PLIS - a Probabilistic Lexical Inference System
The PLIS download package contains the following folders:
- PLIS_Trained_Models - weights used by the lexical_inference project.
- resources - some lexical knowledge resources which you can use in PLIS.
- workspace - source code of PLIS and related libraries:
- Biu-Nlp-Lab - source code developed in Bar Ilan University NLP lab, including the code of PLIS.
- Excitement-Open-Platrofm - some infrastructural classes from the EXCITEMENT project.
The PLIS, BIU and EXCITEMENT code is available under the GPL version 3. The resources are available under various licenses. Please read and agree to the license of each resource.
To install and run PLIS you will need to have:
- 1 GB free hard drive (preferrably more, if you want to add knowledge resources)
- 4 GB memory
- Java Standard Edition 1.7.
- An environment variable DATA that points to the "resources" folder.
Install the source code:
Assuming that you use the Eclipse IDE:
- Install the Maven2Eclipse (M2E) plugin, from this repository: http://download.eclipse.org/technology/m2e/releases (copy this link and paste in Help-->Install New Software-->Work with>).
- In Eclipse, go to File->Import->Maven->Existing Maven Project, select the "workspace" folder, and import all projects.
- Make sure your workspace contains the following projects:
common, core, Excitement-Open-Platform, factories, infrastructure_remains, lap, lexical_inference, lexical_integrator, parse.
- For each project except Excitement-Open-Platform, open it in the Package Explorer, right-click "JRE System Library", select "Properties",
and change the "Execution environment" to Java SE 1.7 (this is a workaround for a bug in M2E plugin).
- If there are compilation errors, try Project -> Clean -> Clean all projects.
The lexical_integrator package
Here you will find the code that integrates various input lexical resources in a lexical graph data structure (notes are terms and edges are inference rules).
Paths can be extracted from this graph to deduce lexical relations between source and target terms.
The entry point to this package is LexicalResourcesIntegrator which has a main method that demonstrates a common usage:
- Go to the class LexicalResourcesIntegrator.
- Run as -> Java application. You should see several paths that link a source word to a target word.
- A configuration file defines the system configuration for this example. It can be easily edited.
For instance, choosing which lexical resources to use as input is done through the parameter "lexical resources".
The file location is: src/main/java/configuration/lexResIntegr_config.xml
The lexical_inference package
This package adds the layer of Probabilistic Lexical Model (PLM)
on top of the Lexical Integrator. This project include the various
models developed to estimate that reliability level of input lexical
resources and to estimate the entailment probability of each rule step,
inference chain and full Text to Hypothesis inference.
Using this package spare the user the need to handle the lower level
of the Lexical integrator. This package encapsulates LexicalResourcesIntegrator and the
user just use its API.
The important (abstract) class of this package is PLM (Probabilistic Lexical Model). It implements the inference process (including learning).
The provided probabilistic models, as well as any new model the user would like to add, should extend this class.
A suggested entry point for new users is the examples with detailed explanations (under ac.biu.nlp.demo.examples):
- Go to the class PLMfullSentencesExample.
- Run as -> Java application. You should see several lexical inference chains with probability estimations.
- The configuration file location for this example is: src/main/java/configuration/lexResInferExampleCode_config_local.xml
Install additional resources
There are several MySQL-based lexical resources, which you can install using the following links:
|Schema Name||Configuration Module||MySQL Schema||File Size|
|BAP (Directional Similarity)||DIRECT||Download||111 MB|
|Lin Similarity||DistSim||Download||236 MB|
|Wikipedia Knowledge Resource||Wikipedia||Download||214 MB|
After you download the Schema, use MySQL WorkBench
→ Server Administration → Data Import to import them to your databases.
Then change the relevant configuration modules in the configuration file, and insert the correct
MySQL connection string.
The PLIS system and its probabilitsic models are explained in the following papers:
- Eyal Shnarch, Erel Segal-haLevi, Jacob Goldberger, Ido Dagan. PLIS: a Probabilistic Lexical Inference System. ACL (Demonstration), 2013.
- Eyal Shnarch, Ido Dagan, Jacob Goldberger. A Probabilistic Lexical Model for Ranking Textual Inferences. *SEM, 2012.
- Eyal Shnarch, Jacob Goldberger, Ido Dagan. Towards a Probabilistic Model for Lexical Entailment. TextInfer, 2011.
- Eyal Shnarch, Jacob Goldberger, Ido Dagan. A Probabilistic Modeling Framework for Lexical Entailment. ACL, 2011.