NAME
Treex::Core - interface to linguistic structures and processing units in Treex
VERSION
version 0.06571
SYNOPSIS
use Treex::Core;
my $doc = Treex::Core::Document->new;
my $bundle = $doc->create_bundle;
my $zone = $bundle->create_zone('en');
my $atree = $zone->create_atree;
my $predicate = $atree->create_child({form=>'loves'});
foreach my $argument (qw(John Mary)) {
my $child = $atree->create_child( { form=>$argument } );
$child->set_parent($predicate);
}
$doc->save('demo.treex');DESCRIPTION
Treex::Core is a library of modules for processing linguistic data, especially tree-shaped syntactic representations of natural language sentences, both for language analysis and synthesis purposes.
Treex::Core is meant to be as language universal as possible. It makes only a few assumptions: the language's written form must be representable by Unicode characters, and it should be possible to segment texts in such language into sentences (or sentence-like units) and words (or word-like units).
Treex::Core is tightly coupled with the tree editor TrEd, which makes browsing the linguistic data structures very comfortable.
Treex::Core uses TrEd's Treex::PML for the memory representation, as well as for storing the data into *.treex files, using the XML-based Prague Markup Language.
Zones parametrized by language codes and selectors
Treex documents can contain parallel texts in two or more languages, as well as alternative linguistic representations (such as two dependency parses of a same sentence, resulting from different parsers). Such contents of the same type are separated by introducing zones.
Zones (classes derived from Treex::Core::Zone) are parametrized by language ISO codes, and optionally also by so called selectors. Selector can be any string identifying the source or purpose of the given piece of data. It can distinguish e.g. reference translation from machine-translated text, or the most probable parse of a given sentence from the second most probable parse. In Treex data structures, zones are used at two levels:
- Treex::Core::DocZone - allows to have multiple texts stored in the same document
- Treex::Core::BundleZone - allows to have multiple sentences and their representations in each bundle.
As for Treex processing units (scenarios and blocks, see below), each processing unit either limits itself to a certain zone, or it can be zone-parametrized (especially in the case of language-universal blocks).
Data structure units
In Treex, linguistic representations of running texts are organized in the following hierarchy:
Documents
The smallest independently storable unit is a document (Treex::Core::Document).
Technically, each document consists of a set of document zones, and of a sequence of bundles.
Document zone
A document can contain one ore more zone (Treex::Core::DocZone), each of them containing a text. =head3 Bundle
A bundle (Treex::Core::Bundle) corresponds to a sentence (or a tuple of parallel or alternative sentences) and all its (or their) linguistic analyses.
Technically, a bundle contains a set of bundle zones.
Bundle zone
Bundle zone (Treex::Core::Bundle) contains one sentence and at most one its linguistic analysis for each layer of analysis. The following layers are currently distinguished:
* a-layer - analytical layer (surface syntax dependency layer) merged with the morphological layer, as defined in the Prague Dependency Treebank.
* t-layer - tectogrammatical layer (deep-syntactic dependency)
* p-layer - phrase-structure layer
* n-layer - named entity layer
Each layer representation has a form of a tree, represented by the tree's root node.
Node
Each node has a parent (unless it is the root) and a set of predefined attributes, depending on the layer it belongs to. There is an abstract class Treex::Core::Node defining the functionality which is common to all types of trees (such as functions for accessing node's parent or children). Functinality specific for the individual linguistic layers is implemented in the derived classes:
Attributes
Nodes contain attribute-value pairs. Some attributes are universal (such as identifier), but most of them are specific for a certain layer. Even if node instances are regular Moose objects (i.e., blessed hashes), node's attributes should be accessed exclusively via predefined accessors.
Attribute values can be plain or further structured using PML data types (e.g. sequences), according to the PML schema.
Processing units
Block
Blocks (descendants of Treex::Core::Block) are the smallest processing units applicable on Treex documents.
Scenario
Scenarios (instances of Treex::Core::Scenario) are sequences of blocks. Blocks from a scenario are applied on a document one after another.
Support for visualizing Treex trees in TrEd
Treex::Core also contains a TrEd extension for browsing .treex files. The extension itself is only a thin wrapper for the viewing functionality implemented in Treex::Core::TredView.
AUTHOR
Zdeněk Žabokrtský <zabokrtsky@ufal.mff.cuni.cz>
Martin Popel <popel@ufal.mff.cuni.cz>
David Mareček <marecek@ufal.mff.cuni.cz>
COPYRIGHT AND LICENSE
Copyright © 2011 by Institute of Formal and Applied Linguistics, Charles University in Prague
This module is free software; you can redistribute it and/or modify it under the same terms as Perl itself.