This article presents an automated construction of a logical information context from a terminological resource, available in xml ; we apply this to the resource FranceTerme and to Camelis tool and we discuss how the resulting context can be used with such a tool dedicated to logical contexts.
The purpose of this development and the choices related to this experiment is twofold : to facilitate the use of a rich linguistic resource available as open-data in xml ; to test and envision a systematic transformation of such xml resources to logical contexts. A logical view of a context allows to explore information in a flexible way, without writing explicit queries, it may also provide insights on the quality of the data.
Such a context can be enriched by other
information (of diverse natures), it can also be
linked with other applications (according to
arguments supplied by the context).
and their goals as well :
In this proposal, we want both : – to facilitate the use of a valuable linguistic resource (with a rich structure) and available in XML, and to allow its flexible querying and exploration without prior knowledge ; – we want to test and consider a systematic transformation (a transducer) from such resources (in XML) to logical contexts ; such contexts can be loaded in a software allowing rich and flexible browsing on data, combining various heterogeneous criteria ; the way we represent the information in the context may also have an impact on its ease of use.
The aim is here to perform a transducer so as to present the data in a logical information system without losing information content, but gaining in ease of exploration. Other advantages are provided by a safe navigation (no dead-end property) and serenpidity.
The resulting context is freely available 1. Terminological resource. The selected resource concerns the scientific and technical fields, it also interests us for the richness of its structure : its multilingual aspects, with definitions, synonymous relations, etc. its confirmed status (with source and date of publication), variations according to domain/subdomains or according to linguistic criteria (several variants of English, for 1. at http://www.irisa.fr/LIS/softwares example), possible absence or possible repetition of certain types of information.
This rich structure also allows further extensions : either with existing data or with new data that we organize in a similar pattern.
Logical context. A logical context is defined by a finite set of objects O, 2 and a finite set of logic descriptions d(oi) expressed using a well-formed logical language L.
A Logical context management system can load
and manage such a context, allowing querying
a context by logical requests (explicit or
interactive) ; then the answer is a sub-context of
objects satisfying the query. We used CAMELIS
(version 1) 3 for the experiment reported in this article.
This software is based on Logical concept
analysis (LCA) as defined in
Thereafter, we present in section 2 our transducer implemented in XSLT 4 and we specify the construction methodology. We present in section 3 how to exploit the transducer output on the FranceTerme resource containing terms of different scientific and technical fields ; we discuss several scenarios and benefits of this approach through this experiment. Additions and adjustments are proposed and discussed in 4 before concluding in section 5. 2 2.1
The transducer is designed to present data in a logical information system without losing information content, but gaining in ease of exploration. The diagram in figure 2 illustrates the approach, where the automated steps (solid arrows) are distinguished from manual or semi-manual ones (dotted line). Source Document Schemas. The transducer is designed to apply automatically on documents conforming to some document specifications. In general, such a specification can be automatically produced from an XML instance. We used DTD 4. a web availability is planned. Logical information system capabilities. We recall that a logical context system must allow the loading and management of a context described by its objects and its objects properties. More precisely, we assume that : - the logical context allows for some inferences : at least from classical logic (such as if A then A or B) possibly with axioms useful to model the context and organize its presentation (for example to reflect a taxonomy and only see some salient properties) ; - some general form of information retrieval and multi-faceted means are provided, as ”logical facets” and ”logical criteria” combinations ; - a modular and dynamic construction is allowed, for both sets of objects and sets of properties. CAMELIS system. In our experiment, we used the logical context tool CAMELIS, that, to our knowledge, is the only logical context management system. The tool interface displays three connected windows (see figure 3) : - a query window (top) ; - an object window (right) ; - an index window (left) as a navigation tree.
Properties in the navigation tree are organised as a clickable summary grouping hierarchy properties : it is important to note that a navigation link there corresponds to a sub-context (as in figure 5, the link/sub-context cardinality is given and a color is also associated with a concept :two navigation links with the same color lead to the same sub-context).
In the browsing mode, this tool allows three forms of query that select a sub-context : expert/query mode by editing in the query window ; the displayed objects become those statisfying the logic query ; the navigation tree is then presented in a form adapted to the new context ; example/object mode by selecting a set of objects in the object window ; the query automatically becomes an expression for the common properties of objects ; index/property mode by selecting a property (or more) in the navigation tree ; the objects 5. it is available at http ://saxon.sourceforge.net/dtdgen.html ;the terminological XML source file we used is accompanied with an XML Schema xsd, but without guarantee displayed become all those verifying the selected property (links) ; the query window is then automatically updated.
between the three windows is ensured. In addition, a session will not lead to an empty set of objects when following the links in the navigation tree : this important property is the navigational safety.
CAMELIS update and logic modularity. Using the
same interface, we can add and dynamically
update objects and properties, then export as a new
logical context file (useful for example to generate
a documentation for the objects of a selected
subcontext). The tool can also be adapted to choose a
dedicated logic, obtained by combination of logic
functors
We do not detail these last aspects in this article. Control. Part of the context information visible in the tool, can be retrieved by other means. We built some XPath queries to control the process and to produce complementary indicators.
We also built a control-context (figure 7). 2.2
We give here the characteristics and main stages of realization of the transducer, in its basic version. This construction is guided by the information in the DTD generated by XPath queries and control. Key selection. Defining a key in the source (by means of its document schema or of an XPath expression) is a preliminary stage, and the key definition plays a central role in the context construction. For this experiment with FranceTerme.xml, we considered //Article/@id (XPath).
Components. The program is designed to facilitate its updating, structured by source components and similar typical treatments. The treatment of a given source component depends on its kind (XML element, attribute) the relevant part of context, its status (optional, repeatable or not), the domain of the source content, and the desired rendering (data type, property name, property hierarchies).
Main Loop. For each source item Article : — each object get a unique label (extracted by Terme[@statut=’privilegie’]), used for the object presentation and for a string property in the navigation index ; — the key becomes a number property articleID = ... (xslt 6) ; — the publication date is processed to appear in the index at different levels of date detail ; — most other components are processed to produce strings of the form : property name is "string value" — property names may depend on several XML components (such as Terme element with a @statut attribute), they are organized to allow their grouping and multiple levels of detail (Terme ? is more general than Terme SYNONYME) ; — we also give a common prefix Plus to properties for data in the source, but not visible in the ressource site (such as data about committees) ; 7 — for XML elements that can be repeated for a given key/object, (such as Terme) we use an inner loop ; at this stage the output file context contains the description of one object per line, with its main properties ; — other components (such as foreign equivalents or antonymes, optional or repeated) are rendered by rules of the form : rule extr (key=id) --> (prop1 is val1) that automatically associate the property to the object designated by the key. 2.3
For treating a logical property related to an XML component, such as <Attention> child (optional and repeatable) of an <Article> identified by its attribute id, several alternatives are possible : -to indicate the name of property and its value by assembling and repeating the property name for the object, using this pattern : mk "object" key=id, ..., is prop1 val1, val2 prop1 is ... is ... prop2 -to indicate each property value by transformation rules, using this pattern, when the object is assumed to be already created and associated with the key : rule extr (key=id) --> (prop1 is val1) 6. <xsl:value-of select="concat(’articleID=’$varArtId)"/> 7. in the navigation tree, compound names of the properties are grouped by prefixes, details appear by opening a link Plus ? rule extr (key=id) --> (prop1 is val2) this will automatically add each property value to the object with key id.
This second solution brings some modularity since we can put rules in separate files, the properties being effectively added to objects after a file import. We chose this approach by means of rules and keys for some repeatable components. 8 3
In this part, we discuss several possible search modes in the resulting context, where navigation links (incremental) correspond to logical concepts that can be selected.
contains translations in several languages, with variations for the same language. Those data are attached to various domains and subdomains (possibly several ones for a given object).
Scenario. An exploration of the logical context can be conducted that way, for example : - open the Domaine ? property in the index ; - select-click Domaine is "Informatique" (computer science), this yields the corresponding sub-context (with 3 coherent views) ; - we may further select-click Domaine is "Droit" (Law), also automatically expressed as Domaine is "Informatique" and Domaine is "Droit" in the top window ; - open the Equivalent ? property then open
Equivalent en is "..." in the index etc. - open the PubliArticle ? property then
PubliArticle date = 2014 in the index etc.
Another simple search on ”streaming” is shown in figure 3.
Sub-context Cardinalities. In the property index-tree, we may choose an order for displaying a given facet. This is useful for example to read directly which Equivalent en correspond to the greatest number of French terms (figure 3).
8. Here is a typical xslt fragment (with some special symbols treatments for compatibility) :
<xsl:for-each select="Attention"> <xsl:value-of select="concat($varRulePart1,’idArticle=’, $varArtId,$varRulePart2, ’Attention is ’, $varQuote, translate(normalize-space(./text()), $varQuote, $varBQuote), $varQuote)"/> <xsl:text> </xsl:text> </xsl:for-each> Data types. Data types other than attributes and strings can be handled, Figure 4) shows a possible use of dates, allowing for more or less fine-grained selections. kind of subcontext summary and extra informations (here Attention is the focus element).
comes from selections in the index tree ; links of a same color characterize the same set of objects (concept).
This example illustrates the identification of potential linguistic errors (in a domain / subdomain)
Other examples such as ”package” (Equivalent en) or some abreviations (”ABS”) can be highlighted as ambiguous : the dynamic navigation links (domains, etc.) then provide hints to disambiguate. 3.2
Using the Not button at a given stage, we can arrive at a subcontext characterized by A2 = not(A1) and A0 as in Figure 6. Property A2 expresses a search for exceptions to A1, we get a
This navigation mode allows to detect abnormalities, in particular pseudo-empty properties appear on other facets (through a link like Definition is "") these cases often correspond to existing but empty XML elements (but are not XML errors). Low cardinalities in the navigation tree may suggest to explore the link, by selecting it and opening other facets simultaneously ; we can analyse this way ”the words without translation, following the not Equivalent? link.
In case of redundancies, these may become easily noticeable throusg browsing : exploring the Antonymes facet, we can see XML structuring redundancies (this information being carried by two source elements).
The logical context software can assign actions to objects by properties ; clicking on an object label then shows a contextual action menu.
This is useful in particular to inspect objects in their source xml file.
Other kinds of control (for coverage, counts, etc.) are made - by XPath queries on the XML document used to verify if certain characteristics of particular subcontexts (planned or explored) are consistent with the source document ; - a meta context built, following the DTD schema, whose objects are : element names, and the pairs (attribute name, element name). These objects are associated with actions parameterized by their label, in our case (Figure 7), the action is an XPath query using BaseX ; This can be adapted easily to another set of controls.
A logical context tool such as CAMELIS by its genericity and its features, allows many alternatives to represent and use a terminological resource like FranceTerme.
Some initial choices can be easily revised or completed ; for example the name of a property can be changed directly through the interface (or by simple transformation of the context file).
The choices and refinements should provide a better context. Several quality criteria can be considered : effectively obtaining a desired result (usefulness / completeness) ; the number of steps to get there (effectiveness) ; rich browsing indexes (multiple views) and efficient indexes to pursue the navigation (through the proposed increments) ; a flexible mode of interrogation and ease of interpretation.
We indicate some possibilities through modifications in the experiment. 4.1
Domains and SubDomains data have been translated. The resulting two context files contain update rules of the form :
rule extr Domaine is "Acoustique" --> Domain is "Acoustic"
when loaded in the context, properties on the right hand side are added to all objects verifying the left hand side.
In the original context, we can see (with an appropriate ordering) that among the terms with a category attribute, the names (categorie is "nm" or categorie is "n") are the majority, followed by categorie is "adj.". However these grammatical categories are listed with various values, we can observe : which may include in particular : - a disjunction, as in : categorie is "adj. ou n.m." and Equivalent en is "crossmedia (n. ou adj.)" which selects the term transme´dia ; but we also observe its permutation categorie is "n.m. ou adj." ; - a more or less fine granularity, as in : categorie is "n.m.inv."
The addition of rules and axioms in the logical context permits to harmonize these properties, resulting a more structured navigation tree according to this facet. A few lines in the resulting context define a hierarchy of categories, such as : rule extr categorie is "n.m.inv." --> Categorie n m inv ...
Categorie n m inv axiom, Categorie n m Categorie n m axiom, Categorie n ...
Note that such improvements could apply to other terminological resources and result from linguistic analysis or other lexicons. 4.3
We have seen that a property Terme? covers three statutes (PRIVILEGIE, SYNONYME, ANTONYME). A context of axioms can facilitate a search on all or a subset of these variants : axiom PRIVILEGIE, SET axiom SYNONYME, SET axiom SET, ANY axiom ANTONYME, ANY
A query may then group several properties (having a status below another expression like SET) as examplified in figure 8.
This example shows a query for Terms (including status variants) containing negation (”non”).
Note that such axioms can be added or modified in a modular way. 4.4
The website FranceTerme allows to select new terms, from the description of a current term (denoted as t1) by a link See also. This is rendered in the context navigation tree by the facet Voir aussi (See also) is "... information on tj" (denoted as fj ) where the term to see tj is shown with its key Article . Two modes of translation of this piece of information have been tested : - in a basic mode, as a simple property fj of the current term t1 - in a full (reflexive) mode, where t1 has fj and also f1 : Voir Aussi is "... information on t1". 9
This second mode allows this type of scenario : while t1 has property fj , select this fj link in the navigation tree ; by reflexive closure, tj is also in the sub-context, and can be further selected.
We treated in the same way the reciprocal link of Voir Aussi, by adding (by the transducer) a Voir Depuis (See From) "..." property. This enables to group into a sub-context the terms poin9. no addition to the terms that have no link ting to (or pointed to by) a particular word (or more).
- In the context for FranceTerme, we observe some terms (15) satisfying this query : Voir Aussi ? and not Voir Depuis ? these are the ”terms pointing to an article, but not pointed to” ; - according to the resource schema, other elements (synonyms, antonyms, ...) could be treated similarly, with navigation links in context. 10
tions. As explained about control, the logical context management software allows to associate actions to objects.
We can use this mode to associate an object with a process (or more) on this item that may be introduced in the interface from a context menu related to the object (or group of objects). The setting can be provided at the transducer level. A file describing these actions can be later loaded from the interface.
We generated connections to :
- a parser, installed locally : the processing chain
(open) Bonsai
This action list is not exhaustive and can be adapted. In particular, we could consider links (local or not) with other analyzers, or other linguistic resources and retrieve results to enrich the logical context with new properties. The capabilities of Full text search (of BaseX) could also be ex10. this treatment is not currently done for the other elements (in the source xml these terms do not necessarily correspond to an article/object). ploited. 5
The general aim of this proposal was to show how a logical concept analysis (LCA) framework and tools could be beneficial for browsing terminological resources ; through this experiment the purpose was twofold : - to facilitate the use of a useful language resource (rich structure) and available in XML, - to envision a systematic transformation of such resources as XML to logical contexts.
Improvements may also be suggested and brought to the data ; other treatments may also be eased, for example a selected sub-context can be exported as text and generate other results (such as a documentation).
We illustrated how a logical context allows to explore linguistic information, in a flexible way, without a priori knowledge, and also get guidance on data quality (in the navigation tree, counts and colors for concepts, ...) New linguistic information (personal, enterprise, ...) could be incorporated easily in the initial context (if they comply with the document model and the key assumption).
Additional data to compare and enrich the
content can also be added in several ways and for
many languages, (for French : Wordnet Wolf
Other actions corresponding to linguistic processing can be added to the context :parsing the expression (several languages), syntactic head, etc. We could also consider inverse properties (such as translation) and enrich the context with these objects.
Moreover, it seems that the development method could be transposed to other open data and linguistic xml ressources. To some extent, the construction of the transducer presented here could be automated if it relies on a determination of a key and a grid indicating for a source component, its label, its type, its repeatability, and the way it should be rendered. A similar experiment could be carried out by adapting the standards and software tools of the semantic web. Finally, we mainly discussed browsing, future work could also concern updates.