Conversion of an existing termbank: EAGLET to MARTIF

  The EAGLET database (see 2.4.4) is available in the form of an ASCII-file, containing all information on every single term in the database.

Table 3.2 shows a possible model of a single, one lined database-entry, also indicating the structure used within the individual data categories. It is printed as is, including all signs that are present in the ASCII source of the EAGLET data-base; for printing purposes only line breaks were added. The real entry does not have any line breaks or carriage returns, but every single term entry consists of one and only one line. The line breaks do not represent spaces here. The data-categories are limited by the comma-character `,', the escape character is the backslash `'. 20 separate columns are defined for the data-base table. Instances of the data categories are represented by expressions referring to the name of the relevant data-categories, for some categories with some generally included markup and for others additional numbers, at least in those cases where the rule of elementarity of data categories is violated. In these latter cases one or two more elements are included as they occur in the database (except of definitions, where three is currently the maximum number).

 

A conversion of such a complex database entry is not easy but it is possible to manage it with the help of UNIX tools . The actual script is given in appendix C. For a UNIX script it is fairly complex; most scripts are written only with few commandlines as easy and fast written tools for computer experts needing tools for otherwise annoying and repeatedly appearing tasks. This script consists of some hundred lines of source code, most of them containing substrings of subroutines.

The most important criterion for this program was that it has to work on UNIX machines, has to be capable of handling great amounts of data and has to be able to be changed quickly. Consequently, programming languages needing compilers such as C and C++ did not match the task. UNIX tools were easy to change in a few seconds, easy to debug, and what is even more important, if changes in the database itself take place, such as the introduction of new or/and other categories, these scripts can easily be adapted to new needs.

Data loss is a serious problem in database conversion, as was previously pointed out on page . To evaluate the efficiency of the script developed for the EAGLET to MARTIF conversion the following tables 3.3 up to 3.7 show the degree of equivalence of the EAGLET data category in comparison to the MARTIF standard with a description of the used tool, comments on possible data loss and some comments.

 

Table 3.3: Equivalence of Database formats EAGLET and MARTIF 

EAGLET-data-category	Individual Format	Script for MARTIF-conversion	Loss of information	comment
Orthography	ASCII-format: an orthographic representation of a term	awk script	none
Sorter	ASCII-format: the same orthographic representation in only small characters	none	none
Pronunciation	SAMPA transcription preceded and ended by slash `/'	awk script	none
Part of Speech	abbreviated POS; also including POS analysis of term components (for compounds or multi word terms)	awk script	none but	violation of elementarity

 

Table 3.7: Equivalence of Database formats EAGLET and MARTIF (continued) 

EAGLET-data-category	Individual Format	Script for MARTIF-conversion	Loss of information	comment
Inflections	inflectional ending for plurals (if any) or absent singular	awk script	none
Domain	Name of the domain, subdomain separated by colon	awk script	none
Hyperonyms	orthographic representation of hyperonym (sometimes multiple)	awk script with subprocess extracting multiple hyperonyms	none	violated principle of elementarity solved by subprocesses
Hyponyms	orthographic representation of hyponym (sometimes multiple)	awk script with subprocess extracting multiple hyponyms	none	violated principle of elementarity solved by subprocesses
Synonyms	orthographic representation of synonym (sometimes multiple)	awk script with subprocess extracting multiple synonyms	none	violated principle of elementarity solved by subprocesses

 

Table: Equivalence of Database formats EAGLET and MARTIF (continued) 

EAGLET-data category	Individual Format	Script for MARTIF-conversion	Loss of information	comment
Antonyms	orthographic representation of antonym (sometimes multiple)	awk script with subprocess extracting multiple antonyms	none	violated principle of elementarity solved by subprocesses
Meronymic superordinate	orthographic representation of meronymic superordinates (sometimes multiple)	awk script with subprocess extracting multiple meronymic superordinates	none	violated principle of elementarity solved by subprocesses
Meronymic subordinate	orthographic representation of meronymic subordinates (sometimes multiple)	awk script with subprocess extracting multiple meronymic subordinates	none	violated principle of elementarity solved by subprocesses

 

Table: Equivalence of Database formats EAGLET and MARTIF (continued) 

EAGLET-data-category	Individual Format	Script for MARTIF-conversion	Loss of information	comment
Definitions	definition, source in parenthesis	awk script with subprocess creating a separate source category with a link to the source definition	undefined	not uniquely structured multiple definitions with sources result in manually post-editing of 10 to 20 definitions in EAGLET
Examples	Short description of examples in text form	awk script	none
Graphic models	Link to external graphic-files	none	none	currently not available for any term
Audio models	Link to external audio-files	none	none	currently not available for any term

 

Table: Equivalence of Database formats EAGLET and MARTIF (continued) 

EAGLET-data category	Individual Format	Script for MARTIF-conversion	Loss of information	comment
Formulas	Link to external graphic-files	none	none	currently not available for any term
Reference	List of URL's	none	entirely	because of outdated references not included
Date	List of dates	awk script with subprocess extracting multiple date records	none	the relation of name and date (which is latent in the source) is not represented
Author	List of abbreviated author names	awk script with subprocess extracting multiple author records	none	the relation of name and date (which is latent in the source) is not represented

Thorsten Trippel
Fri May 21 13:04:11 MET DST 1999