JarCastingDocument Analysis Platform
What it is:
The Document-Analysis Platform, or DAP, is a programming platform for integrating several NLP tools, making them:
- interact with each other, and
- conform to the same interface.
DAP is a lightweight, simple and easy-to-use alternative to UIMA. While UIMA is a revolutionary and strong platform, it suffers from significant drawbacks, which turned into high barriers for new-comers.
The need for a simple, easy-to-learn and easy-to-use alternative, which preserves only the core ideas of UIMA, is the motivation behind DAP development.
The advantages of DAP over UIMA are:
- UIMA takes several weeks to learn, and requires reading of hundreds of user-manuals pages. Getting started with DAP takes no longer than 5-10 minutes. Learning DAP 100% A-to-Z takes only 20 minutes.
- UIMA requires long and hard-to-maintain XML files. DAP requires nothing but pure-Java programming.
- UIMA employs unusual paradigms for exception throwing, logging, constructing objects, etc. DAP follows normal Java conventions.
The core idea
NLP tools tend to depend on each other. Part-of-speech taggers operate over tokenized texts. Syntactic parsers operate over part-of-speech annotations. Coreference-resolvers operate over syntactic analyses. etc. In short, higher level tools rely on the output of lower-level ones.
This brings up the challenge of integration. Both the syntactic-parser and the part-of-speech tagger should agree on the data-structures and the format of a POS-tagged text. In other words, the POS-tagger output should be what the syntactic-parser expects. This requirement applies to every set of tools with dependencies between them.
Moreover, if all POS-taggers conform to the same format, then replacing one tagger by another is transparent to the syntactic-parser. Similarly, if all the parsers conform to the same format, then replacing one parser by another is transparent to the coreference-resolver.
The goal of DAP is to target this integration challenge. DAP provides data-structures with characteristics and utilities that make them fit for virtually every standard NLP tool. The main two data-structures are document and annotation. The output of every NLP tool can be stored as annotations in documents, with features, attributes, and inter-annotation relations.
In addition to data-structures, an actual set of part-of-speech tags, syntactic phrases types, syntactic-dependency-relations, etc. is required. The project DAP-DKPro_1_8 provides a standard set of NLP types, borrowing them from the DKPro project.
Batteries included
Users can start working with DAP right-away with dozens of state-of-the-art NLP tools for several languages, by using the DAP-DKPro_1_8 library, which wraps DKPro tools inside DAP.
A demo is provided in DAP-DKPro_1_8-demo.
Usage in Maven
The project has been uploaded to Maven central repository.
In a Maven project, add the following:
<dependency>
<groupId>com.github.document-analysis</groupId>
<artifactId>dap</artifactId>
<version>0.1.1</version>
</dependency>
To get started, related projects should be imported as well. See:
Your first steps
Start by reading the 20-minutes-tutorial.
Then jump to the demo.
License
DAP is licensed under Apache 2.0 license, which is a permissive license that is good also for commercial use.
Note that DAP-DKPro_1_8-demo depends on external libraries, which have more restrictive licenses.