For almost 14 years in the Language Engineering Group we have worked on a wide variety of Natural Language Processing (NLP) problems, being one of the earliest in the creation and operation of onomasiological dictionaries. During that time we have focused on search engine dictionary improvement, but recently our aim has been a development methodology for creating specialized onomasiological dictionaries in a semi-automatic way. To automate the creation of onomasiological dictionaries necessarily implies the automatic execution of used processes to populate the dictionaries knowledge base. Due to the nature of these dictionaries, the definitions that must be included in the knowledge base are both normative and colloquial. In this paper we present a proposal for semi-automatically populating the knowledge base of these dictionaries.
An onomasiological dictionary is a dictionary that works in
back to front way from “regular” or semasiological
dictionaries. In onomasiological dictionaries users already know
the definition of a term, but they do not know or have
forgotten the name for that concept (this last problem is
commonly known as having a word on the tip of the tongue)
Onomasiological dictionaries have been classified into visual dictionaries, reverse dictionaries, thesaurus and synonym dictionaries. These dictionaries were created in order to solve the tip-of-the-tongue problem, but people still have difficulty using them because they require either that the user knows the precise words to describe the term, or its classification (i.e. when using a reverse dictionary to find the word potato, you might have to know that a potato is a tuber, and that tubers are a kind of plant). With visual dictionaries there is also the problem that not every concept has a visual image to represent it. For these reasons it has been suggested that free-text searcher ―also known as Natural Language searching― is a viable option for solving this problem (Lancaster, 1972) since they allow the user to describe their idea of the concept in the way they would use to explain it to another human.
The creation of onomasiological dictionaries that solve
inputs written in natural language improves the user
experience, but it creates some major challenges that the
developers must handle
In short, natural language onomasiological dictionaries
need a rich knowledge base which includes not only formal,
but also informal definitions. Knowledge bases can be
obtained from ontologies, like in the projects Genoma KB
(
DEBO is the first onomasiological dictionary developed in the Language Engineering Group and it works with user queries given in natural language. DEBO is a specialized dictionary and it was originally made as a dictionary of Natural Disasters, but today its structure and search engine has been extrapolated to other areas such as Linguistics, Metrology, Veterinary, and Sexuality. 2.1
The dictionary works with a search engine developed by
• A set of groups of words called paradigms, which are groups of words with similar meaning either in area of specialization or in regular speech.
(1) The system receives the query of the user as an input. 2.2
The search engine performance (2) The system analyzes de input and extracts its keywords by filtrering them with the aid of the stop list. (3) The system searches among the paradigms the ones to which each keyword of the input corresponds. (4) The system searches for terms that coincide in at least one paradigm with the input’s one. (5) The system retrieves the terms ordered by the number of paradigms that each term has in common with the input ―in case of a tie, the system ranks each term according to the order in which the paradigms are presented in the definition against the input―. The terms are divided in “very probable”, “probable” and “not too probable” columns.
For example, suppose someone enters as an input of the dictionary “someone who hates gays”, and in the knowledge base there is the definition “homophobic: a person who despises homosexuals”, and in the knowledge base there are also the following three paradigms.
someone person people Individual dude
hate loathe contemn despise abhor
gay homosexual lesbian queer dyke Fig. 2. Example of paradigms in the knowledge base of an onomasiological dictionary of sexuality.
The user’s definition is related to paradigms 1, 2 and 3, while one of the definitions of the term homophobic is related to the same paradigms in exactly the same order, which means that the term homophobic will be on the top of the output for this query.
This dictionary had an added difficulty since it also had to be able to handle double-meaning words and phrases that are very commonly used in Mexico when talking about sex. In order to cope with this additional component, the dictionary’s paradigms were extended to include double meaning words and even pejorative terms (see Paradigm 3 on Fig. 2), taking into account not only formal synonyms but also colloquial equivalents. In total there were over 33,000 different words organized in over 25,000 paradigms.
Fig. 3. Example of an output of the Onomasiological Dictionary of Sexuality for Mexican Spanish (DOS-MX) There was an experiment to test the precision of the DOS-MX. This experiment consisted on making students write definitions of sexuality terms and to give their definitions to another student who wouldn’t know which terms were described and would try to guess.
The precision of the dictionary was 71% when tested with natural language entries from users that weren’t involved in the development of the dictionary, which is not bad compared to other non-English onomasiological dictionaries such as the one of El-Kahlout et al (2004) which has a precision of 66% in similar tests. However, a vast opportunity to improve exists. 2.3
After the experience of the sexuality dictionary, it was concluded that the use of paradigms is not enough to try and cover all the ways in which a person can describe a concept. It was clear that there is a need to obtain many more different definitions in order to have a wide variety of expressions and ideas for every concept.
But increasing the number of definitions will also tend to increase the number of options from which the dictionary will have to choose, which is why there is also a need for organizing the definitions and terms into some sort of categories that will facilitate the selection of the correct terms.
The main problem then is to find a way to obtain a large number of definitions for the terms and classify them. This should be done in an automatized way, because by doing it manually will take too long and imply high resource usage. 3
ECODE is a program that was developed in the
Language Engineering Group with the objective of
automatically detecting definitional contexts from specialized texts
(
According to
These patterns in Spanish can be punctuation marks, such as comas, colons and parenthesis; verbs like definir (to define) or significar (to mean); discourse markers similar to en otras palabras (in other words), o sea (that is); and even pragmatic patterns like en este conexto (in this context) or en términos generales (in general terms). For example: Desde el punto de vista de la sexología, se puede definir una relación sexual como el acto en el que dos personas mantienen contacto físico con el objeto de dar y/o recibir placer sexual, o con fines reproductivos. (From a sexology point of view, a sexual intercourse can be defined as the act in which two people have physical contact with the objective of giving and/or getting sexual pleasure or with reproductive purposes) The following features can be obtained from this example:
Definition: “acto en el que dos personas mantienen contacto físico con el objeto de dar y/o recibir placer sexual, o con fines reproductivos” (act in which two people have physical contact with the objective of giving and/or getting sexual pleasure, or with reproductive purposes).
(can be defined as).
de vista de la sexología” (From a sexology point of
view)
In order to automatically detect the features or components
of a definitional context,
Simple verbal definitional patterns
Compound verbal
definitional patterns
The program processes the specialized texts and searches
for definitional contexts. However, not every verbal
definitional pattern that is found truly corresponds to a definition.
There are some other expressions that use the same patterns
with objectives other than give a definition. For this reason,
Those sequences were found in some specific positions. For instance, some negation words like no (not) or tampoco (either) were found in the first position before or after the definitional verbal pattern; also adverbs like tan (so) as well as sequences like no más de (not more than) were found between the definitional verb and the nexus como (like); finally, syntactic sequences like adjective + verb were found in the first position after the definitional verb.
Once the system has eliminated non-definitional contexts, it proceeds to identify the features that form the definitions. For this, it uses a decision tree based on regular expressions which allows the system to identify and tag the position of every feature. These regular expressions are:
Term = BORDER (Determinant) + Noun + Adjective. {0,2} .* BORDER
Pragmatic Pattern = BORDER (sign) (Preposition | Adverb) .* (sign) BORDER
Definition = BORDER Determinant + Noun .* BORDER
The whole process of definitional contexts detection is shortened in the following diagram. ECODE was originally developed as a definitions extractor from specialized texts. However, the same definitional verbal patterns that are used in formal documents are also used in informal ones.
With this in mind, the Language Engineering Group has been working on the development of DESCRIBE, an extended scope of ECODE which extracts definitions from texts written in colloquial language.
This new adaptation consists on a module that automatically extracts search results from the Internet about a particular term, and then retrieves the contents of the web pages that match that search and analyses them looking for new definitions.
This tool removes all definitions that are repeated, and looks not only in formal websites, but also in open forums, personal webpages, blogs and chats, which provides a rich variety of definitions.
In the end, DESCRIBE retrieves a list of definition candidates that still have to be depurated, since some of the candidates might not really be definitions. 5
In order to give the dictionary search engine another feature to help the correct identification and ranking of output terms, it has been considered classifying the definitions to match not only the words and the order in which they appear, but also the type of definition given by the user and the ones in the knowledge base.
There are four kinds of definitions based on the
Aristotelic definitional model: analytic, extensional, synonymic
and functional
In order to automatically provide a category for each definition obtained through DESCRIBE, the verbal patterns have been divided accordingly to the kind of definition in which they usually appear.
Definition type
Verbal definitional patterns Analytic Extensional Synonymic Functional
As most systems in Natural Language Processing, DESCRIBE is not perfect and sometimes the definition candidates turn out to be wrong, or the definition might be misplaced in a particular category.
The Language Engineering group has developed a tool to help the manual revision of the definition candidates’ validity and their categorization correctness. This tool presents a series of definition candidates to dictionary developers. Every candidate shown to the user has also the category in which DESCRIBE placed it.
The system allows the developers to easily accept or reject a candidate as a definition and it also allows them to change the category into which the definition was originally placed.
This system helps in the task of polishing the definitions that will be part of the knowledge base of the dictionary, but it also keeps a record of the definition candidates that have been rejected. This record is intended to be used as a corpus that will serve as training data for a machine learning system that will be used to improve the precision of ECODE and, in consequence, of DESCRIBE itself.
The definitions included in the knowledge base of specialized onomasiological dictionaries must cover both formal and informal concepts, and they also must cover as many forms of expressing them as possible in order to procure more accurate solutions for its users.
It is also convenient to classify the definitions in the knowledge base and the ones given by the user according to their type, so as to provide the search engine with more features to compare and match the user definitions with its own, hence improving its precision. Definition classification is the first step in the creation of ontologies.
In this paper we presented a methodology to automatically obtain definition candidates to fill the knowledge base of onomasiological dictionaries and also classify these definitions according to the Aristotelic definitional model. The source of these definitions is the Internet, which allows us to a very wide variety of speakers and, for that reason, a means of expressing concepts. This methodology has been used and tested in the creation of onomasiological dictionaries of Sexuality and Linguistics, among others, and can be applied to other subject areas, such as Biomedicine, Epidemiology, Veterinary, Laws, etc..
We also presented a tool which will make possible the creation of a corpus with both good and bad definition candidates marked as such. The purpose of creating this corpus is to obtain training data for a machine learning system directed to improve the automatic detection of definitional contexts.
We would like to acknowledge DGAPA for the sponsorship of the project “Análisis estilométrico para la detección de similitud textual”. We also thank the CONACYT Thematic Network “Tecnologías de la Información y la Comunicación”.