Introductory Questions

[Julius Hamilton]

Hey,

I was really interested in learning languages and also teaching English and I can’t remember when but at some point it often occurred to me that I often did not know the rules of a language precisely or comprehensively enough. I did read parts of some authoritative reference grammars but I sometimes felt it was not easy enough to look up one single governing a situation or explaining a mistake, and I also increasingly wanted very strict formalization, really just a list of rules, very much like formal logic or the axioms of mathematics.

Since then I feel like I have asked this question so many times in forums and slowly learned a bit more about different topics in linguistics but I really don’t think I found what I was looking for until I finally discovered this project. Someone had tipped me off about Head Driven Phrase Structure Grammar. While it looks fascinating, so far I have had the same impression as I did from the Cambridge Grammar of the English Language, that it’s more of an organized survey over various topics in language, but it doesn’t have that strictly sequential, cumulative aspect of only stating precise rules that build on previous rules.

I do believe HPSG is actually designed to be that formalized and hence it’s used for computerized parsing of languages. The only thing I struggled with was I still could not conveniently find a list of its actual rules, like the complete (or near complete) grammar of English, as a context-free grammar, or in Extended Bacchus Naur form, or in the HPSG formalism, or what have you.

It took me so long to find this project which appears to be exactly what I’ve been trying to seek out for so long, actually years. It’s all about analyzing natural language formally. And there appears to be really interesting applications of it like making a sort of formalized encyclopedia as well. This is totally in line with things I have been thinking about myself for a long time.

A couple of the biggest reasons I would like an explicit list of rules of a language, even if it’s not 100% accurate or there are various theoretical reasons why it’s not fully accurate, are:

1. I feel like it will make teaching English so much easier because you should be able to reduce a given mistake to one exact and obvious rule.

2. I actually want to learn languages in a sort of “combinatorial” way. For the longest time I have been dreaming of a very simple app, sort of like Duolingo, where it basically drip-feeds you one of the rules, and a small set of the lexicon, and you practice arranging the lexicon as it fits into the rule. As you show competence, you get a second rule that you can now use, and some more of the lexicon. You could stairstep your way up to more or less grammatically perfect knowledge of a language by yourself on your smartphone. I really, really want this try to make this.

3. I feel like I will understand the theoretical objections to context-free grammars better when I have actually worked with one and can then understand what its shortcomings are.

I have found here what I believe are the GF rules for Turkish: 

https://github.com/GrammaticalFramework/gf-rgl/tree/master/src/turkish

I guess I’ll start reading the tutorials about how those rules work. But if anyone would like to offer tips or some kind of model code to help me get a clearer idea, I’d appreciate that.

The basic idea is starting with the most fundamental rule, I assume it’s something like “S -> NP + VP” or something. I see a number of folders there so I’m not sure which one I should start with, but I guess I’ll just read any tutorials.

Interested in getting any external feedback, ideas, or advice.

Thanks,

Julius

P.s.:

How do you create such a grammar, for a language that doesn’t have one? Do you have to be versed in that language’s grammar already, or can it possibly be computationally generated?

Thanks very much 

[Julius Hamilton]

[Krasimir Angelov]

Hi Julius,

This is indeed the Turkish grammar:

On Tue, 8 Nov 2022 at 10:18, Julius Hamilton <juliusha...@gmail.com> wrote:

I have found here what I believe are the GF rules for Turkish: 

https://github.com/GrammaticalFramework/gf-rgl/tree/master/src/turkish

but bear in mind that it is far from complete. All languages are developed by volunteers and naturally some grammars are better developed than others.

 

The basic idea is starting with the most fundamental rule, I assume it’s something like “S -> NP + VP” or something. I see a number of folders there so I’m not sure which one I should start with, but I guess I’ll just read any tutorials.

Actually a more natural starting point is the lexicon. You need at least some nouns and verbs before you can even build a single sentence. The rule "S -> NP + VP" is what we call PredVP. This is a function which takes an NP and VP and returns an object of type Cl, i.e. clause.

 

How do you create such a grammar, for a language that doesn’t have one? Do you have to be versed in that language’s grammar already, or can it possibly be computationally generated?

All the grammars so far have been manually created but I believe that part of the process can be automated. In any case if you want to guarantee quality then human touch cannot be avoided. In the best case the grammar writer also understands the language but exceptions are possible. For example some of the grammars were written by people who know GF but have relied on native speakers or grammar books to gain knowledge of the language.

Best Regards,

Krasimir