Applicative Abstract Categorial Grammars

[Alex]

I just wanted to raise awarenes of this research trend:

http://okmij.org/ftp/gengo/applicative-symantics/AACG.pdf

http://okmij.org/ftp/gengo/applicative-symantics/AACG1.pdf

These papers are generalization of categorial grammars and they provide 3 grammars - 1) abstract grammar (one between syntactic and semantic forms), 2) surface grammar for the real word text and 3) logical grammar for the logical expression that is the meaning of the abstract/surface form. Deterministic transformation rules allow to make inverse transformation from the surface form to the abstract form and direct transformation from the abstract form to the logical form/meaning.

So - these transformation rules can (in my opinion) encode the semantic/pragmatic knowledge gathered by the linguists and if we can learn to learn these rules automatically (e.g. by understanding maximization) then fantastic language processing system can be built.

I only wonder by these papers have not found wider application and recognitions, because the are issued in 2015 and I have not found any citation and further developmen. If someone knows then it would be great to get some further developments along these lines.

[Alex]

While learning this approach I was advised to look on transformational grammars - whole new world for me. Steedman says, that categorial grammars were more constrained approach that tried to tame the rich expressivenes of transformational grammars. So - are transformational grammars the generalization of categorial grammars and hence - better and stronger grammars? Can someone say some good words about them?

One should be able to build reasoning chatbots that uses logic and inference for their language understanding and generation. It is not good that so much resources are devoted to the statistical and neural - quite obsucre - methods for chatbots. Machine learning is that it is - it is learning from the given samples. But logical inference can create new conclusions, new concepts, new thoughts, it has inner creativity and that is what is expected from the truly intelligent systems!

[Alex]

i.e. all the grammars in which I have interest have the common feature - capability of the grammatical framework to transform the natural language sentence in the logical formula that gives the meaning of this sentence.

[Ivan Vodišek]

In my insights, logical formulas are merely yet another form by which we can process some reasoning in a certain, more or less well understood way. But, beside logic, I can imagine other forms of simulating human thoughts (take Lambda calculus for example). Moreover, even almighty logic has many forms like classical, fuzzy, modal, Bayesian, and who knows what other forms.

On the opposite part of the communication chain, we have our natural language that also comes in many flavors that are different in many ways. There can also exist other ways for us to send or receive informations, but I believe they can all be reduced to one dimensional streams of data.

So, I concluded that what I need for making a communication between a machine and a human is a set of translating rules that interchange informations between us. I came up with:

1. a syntax definition part that can define both human, and more or less hi-level language built into a machine
2. an operator for translating between those two.

For syntax definition I picked up some sort of context free grammar language, and for a translation operator I chose a classic functional language elements for defining functions and applying parameters. It turned out that the grammar language and the translation operators complement each other, making their disadvantages vanish in an extent I'm aware of. Thinking further, it really doesn't matter what happens at the ends of the communication chain, as long as we have a tool for translating between those two, but there were some people that disagreed with this thought.

- ivan v.-

2017-12-05 23:12 GMT+01:00 Alex <alexand...@gmail.com>:

i.e. all the grammars in which I have interest have the common feature - capability of the grammatical framework to transform the natural language sentence in the logical formula that gives the meaning of this sentence.

--
You received this message because you are subscribed to the Google Groups "opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email to opencog+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/38b46bc5-9322-4807-ad9f-23642d174b74%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.