Downloadable NLG systems
Tools and Software for English - Downloadable NLG systems
For languages other than English, see List of resources by language.
The natural language generation systems listed below are available for download over the web. If you know of a system which is not listed here, please click on Edit in the upper left corner of this page and add the system yourself.
- 1 ASTROGEN
- 2 CLINT
- 3 CRISP
- 4 DAYDREAMER
- 5 FUF/SURGE
- 6 GenI
- 7 Grammar Explorer
- 8 KPML
- 9 LKB
- 10 Multimodal Unification Grammar
- 11 NaturalOWL
- 12 NLGen
- 13 NLGen2
- 14 OpenCCG
- 15 RAGS (Reference Architecture for Generation Systems) software
- 16 SimpleNLG
- 17 SPUD
- 18 STANDUP
- 19 Suregen-2
- 20 SURGE
- 21 SURGE 2.3
- 22 SURG-SP
- 23 SURG-IT
- 24 TG/2
Aggregated deep and Surface naTuRal language GENerator - Prolog based system.
CLINT is a hybrid template / word-based generation system with an example application of business letter generation. The system is written in C++ and runs under Microsoft Windows.
CRISP is Alexander Koller's NLG system that tries to cast both microplanning and sentence realisation as an AI planning problem. The code is a mixture of Java and Scala, a scripting language for the Java virtual machine. CRISP comes with its own implementation of GraphPlan, but it can also output plans in PDDL (“Planning Domain Definition Language”, a successor to STRIPS) for use with other AI planners. License: LGPL.
DAYDREAMER is a computer model of the stream of thought developed at UCLA by Erik T. Mueller from 1983 to 1988. The generator is located in the file dd_gen.cl. Common lisp source code available under GPL v2.
surface realiser for (Feature-Based Lexicalised) Tree Adjoining Grammar and a flat MRS-like semantics (sans top handle and underspecification). Toy example grammars provided for English and French. Largish core grammar for French is under development (contact us for details). GPL, known to work under Linux and Mac OS X (potential for making it work on Windows as well). Written in Haskell. Source code avalailable via darcs.
http://www.fb10.uni-bremen.de/anglistik/langpro/kpml/tutorials/Grexplorer/grexplorer.html (old site unavailable as April, 2011)
provides a means of exploring large-scale systemic-functional grammars in order to see how they are organized and what kinds of things they cover. It can be used to explore the KPML resources. Downloadable standalone executables of the grammar explorer are available for Windows 95/98/NT. These already include a version of the Nigel grammar of English and pre-installed examples.
The KPML system offers a robust, mature platform for large-scale grammar engineering that is particularly oriented to multilingual grammar development and generation. It is particularly targetted at providing resources for realistic but broad-coverage generation applications, where both flexibility of expression and speed of generation are at issue—for example in online webpage generation or spoken dialogue. KPML is also used extensively in multilingual text generation research and for teaching. It is based on systemic functional linguistics.
The KPML system was a direct descendent of the Penman text generation system, as developed further multilingually in cooperative work between the Komet (http://www.darmstadt.gmd.de/publish/komet/index.html) project in Darmstadt and the Systemic Modelling Group at Macquarie University. Downloadable standalone executables of the system are available for PCs running Windows. The source code is written in ANSI Common Lisp and uses the Common Lisp Interface Manager (CLIM). The system has been compiled and tested[ under Franz Allegro Common Lisp (4.2, 4.3, 4.3.1, 5.0, 6.0, 7.0) for Unix and Franz Allegro Common Lisp 3.0 and Harlequin Lispworks 4.0, 4.1, 4.2 for Windows. It is possible to use the system without the window interface as a generator serving requests for generation across sockets or via files.
A growing set of generation grammars are under development for a variety of languages, inlcluding English, Spanish, Dutch, Chinese, German, Czech, and more. See the Generation Bank (http://www.fb10.uni-bremen.de/anglistik/langpro/kpml/genbank/generation-bank.html ) for current examples. The development of further languages and of extensions to existing resources are very welcome!
LKB (Linguistic Knowledge Builder) is a grammar engineering environment for unification-based formalisms, typically HPSG. It includes a realiser that takes as input Minimal Recursion Semantics (MRS). LKB is implemented in Common Lisp, and is freely available under an open source license. It includes also a KNOPPIX-based GNU/Linux live-CD, with all the system installed, ready to use.
Multimodal Unification Grammar
MUG Workbench is a development and debugging tool for Multimodal NLG. The grammar formalism supported is Multimodal Functional Unification Grammar (MUG). The MUG system runs MUG grammars with fixed (test cases) and arbitrary input specifications to produce output in a natural language, graphical user interface and possibly in other modes. It is designed to do three things: - Multimodal Fission (distributing output to interaction/communication modes) - Some sentence planning (chosing information to include in the utterance) - Natural Language and graphical user interface realization (producing some form of output) The MUG system does these three jobs in parallel. MUG Workbench can serve to inspect the data-structures used during generation. It should help you to learn more about the nature of unification grammars used for parsing or natural language generation. Furthermore, the MUG Workbench is helpful in debugging your grammars.
http://www.aueb.gr/users/ion/software/NaturalOWL1.1.tar.gz NaturalOWL (version 1.1)
Generates descriptions of entities and classes from OWL ontologies that have been annotated with linguistic and user modeling resources expressed in RDF. Currently supports English and Greek. Extensions for other languages welcome. NaturalOWL can also be used as a Protégé plug-in. See here for publications describing NaturalOWL. (GPL)
The NLGen natural language generation system applies the SegSim strategy for generating English sentences. Probabilistic inference for sentence construction is based on a statistical analysis of RelEx output. Not to be confused with NLGen2, below, which uses a different sentence generation theory. Java, Apache license. See demo: Demo of AI Virtual Pet Answering Simple Questions.
The NLGen2 natural language generation system uses RelEx dependency parses, together with Link Grammar linkage analysis to generate English-language output. Not to be confused with NLGen, above, which uses a different sentence generation theory. Java, Apache license. Reference: Blake Lemoine, "NLGen2: A Linguistically Plausible, General Purpose Natural Language Generation System".
OpenCCG, the OpenNLP CCG Library (formerly Grok), is both a parser and a realizer for Combinatory Categorial Grammar. It has been used in several dialog systems. The realizer has been enhanced with n-gram models and a supertagging approach called hypertagging. OpenCCG is implemented in Java, and is freely available under the LGPL.
RAGS (Reference Architecture for Generation Systems) software
Deliverables from the RAGS project - RAGSOCKS software for interfacing modules using RAGS data representations, example RAGS module (genetic algorithm based text planner) and RAGS wrapper for FUF/SURGE.
is an ultra-simple Java-based realiser. Its grammatical coverage and syntactic knowledge is minuscule compared to KPML or FUF/SURGE. However, because it is so simple, its relatively easy for people to learn how to use it. It has been used by many people in Aberdeen, and also for teaching. It is set up as a Java package, so it can only be used by Java programs.
SPUD (Sentence Planner Using Descriptions) is Matthew Purver's LTAG-based NLG system. There are two versions: SPUD version 0.01 was written in SML. Later versions, known as SPUD lite, are written in Prolog. The small codebase of SPUD lite makes it ideal for teaching, but it is also used in dialog system prototypes.
The STANDUP project (System To Augment Non-speakers' Dialogue Using Puns) is a collaborative project on generating simple jokes from a graphical user interface appropriate for non-speaking children. The project began in October 2003 and ran until March 2007. The software was written in Java and is available for Windows and Linux, including source code and database files.
Suregen is “a hybrid, ontology based and NLG-oriented formalism for generating text for documents in clinical medicine.” The system Suregen-2 is written in (Allegro) Common Lisp. A demo system which runs under Windows is available for download. A screencast video shows data being entered into computer forms using mouse and keyboard while a feedback text is continually updated and shown below. (Try playing the AVI file in VLC if you run into problems.) Perhaps this system could be considered an instance of the WYSIWYM approach.
Syntactic realization package. (A CommonLisp package providing an interpreter for a functional unification formalism called FUF and SURGE, a large grammar of English written in FUF.) Offers download of SURGE 2.2.
The latest version of Surge, including support for written dialogue, and expanded syntactic coverage based on the Penn TreeBank.
Systemic Unification Reusable Grammar for Spanish is a large scale Spanish grammar allowing systems which already use FUF/SURGE for English NLG to be able to generate syntactically (and many times semantically) equivalent text in Spanish when new lexical items are introduced. SURG-SP makes use of inputs almost identical to the English version Surge 2.3.
The Italian version of Surge 2.3.
is a shallow verbalizer that can be quickly accustomed to new domains and tasks. It combines context-free grammars with templates and canned text in a single formalism. Thus the granularity of the language model may depend on the application needs. The system currently runs under Solaris 2.5. It is available freely under a research license.
| This page was imported semi-automatically from the NLG Resources Wiki which was run by ACL SIGGEN in the years 2005–2009. Please correct conversion errors and help update its contents.|
Now this page is associated with the Natural Language Generation Portal.