[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [lojban] The efficacy of Lojban's grammar.

Kevin Reid, On 10/04/2010 15:47:

On Apr 10, 2010, at 9:51, And Rosta wrote:
I wager that syntactic structures that would be assigned to Lojban sentences by (1) syntacticians and (2) Lojban speakers would differ very substantially from the syntactic structures assigned by the formal grammar. 

[Disclaimer: I Am Not A Linguist.]
In my experience developing software which works with the results of parsing using formal grammar (well, the PEG version), the trees produced by the formal grammar are not like how I internally think of Lojban grammar, but insofar as they are, *they aren't what I would design as a formal AST for Lojban either*.
In particular, the trees have a huge number of nodes which pertain only to the implementation structure of the grammar and are both redundant and unrelated to the semantics of Lojban. This falls out from the fact that the parser produces one tree node per nonterminal, named according to that nonterminal, unconditionally: no appropriate specialized actions/transformations have been defined.
There are two major problems with the usefulness of the produced parse trees: 

  1. There are many nodes with exactly one child which reflect rules that
exist only due to the factoring of the grammar, or nested cases in order 
     to produce the proper parse tree for various optional clauses which
     usually don't exist.

  2. The nodes are named according to the nonterminal, not according to the
matched rule. This means that the names reflect the syntactic role, the slot it fills, rather than what the slot was filled with. The result of 
     this is that an interpretation of a given subtree has be inferred from
     the number and kind of child nodes rather than an actual symbol in the
     tree.
Particularly, note that the second problem is because the information *simply does not exist* in a formal system. The formal grammar(s) we have are simply defined to accept/reject sentences; the information about "what are these particular alternatives called" exists only informally in the CLL and other semantic-description documents. 

Very well put, much better than I could have done.


So:

I agree that the formal grammar produces weird structures.
However, I believe it would be possible to create a parser, or transform the output of the current parser(s), such that the structure *is* similar to what a syntactician, or a Lojban speaker who is familiar with parsers and formal grammars (such as for programming languages), would assign. 

I agree, at least if it is enough to do it on a rough-and-ready basis. Indeed, I once had an undergraduate do a dissertation with me on precisely this exercise. (Which in fact is where I first beheld the full horrors of Lojban parses in tree form.)
Furthermore, I believe this particular project *should* be done, as it would (a) aid the development of computer software which interprets Lojban, and (b) be a useful tool in discussing “what does this Lojban sentence mean” and making sure that the semantics of Lojban are fully defined.
I agree, in an ideal world, but it's not a simple job. Needs a (possibly amateur) syntactician with plenty of free time. Would that be you? 
---And.

--
You received this message because you are subscribed to the Google Groups "lojban" group.
To post to this group, send email to lojban@googlegroups.com.
To unsubscribe from this group, send email to lojban+unsubscribe@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/lojban?hl=en.