XMG - eXtensible MetaGrammar

Benoit Crabbé Denys Duchier Joseph Le Roux Yannick Parmentier

Calligramme / TALARIS projects - LORIA

provides

MetaTAG

MetaIG

Latest news

• Since Oz-Mozart 1.x.x (which includes the Constraint Solving support used by XMG) is only available for 32-bits architecture, we recommand you to install XMG on a virtual machine.
  To do so, you can e.g. follow the steps in this file.
• The eXtensible MetaGrammar workshop took place in Nancy, France on June 2007, 21-22. More information can be found here.
• As of Dec. 2006, the CVS repository is no longer maintained, please use the Subversion repository instead.

---

Overview

eXtensible MetaGrammar (XMG) is a tool for designing large scaled grammars for natural language. It implements an alternative language of grammatical representation to the classical languages used for this purpose such as PATR II. Provided a compact representation of grammatical information, XMG combines elementary fragments of information to output a fully redundant strongly lexicalised grammar.

Some underlying ideas

XMG is a framework for describing grammatical information without redundancy. More specifically it has been designed to ease the development of strongly lexicalised syntactic formalisms by means of elementary descriptive fragments. The system consists of two main components: a control device aimed at controlling how fragments are to be combined together in order to describe meaningful lexical descriptions and a composition system aimed at resolving how elementary descriptions are to be combined.

Technically, the control device is understood as a logic program of the DCG paradigm, where the terminal symbols are fragments describing some structures. The composition system is a model builder that allows to parametrise the structures output by the compiler (currently trees for Tree Adjoining Grammars and D-Trees for Interaction Grammars). The model-theoretic approach to describing the lexicon also allows linguists to further use an extensible library of constraints, called principles, ensuring additional well-formedness conditions on the output structures.

Features

XMG Currently implements :

• The core engine
• Extensions for multiple dimensions frameworks
• Extensions for handling name spaces
• Two possible output structures : trees and d-trees
• A library of principles used for the development of a French TAG
• XML output available (Loria's TAG parser and Dyalog TAGML2 compliant)
• GUI output, allowing to view (parts or the whole of) your generated grammar.

Current version: 1.1.6

Documentation

The XMG documentation is available as a wiki here, you can download a printable pdf version here.

Moreover we suggest that you read the following papers:

• Metagrammar Redux - CSLP'2004 [.ps] [.pdf]
  
• The Metagrammar Compiler: An NLP Application with a Multi-paradigm Architecture - MOZ'2004 [.ps] [.pdf]
• Grammatical Development with XMG - LACL'2005 [.ps] [.pdf]
  
• XMG: Un Compilateur de Méta-Grammaires Extensible - TALN'2005 [.ps] [.pdf]
• XMG: a Multi-formalism Metagrammatical Framework - ESSLLI'2005 [.ps] [.pdf]

Further information is also available on the XMG Project page at sourcesup (last news, forums, archives, etc).

In particular, one may have a look at the documentation of Benoît Crabbé's TAG for French (documentation in French).
The metagrammar from which this grammar is compiled is available in the XMG Subversion repository. That is, once you downloaded the XMG sources via:
svn checkout http://subversion.renater.fr/xmg local_repository The metagrammar is available under local_repository/trunk/METAGRAMMAR/FrenchTAG.

Requirements

To use XMG, you require the Oz/Mozart system, which you can download at the following adress:

• http://www.mozart-oz.org (Current version: 1.3.2)
  

This system works under the following platforms:

MacOS X

Linux

Windows (95/98/ME/NT/2000/XP)

Notice that under MacOS, you need to install the MacOS SDK in order to have a C++ compiler. Similarly, under Windows, you need to have access to a GNU c++ compiler (for instance by using cygwin).

Make sure you have Tcl/Tk installed and activated. In particular, under Gentoo, add dev-lang/mozart tk tcl in the file /etc/portage/package.use.

You also need to install the "Select" library available at:

• duchier-select__1.3.0__source__1.8.pkg
  

Once Oz-Mozart is installed, just go to the directory containing the Select package and type:
ozmake --install --package=duchier-select__1.3.0__source__1.8.pkg NB: the ozmake command is now part of the Oz-Mozart system. Once you have installed Oz-Mozart, you do not need to install any additional package before invoking ozmake.

Please note there is a bug in the version 1.3.2 of the Oz / Mozart system under windows making the installation of the select package hang. This bug does not appear with version 1.3.1 for windows. Furthermore the other (linux, mac, etc.) packages are not affected with this bug.

Download

The easiest way to get XMG is either:

1) to install it as an ozmake package, which is available here (current version: xmg-1.1.6.pkg).

See the installation instructions.

2) to compile it from the sources.

a) You can download the last version of the sources from a Subversion repository with anonymous access (last updates, recommended). The commands are the following:
svn checkout http://subversion.renater.fr/xmg local_repository Once you have imported the sources, invoke: cd local_repository/trunk/MGCOMPILER/
ozmake --upgrade
b) Here is a stable release (current version: xmg-1.1.6.tar.gz)

Furthermore, here are some additional useful documents:

• some examples, toy metagrammars, implemented with XMG :
  • Small French TAG using non-colored tree descriptions
  • Small French TAG augmented with flat semantics (using colored tree descriptions)
  • Small French IG
• DTDs describing the XML format of the (TAG, MCTAG and IG) grammars produced by XMG
  • xmg-tag.dtd,xml
  • xmg-mctag.dtd,xml
  • xmg-ig.dtd,xml
• configuration files to write metagrammars with emacs or vim (courtesy of Eric Kow)

Once you have downloaded the tarballs to a local file e.g. xmg-xxx.tar.gz on your computer, then invoke: tar xzvf xmg-xxx.tar.gz
cd xmg-xxx
ozmake --upgradeThis will install the two executables in your ~/.oz/1.3.2/bin directory (don't forget to update your $PATH).

NB: if you installed the packages built with Oz-Mozart 1.3.1, you may have to create a symbolic link between ~/.oz/1.3.1/bin and ~/.oz/1.3.2/bin.

For Mac OS X users, here is a nice documentation for installing XMG and its dependencies (courtesy of Patrick Watrin).

Installation

This contribution is distributed either as an ozmake package or as sources. To install xmg's ozmake package, download it to a local file e.g. xmg-xxx.pkg on your computer, then invoke: ozmake --install --package=xmg-xxx.pkg If this package was already installed on your computer and you want to upgrade to a newer version, invoke: ozmake --upgrade --package=xmg-xxx.pkg or else ozmake will complain. By default ozmake installs all binary files in the user's ~/.oz/1.3.2/bin directory tree.

NB: the ozmake command is now part of the Oz-Mozart system. Once you have installed Oz-Mozart, you do not need to install any additional package before invoking ozmake.

Usage

MetaTAG metagrammar_file
Compile a TAG metagrammar to produce a TAG 

MetaIG metagrammar_file
Compile an IG metagrammar to produce an IG

MetaTAG's Options

--gui
Prints the TAG trees through a Qtk GUI (default)
--xml
Prints the TAG trees in an XML format, in stdout!
-o file.xml
Specify the name of the output file (here "file.xml")
--chk
Prints the TAG trees in a binary format (persistent data, see Pickle) for using with the XMG tools
-c file.rec
Specify the name of the binary output file (here "file.rec")
--mac
Prints semantic information (macros) in a file specified by the "-s" option.
-s file
To specify the name of the file containing semantic macros (.mac extension added automatically).
--lin
Prints the links between valuations and semantic classes in a file specified by the "-l" option.
-l file
To specify the name of the file containing links valuation - semantic classes (.lin extension added automatically). -s Int
To split the grammar in "Int" several files (m stands for many).

MetaIG's Options

--xml
Prints the IG tree descriptions in an XML format in stdout! (default)
-o file.xml
Specify the name of the output file (here "file.xml")

(A GUI to visualize the d-trees of the Interaction Grammar is available within the LEOPAR system, see Related Softwares).

Snapshots

• XMG's GUI (from valuation to Trees)
• TAG tree representing a transitive verb
• TAG tree associated with a flat semantic formula

XMG tools

We have developed additional tools to ease metagrammar conception or integration with other applications. Among these:

• classbrowser (classbrowser module): graphical interface in perl/Gtk2 allowing one to explore the metagrammar to observe the relations between classes (provides functionalities comparable with javadoc, see the README).
  Usage:
  perl main.pl metagrammar_file
• hierarchy_printer (hierarchy_printer module): perl programme used to extract from a metagrammar the class hierarchy graph. The output is a set of pdf graphs (see the README). Courtesy of Ulrike Fleury.
  Usage:
  perl hierarchie_classes.pl valuation_file
• CheckTAG (part of the CHECKER module): program used to remove duplicated trees (i.e. identical trees that are produced by combining different tree fragments).
  Usages:
  CheckTAG GrammarFile.rec
CheckTAG GrammarFile.rec --gui
CheckTAG GrammarFile.rec --xml -o FilteredGrammarFile.xml
CheckTAG GrammarFile.rec --chk -c FilteredGrammarFile.rec
• ViewTAG (VIEWER module): graphical interface allowing one to search for specific trees in the produced grammar, either from a tree name or a list of fragments (i.e. class names).
  Usages:
  ViewTAG
ViewTAG GrammarFile.rec
ViewTAG GrammarFile.rec TreeName
• SelectTAG (SELECTOR module): anchoring module, taking as input a grammar in a binary format (Pickle) and a specification of lexical entries, and produces as output anchored TAG trees.
  Usages:
  SelectTAG GrammarFile.rec FilterFile --gui
SelectTAG GrammarFile.rec FilterFile --xml -o OutputFile
SelectTAG GrammarFile.rec FilterFile --chk -c OutputFile
• XMGExtractor (XMGExtractor module): graphical interface in Java allowing to display the inheritance hierarchy of classes (see the documentation for installation). Developed by Brice Ambrosiak.
  Usage:
  XMGExtractor metagrammar_file

All these programs are included in the Subversion repository (so they have been imported when you downloaded the xmg sources). The installation process is identical to XMG's (except for the classbrowser, hierarchy_printer and XMGExtractor for which there is no need to type "ozmake --upgrade"), i.e.: cd local_repository/trunk/XMG-TOOLS/MODULE-NAME
ozmake --upgrade

Related softwares

The automatically produced grammars can be used for parsing. We are currently working on integration of XMG with TAG and IG parsers.

TAG parsers

• LORIA LTAG Parser 2 (LLP2)
  The XSLT script to convert grammar from XMG's XML format to LLP2's is available here.
  
  
• DyALog
  The XSLT script to convert grammar from XMG's XML format to DyALog's is available here.
  
  
• The Tuebingen Linguistic Parsing Architecture (TuLiPA)
  No XSLT conversion is needed if you use the XMG-Tuebingen branch, otherwise (i.e. you are using XMG main branch) the XSLT script to convert XMG's XML format into TuLiPA's is available here.
  
  NB: note that if you compiled XMG via the sources in the subversion repository, you can produce XML for TuLiPA directly by invoking: MetaTAG valuations.mg --tulipa -o grammar.xml

You may also be interested in Semantic Construction during parsing with TAG, see SemConst and TuLiPA.

IG parser

• eLEctrOstatic PARser (LEOPAR)

Generation with TAG

• GenI
  

GenI is a surface realiser for TAG, which works with TAG produced using XMG (provided they are equipped with semantics).

---