Japec is a JAPE-to-Java compiler, packaged as a processing resource for GATE 3.1+. The processing resource (Ontotext Japec Transducer) is designed as a replacement for the standard JAPE Transducer, provided for optimized performance. The implementation is based on finite state machines and the compiler uses the standard algorithms for determination and minimization in order to achieve better performance. The actual compiler is a standalone executable written in Haskell because it involves complicated algorithms with dynamic data structures. The processing resource wraps the compiler and translates the grammar under the hood.

Japec is proven to be 2 to 5 times faster than the standard JAPE Transducer. The code of Japec is donated to the GATE project and community under the LGPL licence. The GATE team have adopted Japec internally, and are now working with Ontotext to port Japec to Java for release 4 of the system.

Ontotext took part in the development of GATE after release 2.0 with number of tasks, among which:

• GATE's Oracle support - work on the production version of the GATE's support for ORACLE data stores. The work included analysis and redesign of the relational schema, implementation and optimization of the supporting code, installation and testing. The scalability requirements related with the hosting of huge corpora such as BNC turned this task into a challenging VLDB problem.
• Optimizations - performance analysis and optimization of GATE, concentrating on the IE pipeline. The result of couple of manmonths work was over two times speed up and memory reduction (without considering the HashGazetteer, see below.)
• WordNet API - integration of world most popular lexical database. WordNet is currently available within GATE both via Java API (similar to the original one for C) as well as through GUI allowing the standard queries to it;
• Protege-2000 integration - the most popular ontology editor can now be run within the GATE development enviornment;
• Ontology access, editing, and markup - support for editing viewing and using ontologies. Basic features: DAML+OIL storage, access via API, ontology-based feature-map matching allowing grammar rules taking benefit from hierarchical annotation and look-up types;
• Information Retrieval - fully-featured Information Retrieval (IR) subsystem that allows full-text queries to be performed against GATE corpora. The current implementation is based on the most popular open source full-text search engine - Lucene;

Most of the above mentioned enhancements are part of the GATE distribution and come free. There are also number of OntoText proprietary tools and applications which are not free for commercial use. However, those are still free for reseach or education purposes.

A typical gazetteer - Java-implemented lookup tool that allows occurrences of strings from a predefined lists to be found in texts. The ciritical issues about such routines are the speed and memory usage. A classical implementation approach is making a finite state machine (FSM) recogniser or a kind of suffix-tree.

The OntoText's Hash Gazetteer work is based on hashtables instead of FSM. On average it takes four times less memory and works three times faster than an optimized FSM implementation! It is currently available for free download (for non-commercial use) as a CREOLE component to be used within the GATE framework (see above).

A stohastic module capable in filtering annotations, disambiguation, and other "soft" tasks based on confidence measures. It is based on OntoText's proprietary HMM implementation tunned for Information Extraction applications - see [Scheffer et all.2002] for the most general ideas. Contact us at info-at-ontotext.com for more information and applications.