3 ms·

We need someone like https://twitter.com/justsaysinmice https://twitter.com/justsaysinmice, tagging NLP papers with "in English". The DisCoCat examples look li

by mci 7y ago

We need someone like https://twitter.com/justsaysinmice https://twitter.com/justsaysinmice, tagging NLP papers with "in English". The DisCoCat examples look like another riff on categorial grammar [1], which is not too adequate for languages with the concept of noun case. [1] https://en.wikipedia.org/wiki/Categorial_grammar https://en.wikipedia.org/wiki/Categorial_grammar

stereolambda 7y ago
This is true, seeing the title I was thinking of someone just rediscovering some fun tricks of Montague. Although I think in principle nothing stops categorial grammar from working for inflected languages if we agree to do some preprocessing/intermediate metalanguage conversion. The semantics of cases is largely analogous to what would be accomplished with prepositions in English. Anyway, I love the analogy on the page 4. He says that data-driven models in NLP (such as neural nets) may be similar to stacking Ptoleme's epicycles: they could fit the data better than simpler models from Copernicus to Newton, until we got the ~correct simple model understanding with Einstein. I don't know if this will be actually true for language, but I like it.
mathgenius 7y ago
Indeed, Lambek was one of the founders of categorical grammer, and from the article: """ Around 2004 I was giving a talk at the McGill category Theory seminar about our then new diagrammatic description of quantum teleportation. Lambek immediately pointed out: “Those are pre-groups!” The compact closed category-theory underpinning of quantum theory, have indeed pregroups as the posetal instance. It was this connection, between grammar and teleportation diagrams, that inspired the DisCoCat model, making it look as if word-meaning being teleported around in sentences by means of the channels provided by the pregroup grammar """

3 ms·

We need someone like https://twitter.com/justsaysinmice https://twitter.com/justsaysinmice, tagging NLP papers with "in English". The DisCoCat examples look li

by mci 7y ago

stereolambda 7y ago
This is true, seeing the title I was thinking of someone just rediscovering some fun tricks of Montague. Although I think in principle nothing stops categorial grammar from working for inflected languages if we agree to do some preprocessing/intermediate metalanguage conversion. The semantics of cases is largely analogous to what would be accomplished with prepositions in English. Anyway, I love the analogy on the page 4. He says that data-driven models in NLP (such as neural nets) may be similar to stacking Ptoleme's epicycles: they could fit the data better than simpler models from Copernicus to Newton, until we got the ~correct simple model understanding with Einstein. I don't know if this will be actually true for language, but I like it.
mathgenius 7y ago
Indeed, Lambek was one of the founders of categorical grammer, and from the article: """ Around 2004 I was giving a talk at the McGill category Theory seminar about our then new diagrammatic description of quantum teleportation. Lambek immediately pointed out: “Those are pre-groups!” The compact closed category-theory underpinning of quantum theory, have indeed pregroups as the posetal instance. It was this connection, between grammar and teleportation diagrams, that inspired the DisCoCat model, making it look as if word-meaning being teleported around in sentences by means of the channels provided by the pregroup grammar """