lizmat / ming

A small module for working with Stabler's Minimalist Grammars in Perl6.

MinG

A small module for working with Stabler's Minimalist Grammars in Perl6.

STRUCTURE

As of now there are five (sub)modules: MinG, MinG::S13, MinG::S13::Logic, MinG::EDMG and MinG::From::Text.

In MinG, you'll find the necessary classes and subroutines for creating descriptions of Minimalist Grammars. It's not of much use by itself unless you're planning to implement your own parser/etc. and want to save yourself the time of having to define classes and useful functions.

In MinG::S13, you'll find Stabler's (2013) "Two models of minimalist, incremental syntactic analysis" parser. Currently, this parser analyses all possibilities, while Stabler's parser discards low-probability derivations. There's a helper submodule called MinG::S13::Logic.

MinG::EDMG is currently under construction. To check what has been implemented already, you can view the milestone "EDMG Implementation".

Finally, in MinG::From::Text you'll find a parser that creates MinG::Grammar-s out of text descriptions of MGs.

More documentation can be found in HTML files inside the doc/ directory.

INSTALLATION

If you have perl6 and panda or zef, the following should suffice:

    zef update
    zef install MinG

If you don't, the easiest is probably to install rakudobrew https://github.com/tadzik/rakudobrew and then run:

    rakudobrew build moar --gen-moar --gen-nqp --backend=moar
    rakudobrew build zef

and you should be ready to install this module with zef.

The best option may be to install Rakudo Star http://rakudo.org/how-to-get-rakudo/ which comes with zef and some common modules. There's lots of tutorials about how to get perl6. Follow one of them and make sure you install zef (or panda).

EXAMPLE USAGE

As of now, someone who isn't interested in the inner workings of the module but wants to try out some minimalist grammars can easily create their grammar following this template:

START=F
word1 :: =F =F -F F
word2 :: =F F
:: =F =F F
.
.
.
word23 :: =F F ...

Without the dots, and changing wordi for your phonetic word (and, of course, changing F for whatever features you want your grammars to have). For two example grammars, check the resources/ directory.

You can save that in a file and call it using ming-analyser.p6 (you can use ming-analyser.p6-j to run it on the jvm if you have the jvm backend installed and you so wish). Assuming the file is in the directory $HOME/grammars/ and it's called gr0.mg, you can run:

ming-analyser.p6 $HOME/grammars/gr0.mg

Each line you write of input will be parsed using your grammar. You can parse several sentences in a series by separating them with a ';'. You can modify gr0.mg at any point and restart ming-analyser.p6 to have your new grammar working.

If you want to try out the example grammars, they can be accessed by passing the arguments --eng0 for a very small grammar of something-like-English, copied from Stabler (2013), --espa0 for a not-so-small (but small) grammar of Spanish written by myself, and --nihongo0 for a minimal grammar of japanese used to exemplify complement directionality in EDMGs. Like so:

ming-analyser.p6 --eng0
ming-analyser.p6 --espa0
ming-analyser.p6 --nihongo0

When inputting lines, pay attention not to put a final dot to your sentence. "dance." is a different word from "dance".

Default output is a list of derived tree descriptions in qtree format. If you have a latex distribution with pdflatex and qtree installed (under GNU/Linux) you can run ming-analyser with option compile=<filename.tex> to get a pdf with all derived trees for each sentence you input. For example:

ming-analyser.p6 --espa0 --compile=mytex.tex

Will compile a pdf named mytex.pdf with all derived trees each time you pass a sentence. Do note it will rewrite existing pdfs, so you have to restart the script if you want to get the trees of various sentences in pdf form.

If you want to call a grammar from a file, it should go after all flags (as of now, only the --compile flag can reasonable go with a file). For example:

ming-analyser.p6 --compile=mytex.tex $HOME/grammars/gr0.mg

CURRENTLY

• Has classes that correctly describe MGs (MinG::Grammar), MG-LIs (MinG::LItem) and MG-style-features (MinG::Feature).

• Has a subroutine (feature_from_str) that takes a string description of a feature (e.g. "=D") and returns a MinG::Feature.

• Has lexical trees for Stabler's (2013) parsing method (MinG::Grammar.litem_tree).

• Automatically generates LaTeX/qtree code for trees. (Node.qtree inside MinG)

• Has a working parser for MGs! (MinG::S13::Parser or MinG::S13.parse_and_spit())

• Has a parser that reads grammars from a file! (MinG::From::Text)

• Has an analyser script (ming-analyser.p6) that can be used to read grammars from files and analyse sentences from standard input.

TODO

• Allow some useful expansions of MGs (EDMGs are being implemented).

• Make the parser more efficient by adding probabilistic rule-following.

MAYDO

• Create a probabilistic trainer.

• Use annotated corpora to build lexical entries.

• Use a small subset of predefined lexical entries and a non-annotated corpus to "guess" the feature specification of unknown lexical items.

• Create a Montague-style semantics for MG trees.

• Create a world-model for a knowledgable AI using such semantics.

AUTHOR

Ian G Tayler, <[email protected]>

COPYRIGHT AND LICENSE

Copyright © 2017, Ian G Tayler [email protected]. All rights reserved. This program is free software; you can redistribute it and/or modify it under the Artistic License 2.0.