Overview

The LOGON infrastructure combines about a dozen or so independently developed resources into an integrated MT environment. In most cases, the individual components (e.g. the LKB, PET, or [incr tsdb()]) function much like they would in isolated installation. However, for better integration of the LOGON tree, in a few cases the default configurations may differ from such isolated installations, and there is some functionality that is only activated in the LOGON universe. This page enumerates such divergences and LOGON-exclusive functionality, and aims to enable users to revert to more familiar defaults if desired.

International Character Support

Since around 2006, several DELPH-IN developers have worked with Franz Inc., the developer of the Common Lisp system behind many of the LOGON pieces, to improve display and input of international characters in the Common Lisp Interface Manage (CLIM). CLIM is the widget library used to build the (original) LKB graphical user interface. Since early 2007, the LOGON tree has been the testbed for patches to improve internationalization in CLIM, as they have been provided incrementally by Franz Inc. Sometime early in 2008, Franz Inc. made these patches official, i.e. they are now part of the official patch distribution (and have since been revised a few more times).

The LOGON run-time binaries are built with the latest CLIM patches (as of late 2008), and both display and input of a wide variety of character sets should be possible in this environment (we have tested and confirmed at least the following: German, Japanese, Korean, and Norwegian). As tends to be the case with internationalization problems, it is important to align the Un*x and Lisp locale (value of $LANG or $LC_ALL, and value of the -locale command line argument to ACL), and to make sure emacs(1) and Lisp agree on the appropriate coding system. The LOGON tree attempts to get all of these parameters right ‘out-of-the-box’, mostly by pushing for UTF-8 as the default encoding across languages. While the non-LOGON LKB binaries (see the LkbInstallation page) are not yet built using the latest set of CLIM patches as of late 2008, it is expected that they too will activate improved international character support (on Linux) in the near future.

Linguistic User Interface

The LOGON tree enables the replacement Linguistic User Interface (LUI) for the LKB by default; see the LkbLui page for further information. However, LUI as of late 2008 only supports the display of trees, AVMs, (single) MRSs, and parsing charts, hence the LOGON infrastructure will often still fall back on the original LKB CLIM widgets. This is specifically the case for the LOGON multi-MRS browser, i.e. the widget that interactively displays the results of transfer. To render individual MRSs in the more compact (and, some say, more user-friendly) LUI browser, the Debug | Simple and Debug | Index commands can be used in the multi-MRS window.

Discriminant Mode

The Redwoods tree comparison and treebanking tools support two distinct modes of discriminant extraction, one syntactic in nature (operating on HPSG derivations), the other semantic (operating on variable-free MRS triples). Unlike the stand-alone LKB and [incr tsdb()] installations, the LOGON tree opts for semantic discriminant mode by default, primarily so as to also support comparison and treebanking of analysis results from non-HPSG systems. The two discriminant modes are (somewhat opaquely) named :classic and :modern, for syntactic and semantic discrimination, respectively. The LOGON default is :modern, and can be changed as follows

  (setf lkb::*tree-discriminants-mode* :classic)

To semi-permanently change the default, a command like the above can be put in the per-user .lkbrc LKB start-up file in the user home directory.

[incr tsdb()] Skeleton Directory

The LOGON tree sets the default [incr tsdb()] skeleton directory to Norwegian, reflecting the primary source language of the original LOGON consortium. To globally change the default skeleton directory, a command like the following can be executed interactive, or placed in the per-user .tsdbrc [incr tsdb()] start-up file:

  (tsdb :skeletons "~/logon/lingo/lkb/src/tsdb/skeletons/english")

However, note that the various LOGON batch processing scripts (see the LogonProcessing page for details) will adjust the skeleton directory appropriately, i.e. reflecting the choice of grammar or language pair.

Language Modeling and MRS Ranking

Web-Accessible Demonstrators

MaxEnt Experimentation

Last update: 2018-01-13 by StephanOepen [edit]