An introduction (for absolute beginners) ideas of grammar, grammatical structures, lexicon, lexical types, grammar-based random generation of sentences, parsing of sentences.
Examples of things that can be done by learners after being introduced to the grammar formalism and the library designed for using the grammars to generate or parse sentences.
http://www.cs.bham.ac.uk/research/projects/poplog/examples/robolang/grammar.p
Lightly commented, Pop-11 source code for the LIB GRAMMAR package that makes all the above work. It depends on use of the Pop-11 matcher for list structures. [ADD LINK]
http://www.cs.bham.ac.uk/research/projects/poplog/examples/robolang/newgrammar.p
A modified version of Pop-11 LIB GRAMMAR to support use of an ill-formed substring table to speed up parsing (with spectacular results).
This, like the original library, finds only one parse, even if the input string is ambiguous according to the grammar.
To find where this differs from the original library look for occurrences of the global variable no_parse_found and calls of the procedure setup_noparses();
A grammar for an artificial subset of English relevant to a conversational robot.
This shows how, using Pop-11's formalism for recursive context-free grammars, it is possible to define a fairly complex grammar for assertions, questions and commands, for the domain of a robot with the ability to perceive and manipulate objects on a table.
(As in the CoSy and CogX robot projects at Birmingham.)
Using the standard Pop-11 grammar library, parsing sentences of the sorts generated by this grammar can be very slow.
Using newgrammar.p, with the ill-formed substring memory, speeds that up enormously. (The memory is cleared and rebuilt for each sentence. The program could be changed to create an enduring memory of things not to try.)
This grammar was hand-coded, using an idea suggested long ago by Gerald Gazdar, namely compiling semantic categories into syntactic categories. In a more ambitious project it might be possible to make a system that can learn such a grammar by being exposed to a very large corpus of examples.
A lexicon for use with the above grammar, i.e. words to be used in statements, commands, and questions referring to the robot and its environment and actions.
This package has no understanding of inflections: different forms for the same word, depending on context of use. All words are treated as unanalysed wholes, except for the use of apostrophes.
Example code for compiling and testing the grammar and lexicon, using the tools in newgrammar.p
This includes some examples that invoke the linux 'espeak' program to say the generated sentences out loud.
One test sentence parsed, to show the sort of parse tree produced. The sentence is
[move the big red box on the tray onto the top of the block on the left side of Jeremy"s triangular horse]The parse tree is too big to be displayed easily, though it can be displayed in an xterm-compatible window, or using the poplog XVed editor.
A screen-shot of the tree displayed in an xterm window with a small font is below,
automatically generated by the Pop-11 'showtree' library package.
