in the PMB; there is always a connection between char offsets and derived tokens. If the token segmentation is changed, annotation is preserved for (unaffected) tokens even if the token indices have changed.

@andreasvc Asking for Rik? 😄 How serious is your question? 🤔 I don't think we use standoff annos in some standard format (if that's u r asking). In PMB, tools like parsing and tagging assume tokenized text, and their standoff annos follow token indices (which is not strictly standoff)

The original question wasn't very clear, so i tried to write what i am looking for in a more elaborate / concrete way:

@andreasvc @FiboKowalsky Well, by definition the raw text shouldn't change! In your case the raw input isn't text, but an image of text. An interesting challenge.

@FiboKowalsky No for myself. I got the impression that standoff annotation assumes that raw text never changes but I'd love to be proven wrong. Even character indices may change if e.g. you need to fix an OCR error.

@SapienzaNLP @RealAAAI @RNavigli @rexhina_b @CarlosMalaga26 @babelscape @ERC_Research @knowgraphs @elexis_eu The PMB meaning representations are language-neutral by design (unlike AMR and original DRS).

AMR & friends are language-specific. Looking for language-independent sentence representations? Check out BMR: BabelNet Meaning Representation @RealAAAI #AAAI2022 by @RNavigli @rexhina_b @CarlosMalaga26 @babelscape researchgate.net/publication/35… @ERC_Research @knowgraphs @elexis_eu

About ten days left to apply! Don't miss out on this great opportunity to join our Computational Linguistics group in Groningen! #NLProc @univgroningen @FacultyofArtsUG

This month's ILFC seminar @GdrLift was on variable-free meaning representations, presenting a new format that provides easy human and machine annotation of meanings. Video and slides are available here: gdr-lift.loria.fr/monthy-online-… present.v.08 Agent +1 person.n.01 Name "Johan Bos"

@ryandcotterell Indeed the Parallel Meaning Bank has lambda-expressions attached to the CCG derivations. You can see them in the PMB Explorer (sentence tab, drs, then click on "unfold all"). And yes we could release them!

@zehavoc @FiboKowalsky @rikvannoord @evanmiltenburg @srchvrs @nasrinmmm @stanfordnlp @GroNlp If you click on "discourse" you get the entire picture. But I cheated here and added a bit of wisdom. Anyway, Boxer will not be able to resolve these Winograd pronouns in an "intelligent" way. Because it will need the knowledge that monkeys can be hungry and banana's can't.

@FiboKowalsky @rikvannoord @evanmiltenburg @srchvrs @nasrinmmm @stanfordnlp @GroNlp @MeaningBank there must be a bug somewhere bc from what i'm seeing it got right. Look at the box for "it" where it correctly points to b1.

I'm sick of opening my NLU talks with the same exact classic motivating example for 10 years now!! (and I am sure others are too). Someone at @stanfordnlp should just fix this one-off so that I can't use it anymore and we can all assume the state has changed!😄

@texttheater @evanmiltenburg @srchvrs @zehavoc @nasrinmmm @stanfordnlp @rikvannoord's Neural Boxer is of course available. And I think @Ginger_in_AI has a version of the classic (Prolog-based) Boxer in his github. There isn't much demand anymore, although recently I got a few more requests --maybe people are getting tired of the neural approaches?

@evanmiltenburg @srchvrs @zehavoc @nasrinmmm @stanfordnlp BTW @MeaningBank, are there plans for open-sourcing more recent versions of Boxer? I believe it has come a long way since then...

We are looking for someone who can teach a 5 ECTS Human-Computer Interaction course in our Bachelor in Information Science (second year) at the University of Groningen, NL, in Feb-March 2022. Details 👇

@FiboKowalsky @GroNlp @FacultyofArtsUG Aha, good question! Because AMR/PENMAN has the expressive power required. Negation and universal quantification, for which DRS would be a better choice, is seldomly found within tombstone inscriptions... Plus: the simple structures of AMR are closer to RDF triples.

@MeaningBank @GroNlp @FacultyofArtsUG Why not DRS-like structures? DRTS (Discourse Representation TombStone) sounds much better 😄

Just appeared in the International Journal of Digital Humanities: "A semantically annotated corpus of tombstone inscriptions", by Johan Bos @GroNlp @FacultyofArtsUG. Open access here: link.springer.com/article/10.100… Meanings are represented by AMR-like structures!

Today we released a new stable version of the PMB data, version 4.0.0: pmb.let.rug.nl/data.php More gold data for English, Dutch, German and Italian, and lots of silver data.

Not bad at all...

Chunliu Wang and Johan Bos presented recent work on the Parallel Meaning Bank at the 2021 Workshop of Logic, Chinese and AI (逻辑、汉语与人工智能研讨会）

📢 new Dr in the group! 🥳🥳🥳 We are proud to announce that our own Rik van Noord (@rikvannoord) successfully defended his thesis on neural semantic parsing yesterday, and was awarded a *cum laude* distinction! Congratulations, Rik! Link to the thesis: research.rug.nl/en/publication…

We are hiring! Are you a computational linguist who wants to work in a wonderful group? Then apply for this lecturer position in Groningen with us! Deadline May 9th, start date September 1st, all info at this link 👇 rug.nl/about-ug/work-… #NLProc