index.html

Computing Semantic Representations with Referent Systems

Version 6.1 Date: Tuesday, July 12, 2016

Referent systems are designed to allow for calculation of semantic representations bottom up. This project page is about referent systems. It consists of two parts:

The theory, contained in the manuscript Agreement Morphology, Argument Structure, and Syntax (8th Revision).
The implementation: Archive. Most recent Version: Version 6.1.

The manuscript also contains an introduction into the system so that one can effectively work hand in hand with both theory and implementation.

Source Code
Installation
Tk-Interface
Changelog
Archive
To Do
Foreign Language Support
Other Platforms
Acknowledgments

Source Code

(The latest release.) The version 6.1 comes with source code. You may use it subject to GNU license terms. Use and modification of this software is permitted. We decline any responsibility connected with the use of this software. Notice also that what we describe below will work under Unix only (which includes Mac OS X and Linux). If you insist on using Windows, you should probably install Cygwin first. (Read also the howto manual linked to below).

Back to Top

Installation

If you want a standalone version to play with, here it is. The minimal version requires

Operating System: Linux, Unix or Mac OS X or newer,
LaTeX,
Tcl/Tk
OCaml (current version is 4.01.0, but older ones probably do just as fine). Before installing, make sure you have Tcl/Tk installed so that the graphics works as well. In addition, the following modules should be installed.
- findlib
- Xml-light
- Camomile 0.7.1. Camomile is an extension module for OCaml for locales.
These modules or packages can be obtained from Inria.

To install the program, download the file archiv/referent_v6-0.tar. It contains the following items:

The Manual "howto.pdf". This does not explain the theory, but it tells you about the installation and the use of the program. It is more comprehensive than this page, for example.
The ocaml source files.
The shell scripts. Invoke the scripts with the option "-h" to get some basic help. These are mainly the following:
- hilfe, to get fast help,
- setup, to set the system up (default language is German, so better call with option "-h" beforehand).
- compile, to compile the program.
- ref to initiate the user interface of the program.
- wrap, to be used to create an archive.
- save to create backups.
Some dictionaries, among which a dictionary of Hungarian nominal inflection and the Latin verbs and nouns, which are pretty comprehensive.

Put the archive into a directory where you want to install the entire system, say, <RefSys> . Unpack the archive using

  tar xvf referent_v6-1.tar

You should have directories dict and bin. dict is where the dictionaries have to be put, bin is for the binaries. To get the English installation type

  setup -p -l en

This makes the files in bin executable. Setup also adds directories, and adds <RefSys>/bin to your path (via the option "-p"). Option "-l en" makes the system use English. (If you do not set your paths, you will have to prepend each command by ./bin/ !) It is best to add this line to your .bashrc, otherwise you will have to retype it everytime you restart the computer. Now type

  compile -r

This will compile the system. The errors (if we made any ...) are redirected into compile.log. You are told whether the installation is successful. If not, take a look at compile.log to see what went wrong. Finally, type the magic incantation:

  ref

Back to Top

Tk-Interface

RefSys uses a graphical interface. This should be easy to use, but we am sure there is lot of room for improvement. There is a dialog session. In the bottommost white window you can type commands, the most useful of which is "help". No need to use quotation marks. Hit return and wait for the next thing to happen.

Back to Top

Changelog

Version 3

The main change is in the modularisation of the program. There are also intrinsic changes. The new version separates the syntax and the semantics in the initial parsing stage. During parsing, it only computes the set of viable parse terms. When it is finished, it unravels the viable parse terms into real entries and shows them to the user. Another change is that it now allows for polyadic merge, so it can handle complex predicates correctly.

Version 4

Apart from bug fixes, the new version creates an interactive web page, where users can upload dictionaries and enter sentences. The upload is done in several stages. For security reasons the system will check whether the file looks like a dictionary. If it does not comply with the rules give in the manual, it will not upload. After uploading it will create an executable (if possible) and rehash the pages. A new page is created where one can enter strings in the new dictionary. Entering strings is done by clicking on the items (to avoid awkward keyboard issues for foreign languages).

Back to Top

Version 5.0

The most visible change is the increased support for morphology. Entries have a morphology; this is a morpheme, where a morpheme is a set of morphs. Morphs in turn are complex structures, allowing the use of several strings and features for strings. This allows to treat plural in English, for example, using one entry only, so the semantics need not be iterated. The morphological decomposition tables used in the previous version are now redundant (they were also a source of exponential slowdown).

The Tcl-script is much simplified.

Version 5.1

It is not possible to input non-ASCII symbols with the standalone system using a standard keyboard. The input is through specified sequences, like html, but the coding is flexible.

The exponents are now arrays of so-called glued strings. These are strings with optional conditions what kinds of strings they can be concatenated with. The conditions are of the form "concatenates with a string that has/does not have a suffix suf if appended (prefix pref if prepended)". Variables are now pairs (string, number) and most string output uses buffers, to speed up the output.

Version 5.2

This version is a drastic change from 5.1. There is no standalone version any more, as the graphical interface has become much easier to use. It uses a stack where things can be put and manipulated, and a user dialog box where commands can be entered. Also, we have become frustrated with the dynamic linking tool of OCaml and have not been able to get it to work. Finally, we decided to use an xml-style data format and wrote our own tools to parse from such data and write to it. The web versions are at present not supported and the documentation lags behind seriously.

Version 5.4

New features are:

Data is strictly Xml, and we use xml-light to parse it.
The dictionaries may contain also empty strings, but to enhance efficiency we have added a rank function. Every morph has an optional rank. Adding empty strings must increase the rank. By default, adding a nonempty morph creates a string of lowest rank, adding an empty morph creates a string of maximal rank.
The compile routine creates .ocamlinit automatically. Thus calling ocaml will start the system directly.
There are method for adding dictionary entries using a graphical interface.

Version 5.7

New features are:

The display methods have been overhauled. The command "show all" allows to view all items that have been stored in various tables. They can be clicked to view.
The graphical entry creator has been simplified.
There is a dictionary of Hungarian nouns. It is not complete in the sense that some suffixes are missing, but it demonstrates well the power of the system.

Version 5.8/5.9

Bugs have been fixed (especiall the substitution algorithm).
Many nontrivial dictionaries have been added. Commentary is included.

Version 6.0

Bugs have been fixed. The algorithm now works fully according to theory, with no workarounds.
The handbook explains the algorithm in full detail, everything is being covered.
Many nontrivial dictionaries have been added. Commentary is included.

Version 6.1

Bugs have been fixed. In particular, there were problems in the treatment of fusion with unsaturated arguments as well as the dictionary entries and morphnames.
A partial dictionary for Latin adjectives and nouns has been added that needs to be updated. Commentary is included.

To Do

(Theory:) Add full support for logical connectives.
Improve the messaging system.
Help menus must still be updated.

Foreign Language Support

Interface support is not available other than for English and German. For making the dictionaries notice that although OCaml can only handle ISO-Latin-1, you may define strings for your language in other charsets, too. Tk displays them (it is fully Unicode compatible). LaTeX is a bit trickier, since the typewriter font is somewhat incomplete. Apart from that, the way it works is as follows. For the strings in the language, write them into the dictionary in UTF-8. They are passed by OCaml to Tk and LaTeX, which render them assuming UTF-8 encoding throughout. To input into the Tk-Interface we have built a small converter that allows you to use a standard keyboard to enter foreign symbols (a bit like HTML character codes).

Back to Top

Other Platforms

The software has been successfully installed both on Unix/Linux and Mac OS X. For that you can use the compile installer. For Windows, the only way is to install Cygwin before installing everything else. We will need to find out exactly how this can be done. (If you have any ideas, please let me know.)

Back to Top

Acknowledgments

Referent systems are due to Kees Vermeulen. We am indebted to him as well as Albert Visser for the theory part. The implementation has been done by myself. Its creation has been sponsored by two successive senate grants from UCLA. We have enjoyed the help of Cory Hill, Ben Keil, Adam Skory and Joseph Vaughan. The project has been joined in 2009 by Udo Klein and between November 2010 and September 2011 by Sabine Gründer. They were supported by the Alfried Krupp von Bohlen und Halbach-Stiftung.

support

Send any reports of error (or praise) to Marcus Kracht.

Back to Top