Recently, the Semantic Web has experienced signi cant advancements in standards and techniques, as well as in the amount of semantic information available online. Even so, mechanisms are still needed to automatically reconcile semantic information when it is expressed in di erent natural languages, so that access to Web information across language barriers can be improved. That requires developing techniques for discovering and representing cross-lingual links on the Web of Data. In this paper we explore the di erent dimensions of such a problem and re ect on possible avenues of research on that topic.
The large and growing amount of semantic data available on the Web, mainly
in the form of Linked Data [
In this paper we focus on the second challenge, namely, the need of establishing, representing, and storing cross-lingual links among semantic information on the Web. In fact, in the multilingual Web of Data that we envision, semantic data with lexical representations in one natural language would be mapped to equivalent or related information in other languages, thus making navigation across multilingual information possible for software agents. In the following we will refer to \cross-lingual ontology linking" in a broad sense, including (semi-) automatic ontology and instance matching methods and techniques applied to the linking of semantic data documented in several natural languages.
The problem of cross-lingual linking is a fundamental one, since more and
more legacy data sources available in di erent natural languages are being
transformed into linked data, and have to be linked to be exploited at its full potential.
In fact, the establishment of links between or among multilingual data sources
would also contribute to the localisation issue, since it would transform
monolingual, isolated, data resources into \multilingual resources" just thanks to the
links. However, the linking of resources documented in di erent languages is
not so immediate. Several issues that arise in the localization of semantic web
resources [
The main purpose of this position paper is to give an insight into the problem of cross-lingual linking on the Web of Data and identify some research topics that will allow us to advance towards a truly multilingual Web of Data. In the rest of the paper (Section 2) we refer to the di erent knowledge representation levels in which cross-lingual links can be established. Then, we explore the problem and identify possible research lines grouped in three aspects: cross-lingual link discovery, representation, and reuse. Finally, the main conclusions of the paper are summed up in Section 3. 2
Cross-lingual links between ontologies and data sources can be established at di erent knowledge representation levels: 1. Conceptual level: links between ontology entities at the schema level. 2. Instance level: links between data underlying ontologies. 3. Linguistic level: links between lexical representations associated with ontology concepts and/or instances.
The last one is particularly important if certain lexical relations have to be represented across ontologies (e.g., translations or term variations). Each of these levels will require its own link discovery/representation methods and techniques.
In the following we propose some enhancements of available methods and techniques and suggest new avenues of research that could help overcome the problem. 2.1
Current ontology matching techniques have to be extended with multilingual
capabilities, and novel techniques need to be investigated as well. Cross-lingual
links can be discovered by means of some of these techniques:
1. Projecting the lexical content of the mapped ontologies into a common
language (either one of the languages of the aligned ontologies or a pivot
language) e.g., using machine translation.
2. Comparing the ontology entities directly by means of cross-lingual semantic
measures, that is, measures capable of evaluating similarity or relatedness
between (ontology) entities documented in di erent natural languages (e.g.,
cross-lingual explicit semantic analysis [
Both avenues have to be further explored, compared, and possibly combined. There are a number of early cross-lingual ontology alignment tools that already implement the rst technique1, while the second one remains unexplored yet. Notice that such preliminary systems are intended to discover cross-lingual links at the conceptual level and that cross-lingual alignment systems operating at the instance and linguistic levels are still to come.
An alternate way to discover cross-lingual links is by using the Web of Data
as a source of background knowledge. The idea is to infer links from other links
already existent among online ontology entities (that are similar to the entities
I intend to link). Such an approach was explored in a monolingual context by
the Scarlet system [
In principle, existing constructs of ontology languages can be utilised for representing cross-lingual mappings at the conceptual and instance levels (e.g., owl:sameAs or owl:equivalentClass), whenever the two concepts or instances can be considered cross-lingual equivalents.
Other commonly used vocabularies (e.g. rdfs:subclassOf, skos:narrower or skos:broarder) could also be re-used in case of granularity discrepancies, i.e., when one conceptualization regards a certain concept with a granularity level di erent from the other conceptualization. In this case, we would suggest an adaptation or enhancement of such relations for a multilingual scenario, so that ner language distinctions are captured.
In the case no equivalence exists (the one language does not conceptualize a certain phenomenon of the world, whereas the other has a concept for it), we could still provide a lexical description for the \inexistent concept" in the target language, provide a link to its closest concept, and signalize it as a speci c crosslingual case. We believe this kind of links should also be accounted for in the Web of Data.
Regarding cross-lingual mappings at the linguistic level, mappings could be
established between the natural language descriptions of their concepts. At this
level, lexical-semantic relations could be used (hypernym-hyponym, synonym,
antonym, translation, etc.). In the simplest case in a cross-lingual scenario, a
property labelled \translation" or \cultural equivalent" (for instance) might be
established between the lexical realizations of the concepts [
We argue that speci c representation models have to be able to de ne speci c relations between natural language descriptions in di erent languages, what
http://oaei.ontologymatching.org/2011.5/multifarm/index.html in
OAEI2011.5
we term translation relations or cross-lingual relations. Highly related with this
issue is the representation of term variation at a monolingual or multilingual
level. A term variant has been de ned as \an utterance which is semantically
and conceptually related to an original term" [
Further, to facilitate processing and interchange of alignments, speci c formats has been proposed in the literature such as the Alignment Format 2 or the EDOAL language 3. They should be explored and, if needed, extended to accommodate the representation of cross-lingual and multilingual alignments. 2.3
Cross-lingual links can be discovered runtime/o ine. However, owing to the growing size and dynamic nature of the Web, it is unrealistic to conceive a Semantic Web in which all possible cross-lingual links are established beforehand. Thus, scalable techniques to dynamically discover cross-lingual links on demand of semantic applications have to be investigated. Although the scalability requirement is not inherent to the multilingual dimension in ontology matching, multilingualism exacerbates the problem due to the introduction of a higher heterogeneity degree and the possible explosion of compared language pairs.
On the other hand, one can imagine some application scenarios (in restricted domains for a restricted number of languages) in which computation and storage of links for later reuse is a viable option. In that case, suitable ways of storing and representing cross-lingual links become crucial. Also links computed runtime could be stored and made available online, thus con guring a sort of pool of crosslingual links that grows with time. Such online links should follow the Linked Data principles to favour their later access and reuse by other applications. 3
In this paper we have motivated the study of cross-lingual ontology links as one of the fundamental challenges to solve in order to attain the goals of a truly multilingual Web of Data. There are, in particular, three subproblems to treat, namely cross-lingual link discovery, representation, and reuse. We have given an overview of the characteristics of each of them, as well as identi ed some relevant research topics that have to be further explored to be part of the solution. For instance, representation of cross-lingual links at the linguistic level, as well as the study of cross-lingual semantic measures and cross-lingual ontology alignment techniques. In our view such topics require more atention by the community and
will be crucial to enable the multilingual capabilities on the Web of Data. Acknowledgments. This work is supported by the EU project Monnet (FP7248458), the Spanish national project BabeLData (TIN2010-17550), and the Spanish Ministry of Economy and Competitiveness within the Juan de la Cierva program.