edu.stanford.nlp.trees

Class GrammaticalStructureConversionUtils

java.lang.Object

edu.stanford.nlp.trees.GrammaticalStructureConversionUtils

public class GrammaticalStructureConversionUtils
extends java.lang.Object

Contains several utility methods to convert constituency trees to dependency trees. Used by GrammaticalStructure.main(String[]).

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	GrammaticalStructureConversionUtils.ConverterOptions Enum to identify the different TokenizerTypes.

Field Summary

Fields

Modifier and Type	Field and Description
static java.lang.String	DEFAULT_PARSER_FILE

Method Summary

Modifier and Type	Method and Description
static void	convertTrees(java.lang.String[] args, java.lang.String defaultLang) Given sentences or trees, output the typed dependencies.
static java.lang.String	dependenciesToCoNLLXString(java.util.Collection<TypedDependency> deps, CoreMap sentence) Returns a dependency tree in CoNNL-X format.
static java.lang.String	dependenciesToCoNLLXString(GrammaticalStructure gs, CoreMap sentence) Calls dependenciesToCoNLLXString with the basic dependencies from a grammatical structure.
static java.lang.String	dependenciesToString(GrammaticalStructure gs, java.util.Collection<TypedDependency> deps, Tree tree, boolean conllx, boolean extraSep, boolean convertToUPOS)
static void	printDependencies(GrammaticalStructure gs, java.util.Collection<TypedDependency> deps, Tree tree, boolean conllx, boolean extraSep, boolean convertToUPOS) Print typed dependencies in either the Stanford dependency representation or in the conllx format.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

DEFAULT_PARSER_FILE

public static final java.lang.String DEFAULT_PARSER_FILE

See Also:

Constant Field Values

Method Detail

printDependencies

public static void printDependencies(GrammaticalStructure gs,
                                     java.util.Collection<TypedDependency> deps,
                                     Tree tree,
                                     boolean conllx,
                                     boolean extraSep,
                                     boolean convertToUPOS)

Print typed dependencies in either the Stanford dependency representation or in the conllx format.

Parameters:

deps - Typed dependencies to print

tree - Tree corresponding to typed dependencies (only necessary if conllx == true)

conllx - If true use conllx format, otherwise use Stanford representation

extraSep - If true, in the Stanford representation, the extra dependencies (which do not preserve the tree structure) are printed after the basic dependencies

convertToUPOS - If true convert the POS tags to universal POS tags and output them along the original POS tags.

dependenciesToCoNLLXString

public static java.lang.String dependenciesToCoNLLXString(GrammaticalStructure gs,
                                                          CoreMap sentence)

Calls dependenciesToCoNLLXString with the basic dependencies from a grammatical structure. (see dependenciesToCoNLLXString(Collection, CoreMap))

dependenciesToCoNLLXString

public static java.lang.String dependenciesToCoNLLXString(java.util.Collection<TypedDependency> deps,
                                                          CoreMap sentence)

Returns a dependency tree in CoNNL-X format. It requires a CoreMap for the sentence with a TokensAnnotation. Each token has to contain a word and a POS tag.

Parameters:

deps - The list of TypedDependency relations.

sentence - The corresponding CoreMap for the sentence.

Returns:

Dependency tree in CoNLL-X format.

dependenciesToString

public static java.lang.String dependenciesToString(GrammaticalStructure gs,
                                                    java.util.Collection<TypedDependency> deps,
                                                    Tree tree,
                                                    boolean conllx,
                                                    boolean extraSep,
                                                    boolean convertToUPOS)

convertTrees

public static void convertTrees(java.lang.String[] args,
                                java.lang.String defaultLang)

Given sentences or trees, output the typed dependencies.

By default, the method outputs the collapsed typed dependencies with processing of conjuncts. The input can be given as plain text (one sentence by line) using the option -sentFile, or as trees using the option -treeFile. For -sentFile, the input has to be strictly one sentence per line. You can specify where to find a parser with -parserFile serializedParserPath. See LexicalizedParser for more flexible processing of text files (including with Stanford Dependencies output). The above options assume a file as input. You can also feed trees (only) via stdin by using the option -filter. If one does not specify a -parserFile, one can specify which language pack to use with -tLPP, This option specifies a class which determines which GrammaticalStructure to use, which HeadFinder to use, etc. It will default to edu.stanford.nlp.parser.lexparser.EnglishTreebankParserParams, but any TreebankLangParserParams can be specified.

If no method of producing trees is given other than to use the LexicalizedParser, but no parser is specified, a default parser is used, the English parser. You can specify options to load with the parser using the -parserOpts flag. If the default parser is used, and no options are provided, the option -retainTmpSubcategories is used.

The following options can be used to specify the types of dependencies wanted:

• -collapsed collapsed dependencies
• -basic non-collapsed dependencies that preserve a tree structure
• -nonCollapsed non-collapsed dependencies that do not preserve a tree structure (the basic dependencies plus the extra ones)
• -CCprocessed collapsed dependencies and conjunctions processed (dependencies are added for each conjunct) -- this is the default if no options are passed
• -collapsedTree collapsed dependencies retaining a tree structure
• -makeCopulaHead Contrary to the approach argued for in the SD papers, nevertheless make the verb 'to be' the head, not the predicate noun, adjective, etc. (However, when the verb 'to be' is used as an auxiliary verb, the main verb is still treated as the head.)
• -originalDependencies generate the dependencies using the original converter instead of the Universal Dependencies converter.

The -conllx option will output the dependencies in the CoNLL format, instead of in the standard Stanford format (relation(governor,dependent)) and will retain punctuation by default. When used in the "collapsed" format, words such as prepositions, conjunctions which get collapsed into the grammatical relations and are not part of the sentence per se anymore will be annotated with "erased" as grammatical relation and attached to the fake "ROOT" node with index 0.

Keeping punctuation is the default behavior. This can be stopped with -keepPunct false

The -extraSep option used with -nonCollapsed will print the basic dependencies first, then a separator ======, and then the extra dependencies that do not preserve the tree structure. The -test option is used for debugging: it prints the grammatical structure, as well as the basic, collapsed and CCprocessed dependencies. It also checks the connectivity of the collapsed dependencies. If the collapsed dependencies list doesn't constitute a connected graph, it prints the possible offending nodes (one of them is the real root of the graph).

Using the -conllxFile, you can pass a file containing Stanford dependencies in the CoNLL format (e.g., the basic dependencies), and obtain another representation using one of the representation options.

Usage:
java edu.stanford.nlp.trees.GrammaticalStructure [-treeFile FILE | -sentFile FILE | -conllxFile FILE | -filter]
[-collapsed -basic -CCprocessed -test -generateOriginalDependencies]

Parameters:

args - Command-line arguments, as above