Overview

The LOGON infrastructure (and source tree) is a collection of software, grammars, and other linguistic resources to facilitate experimentation with transfer-based machine translation (MT), or other experimentation with 'deep' parsing or generation. To a large degree, the LOGON tree packages resources that exist independently, specifically the core of the open-source DELPH-IN toolchain and several of the DELPH-IN grammars. These include, among others, the LKB, PET (see the LogonPet page), and [incr tsdb()] software systems, and the LinGO ERG, GG, JaCY, and SRG broad-coverage grammars for English, German, Japanese, and Spanish, respectively. Additionally, the tree includes pre-compiled versions of other packages, for example ChaSen (for Japanese pre-processing), the CMU Language Modeling Toolkit (SLM), FreeLing (Spanish pre-processing), TADM (for MaxEnt experimentation), TnT (for English PoS tagging), and UTool (for MRS manipulation).

The LOGON tree was originally developed by the Norwegian LOGON and HandOn research projects, working on quality-oriented translation from Norwegian to English. For Norwegian analysis, these projects employed (an extended version of) the Oslo-Bergen Tagger (OBT) and the NorGram LFG implementation. Both resources are released under open-source licenses as part of the LOGON tree. However, to actually run the Norwegian–English instantiation of the system (dubbed NoEn), the proprietary XLE LFG system and a commercially licensed bilingual dictionary (dubbed KF) are required, which cannot be part of the freely available LOGON distribution. Please see the LogonExtras page for instructions on how to install proprietary add-ons to the LOGON tree, e.g. for sites holding valid XLE and KF licenses. To see the Norwegian–English LOGON system at work, there is an on-line interface to the MT demonstrator.

In subsequent collaborations between the original LOGON developers and DELPH-IN researchers in Germany, Japan, and the USA, additional language pairs were added. As of late 2008, these include German–English and Japanese–English (and, albeit lesser developed, the inverse directions of translation), as well as a battery of 'baby' MT systems built from a collection of smaller grammars based on the LinGO Grammar Matrix. In a sense, the LOGON tree functions similar to a GNU/Linux distribution: it combines a complex set of individual components, aiming to provide ease of installation and a certain degree of uniformity, inter-operability, and quality assurance. The system is available exclusively for GNU/Linux (on 32-bit or 64-bit x86 architectures). As of November 2008, all system development and distribution is through the SubVersion (SVN) revision management system. Please see the LogonInstallation page for details. Regrettably, only a very limited amount of documentation is available, a property that the LOGON tree shares with a number of the core DELPH-IN resources. The LogonReports page summarizes the documentation misery as of late 2008.

Table of Contents

Following is a list of topics for which at least some documentation exists. Feel free to add additional materials, but please make sure to create adequate wiki names for new pages, typically prefixed with Logon where they pertain to specifics of the LOGON infrastructure.

Background Materials

Further information on the LOGON software and consortium can be found at the project web site; the following publication provides an overview of most of the core pieces:

The first paper discussing the use of Minimal Recursion Semantics in machine translation is:

An example of the extension of the LOGON machinery to a new language pair can be seen in

For additional information, there is an archived mailing list for the LOGON tree. For additional questions, please feel free to contact Stephan Oepen (oe at ifi.uio.no), the technical manager for the original Norwegian LOGON consortium.

LogonTop (last edited 2013-07-30 15:18:41 by StephanOepen)

(The DELPH-IN infrastructure is hosted at the University of Oslo)