🧵

New developments show promise for mathematically modeling natural language in alignment with human cognition. A thread on integrating Lambek calculus, quantum computing, and text circuits:

1/n Lambek calculus elegantly captures the compositionality and context-dependence of language using typed logic. Sentences have atomic types like Noun, Verb, etc. Rules compose types into larger structures. But modeling large texts classically becomes intractable.
2/n

A fantastic talk by Prof. @gemmaboleda of @putaupf, on #colexification at @abralin_oficial covering almost anything discussed in her paper jointly by Prof. Thomas Brochhagen also of @putaupf (Link to the paper and the author's webpages can be found in the comments). 1>11 🧵 "Here is summary of what the talk and the corresponding paper are all about:
"When do languages use the same word for different meanings? The #Goldilocks #Principle in the #lexicon 2>11

Finally, somebody in NLP started talking about structures! Semantic, syntactic, and their interplay.
If you're profoundly curious about the 'Semantic Structure in Deep Learning', this paper by Dr. Ellie Pavlick of @BrownUniversity might be a good fit for you. 1>3 🧵 #ai #nlp annualreviews.org/doi/abs/10.114… 2>3

Check out this wonder paper by Prof. @JudithTonhauser & Dr. Judith Degen. After reading it, I can say I recovered somehow from depression and got a bit more motivated. Tribute to both of these venerable scholars!
1/4 🧵 I'll put the link to their pages, in case you're interested in their research interests.
Link to the paper:
ling.auf.net/lingbuzz/006771
2/4

What happened to stratificational (neuro-cognitive) linguistics? It had some drawbacks, but was (and still can be, imho) a very efficient approach to bridge the gap btw. the symbolic and connectionist approaches to language. Even Prof. Lamb's Lab, has not been updated 1/4 since 2010. Here are the links to his Lab and his most recent paper 'Linguistic structure: A plausible theory (2016)'. When it comes to the interfaces, the need for transduction (unlike computation) increases, since symbolic (and digital) items are to be encoded to fuzzy 2/4