-
Notifications
You must be signed in to change notification settings - Fork 4
LogonExtras
The LOGON tree makes a distinction between core components (a meager three gigabytes) and extra resources. Of these, the latter are not required for basic functionality but rather extend the LOGON tree with either proprietary third-party components (which can only be released to sites with appropriate licenses, typically members of the original LOGON consortium and external co-developers) or 'bulky' add-on resources, for example the LinGO Redwoods treebanks. This page provides information on which extra components exist in the LOGON SVN repository, and on how to activate these. Please note that, as of November 2008, the extra mechanism is new and remains to be tested. The information on this page is thus expected to change more frequently than some of the more basic documentation.
All access to the LOGON SVN repository is via the HTTP protocol, typically to the address http://svn.emmtee.net/. The repository makes available the open-source LOGON core for anonymous read-only access. However, some of the add-on components can only be made available to authenticated SVN users, for example members of the LOGON consortium who have obtained licenses for the XLE LFG system or Allegro Common Lisp. Thus access to proprietary LOGON add-ons is regulated by SVN user accounts, where the original CVS accounts (and passwords) of the LOGON core developers team have been imported into the SVN repository. Please contact Stephan Oepen (oe at ifi.uio.no) for all questions relating to SVN user accounts and access rights.
Due to what appears to be a quirk in how SVN handles [http://subversion.tigris.org/issues/show_bug.cgi?id=2712 mixed-access repositories] (i.e. repositories where some content is available to anonymous users, while other parts require user authentication), LOGON co-developers with SVN user accounts should connect through a different URL, viz. http://logon.emmtee.net/. In fact, both URLs (svn.emmtee.net and logon.emmtee.net) connect to the same physical SVN repository, they only differ in how they handle authentication. We hope to be able to consolidate access methods as newer SVN server revisions become available. In the following examples, we will use the world-readable SVN address for add-on resources that are publicly available, and the authenticated URL otherwise. Note however that, for those with valid SVN user accounts, it is always possible to uniformly use logon.emmtee.net as the base SVN URL.
Finally, the repository structure for add-on components reflects the distinction between tagged releases and the on-going trunk of development. Again, in adapting the example SVN commands suggested below, one may have to adapt directory names systematically (i.e. in ways that should be obvious), to reflect the choice of LOGON version currently installed. As of November 2008, we will assume that users install the HandOn release.
The core directory lingo/redwoods/ contains the environment for semi-automated parse selection, realization ranking, and MR re-ranking experiments, using treebanked versions of the LOGON tourism corpora. In the core installation, this directory only contains the scripts and various configuration files for such experimentation, but not the treebanks or corresponding version of the ERG. To download the complete data, execute the following:
cd $LOGONROOT/lingo/redwoods
svn switch http://svn.emmtee.net/extras/handon/lingo/redwoods
The SVN switch command makes a sub-directory of an SVN-controlled tree point to a different module from the same repository (i.e. it does not allow 'mixing and matching' across repositories). Please see the [http://svnbook.red-bean.com/ SVN documentation] for further information.
For several of the LOGON development data sets and the final test set (which only became available after completion of system development), historic incr tsdb() profiles (and fan-out log files) are available. These profiles document (in great detail) development progress made in the course of the original [http://www.emmtee.net/ LOGON] and [http://www.emmtee.net/index.php?page=7 HandOn] projects. To add this data to your LOGON tree, do
cd $LOGONROOT/uio/evolution
svn switch http://svn.emmtee.net/extras/handon/uio/evolution
To analyze the evolution of the LOGON Norwegian–English instantiation, point incr tsdb() to uio/evolution/ as its datatabase directory and experiment with the commands in its Evolution menu.
The LOGON tree includes so-called run-time binaries, precompiled executable versions of the main LOGON software (which comprises the LKB, [incr tsdb()]], and LOGON extensions). These run-time binaries are sufficient for grammar development (including transfer and realization), experimentation with the MT functionality, treebanking, web-accessible on-line demonstrations, and (we believe) MaxEnt experimentation. However, to make modifications to the Lisp (source) code of individual components or compile in additional software, a complete Common Lisp system is required. The LOGON tree is built using [http://www.franz.com/products/allegrocl/ Allegro Common Lisp] (ACL), and this is the only Lisp compiler that supports the complete LOGON functionality (this is mainly due to some parts of the LOGON system making use of Lisp functionality that is not part of the ANSI Common-Lisp standard, i.e. not portable across Lisp implementations).
The directories franz/linux.x86.32/ and franz/linux.x86.64/ provide stubs for the site-specific installation of Allegro Common Lisp. The core versions of these directories only contain the configuration files used by LOGON developers to tune Allegro CL to their needs. For ease of installation, complete ACL packages for 32- and 64-bit Linux are available through SVN to authenticated users whose sites already hold a valid ACL 8.1 license. Use the following command to download the Lisp compiler:
cd $LOGONROOT/franz
svn switch http://logon.emmtee.net/extras/trunk/franz
However, even though we provide the base Lisp files (including all software updates provided by the original vendor) and customized Lisp binaries, you will still need to obtain the license files owned by your site. These files are called devel.lic and are provided by Franz Inc. customer support to the contact person at your site. Depending on your platform (32- vs. 64-bit, or both), copy your own devel.lic into the appropriate directory. For example, assuming a 32-bit installation:
cp devel.lic $LOGONROOT/franz/linux.x86.32
To confirm that your license file is valid and compatible with the LOGON version of ACL (which is version 8.1, as of November 2008), consider running the following test (which should complete without error messages):
cd $LOGONROOT/franz/linux.x86.32
./alisp -I base -e "(excl:exit)"
For the full Norwegian–English system (but not other language pairs), a copy of the proprietary [http://www2.parc.com/isl/groups/nltt/xle/ XLE] software is required. If your site holds a valid license, the following command will agument the core LOGON tree with the XLE software:
cd $LOGONROOT/parc/xle
svn switch http://svn.emmtee.net/extras/handon/parc/xle
The LOGON project licensed Norsk–Engelsk Stor Ordbok from Kunnskapsforlaget for use in the project. For the remaining duration of the LOGON project proper, members of the original consortium can install a copy of the XML files for the dictionary as follows:
cd $LOGONROOT/kf
svn switch http://svn.emmtee.net/extras/handon/kf .
This dictionary was used for the Norwegian–English demonstrator to auto-generate transfer rules for open-class words with predictable properties, mostly nouns and adjectives. The licensing agreement with Kunnskapsforlaget (probably) does not allow public distribution of these derived transfer rules, hence they are not part of the core LOGON tree. To re-generate these transfer rules, once you have successfully installed the dictionary, execute
cd $LOGONROOT/uio/noen
./trag --kf
This process should take a little while (ten to sixty minutes, maybe, depending on the type of machine you have available). It should result in new files kf.a.mtr, kf.n.mtr, and kf.nn.mtr, containing on the order of 15,000 transfer rules.
Home | Forum | Discussions | Events