This application is a command line tool to map synsets along with their respective Supersenses[1][2] (Lexical Files) to the Semantic Types described by Dixon (2005)[3]. This code is intended to be used along with a local installation of Princeton's WordNet (US-English).
- Why mapping Supersenses to Semantic Types?
- Preparing the environment
- Running the application
- After Running
- License
Supersenses are high level generic classes intended to serve as top hypernyms to WordNet's synsets. They are derived from the work by Ciaramita and Johnson (2003)[1] and Ciaramita and Altun (2006)[2], and have been implemented in WordNet since early versions. Currently WordNet describe 45 Supersenses according to the distribution shown by the table below:
| Open Word Class | Quantity of Supersenses |
|---|---|
| Noun | 26 |
| Verb | 15 |
| Adjective | 3 |
| Adverb | 1 |
On the other hand, another kind of high level classes have been proposed, not only grouping words by common meaning, but also describing some characteristics they share. These classes are the Semantic Types, proposed by Dixon (2005)[3]. Semantic Types are able to indicate whether a word refers to something that should be counted by mass instead of unit (Semantic Type "Quantity") or something that refers to an abstract concept (Semantic Type "Other").
As proven by other works such as Castro et.al. (2011)[4], the identification of which Semantic type is related to a word may provide enough semantic to associate a word to a construct of a well-founded ontology language such as OntoUML, based on the Unified Foundational Ontology by Guizzardi (2005)[5].
This mapping tool is being developed under as a part of the research I'm currently developing as a M.SC. candidate at the UNIRIO (Universidade Federal do Estado do Rio de Janeiro). For further information please refer to my website.
The application uses a local installation of Princeton WordNet, (version 3.1) and does it through MIT JWI API which requires a environment variable (WNHOME) to be set.
To install WordNet you can simply refer to the website project and download the version that best fits your Operational System. Remember to set the WNHOME environment variable to point at WordNet's home folder. For orientations on how to install WordNet 3.1, refer to the project's official website.
You must also have a MySQL database installed and running. To run the tool the user must configure the resource file hibernate.cfg.xml under src/main/resources folder, indicating the URL, user and password for the MySQL database where the output data will be saved. The application will generate the schema itself on first run.
Observation: If any warnings concerning foreign key show up, ignore them, starting from the second run they won't happen again (For some reason Hibernate can't decide the right order for creating tables and respect the foreign keys).
<hibernate-configuration>
<session-factory>
<property name="connection.url">jdbc:mysql://simba:3306/supersense_mapping</property>
<property name="connection.username">mapper</property>
<property name="connection.password">1234</property>
<!-- Other parameters omitted -->
</session-factory>
</hibernate-configuration>
The schema must have been created prior to the execution of the application. Don't worry about creating any tables, the application will create them in the first run. It will also add some information necessary for running the application itself. If you wish to check or change any of these data, refer to the .sql scripts inside src/main/resources/ folder.
Maven is used to manage all projects dependencies and to build it. So once the repository has been cloned, running maven clean package should import all necessary dependencies and generate an executable JAR file under the target/ folder. If you wish to inspect the code, import it into Eclipse IDE and don't forget to run a maven update.
Observation: If you need help with preparing your environmen from scratch (installing a Unix OS, installing Java, WordNet and MySQL), check the configuration-help.md file available in the root folder.
Once the environment is prepared and the project has been packaged into an executable JAR, the tool can be run simply by calling:
java -jar target/SemanticMapper-1.0.0-SNAPSHOT.jarUse the flag -h to see specific informations on running options. Alternatively the application can be run directly from Eclipse by executing its Main class, under br.uniriotec.ppgi.mapping.view.Main.
Optionally the user may add the -e parameter when invoking the JAR to activate Evaluation Mode. This way, instead of applying the mapping rule to all synsets (noun, leaf synsets) only samples for each supersense are mapped, along with some intentional wrong mappings for additional synsets, that can be used as control group when evaluating the mapping rules. To this mode simple use the following command:
java -jar target/SemanticMapper-1.0.0-SNAPSHOT.jar -eWhen evaluation mode is activted the application loads additional information from the database prior running the mapping, such as z-value for cnofidence level and confidence interval, which are used when determining the sample sizes for each supersense mapping evaluation.
Once the application has run, the database schema will have been generated, relating the sampled synsets to semantic types. if you wish to run the following step of the mapping evaluation described in my Master's Thesis, create the tables User and Evaluation through the DDL scripts on the file other_tables.sql in the project's root folder (same place as this README file).
The evaluation can be conducted through another application, a PHP webapp also available as a repository at the MappingEvaluation project.
You should be able to have the system up and running, along with your MySQL database, on any operational system with at least 4GB of RAM and 2.0GHz processor. A big part of the application consists of I/O, so an SSD drive is welcome. We recommend using Linux or MacOS, since WordNet 3.1 is not yet available for Windows.
This application is provided under the MIT License (check the license.txt file).
[1]: Ciaramita, Massimiliano, and Mark Johnson. Supersense Tagging of Unknown Nouns in Wordnet. Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2003, 168–75.
[2]: Ciaramita, Massimiliano, and Yasemin Altun. Broad-Coverage Sense Disambiguation and Information Extraction with a Supersense Sequence Tagger. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing, 594–602. EMNLP ’06. Stroudsburg, PA, USA: Association for Computational Linguistics, 2006. http://dl.acm.org/citation.cfm?id=1610075.1610158.
[3]: Dixon, R. M. W. A Semantic Approach to English Grammar. 2nd ed. Oxford University Press, USA, 2005.
[4]: Castro, Lucia, Fernanda Baião, and Giancarlo Guizzardi. A Semantic Oriented Method for Conceptual Data Modeling in ontoUML Based on Linguistic Concepts. In Proceedings of the 30th International Conference on Conceptual Modeling, 486–94. ER’11. Berlin, Heidelberg: Springer-Verlag, 2011. http://dl.acm.org/citation.cfm?id=2075144.2075195.
[5]:Guizzardi, Giancarlo. Ontological Foundations for Structural Conceptual Models. Vol. 15. Fundamental Research Series. The Netherlands: Telematica Insituut, 2005.