EUGENE

Revision as of 21:34, 28 July 2012

EUGENE is a natural language generation system. It generates natural language sentences out of semantic networks represented in the UNL format. In its current release, it is a web application developed in Java and available at the UNLdev.

The name

EUGENE is an acronym for dEp-to-sUrface GENErator

Requirements

As a universal engine, EUGENE must be parameterized to the target languages with the following files, to be provided through EUGENE's interface:

• The input document in the UNL document structure, i.e., the universal semantic network to be generated in natural language
• The UNL-NL (generation) dictionary, i.e., a lexical database where UWs are mapped into natural language entries, along with the corresponding features, to be provided according to the UNL Dictionary Specs
• The UNL-NL (generation) transformation grammar, i.e., a set of of transformation rules used to convert the UNL graphs into natural langauge sentences, to be provided according to the UNL Grammar Specs
• The UNL-NL (generation) disambiguation grammar, i.e, a set of disambiguation rules used to improve the results of the tokenization and of the transformation

to be provided according to the UNL Grammar Specs, to be provided according to the UNL Grammar Specs

Functioning

EUGENE performs the three following movements over the input file:

• Segmentation, i.e., the division of the input document into a series of isolated graphs, which are processed one at a time
• Tokenization, i.e., the identification of the tokens (UWs, relations and attributes) of each graph of the input document
• Transformation, i.e., the application of the transformation rules of the grammar over each tokenized graph in order to generate a natural language sentence

Quick start

As part of the UNLdev, EUGENE is available at [1]. You must be registered in the UNL^web in order to log in.
EUGENE has 5 tabs:

• The welcome tab
• UNL input, where you have to provide the UNL document be NLized. You may either create a new file or upload an existing file.
• Dictionaries, where you have to provide the UNL-NL dictonaries (i.e., the dictionaries to be used in natural language generation). You may either create a new file or upload an existing file. In any case, the dictionary must be provided according to the UNL Dictionary Specs. Once you create/upload a dictionary, you have to select it (by clicking the corresponding check box) and load it (by pressing the load button at the top menu). You may have several different dictionaries, and may load many of them to process the same corpus, but be sure that they are loaded in the correct order (because the order of the entries in the dictionary does matter for tokenization). You may reorder the dictionaries through the option "reorder dictionaries" at the top menu.
• T-rules, where you have to provide the UNL-NL transformation grammar (i.e., the grammar to be used to convert the UNL input into the NL output). You may either create a new file or upload an existing file. In any case, the grammar must be provided according to the UNL Grammar Specs, and must contain only transformation rules. Once you create/upload a grammar, you have to select it (by clicking the corresponding check box) and load it (by pressing the load button at the top menu). You may have several different grammars, and may load many of them to process the same corpus, but be sure that they are loaded in the correct order (because the order of the rules does matter for transformation). You may reorder the grammars through the option "reorder grammars" at the top menu.
• D-rules, where you have to provide the UNL-NL disambiguation grammar (i.e., the grammar to be used to control the tokenization and improve the results of the transformation grammar). You may either create a new file or upload an existing file. In any case, the grammar must be provided according to the UNL Grammar Specs, and must contain only disambiguation rules. Once you create/upload a grammar, you have to select it (by clicking the corresponding check box) and load it (by pressing the load button at the top menu). You may have several different grammars, and may load many of them to process the same corpus, but be sure that they are loaded in the correct order (because the order of the rules does matter for disambiguation). You may reorder the grammars through the option "reorder grammars" at the top menu.
• EUGENE console, where you will get the results. The IAN console brings the list of sentences appearing in the NL input, which may be processed one at a time, or in a range. The results are displayed in 5 different trace levels.

Test drive

You may test the system using the resources below:

• UNL input: corpus500_unl.txt, to be uploaded to the tab NL input (don't forget to select and load the file after uploading it)
• T-grammar: eng_gen_tgrammar.txt, to be uploaded to the tab T-rules (don't forget to select and load the file after uploading it)
• D-grammar: eng_gen_dgrammar.txt, to be uploaded to the tab D-rules (don't forget to select and load the file after uploading it)