Introduction to RDF and related Vocabularies. Introduction to SPARQL. Thierry Declerck, John P. McCrae

Transcription

1 Introduction to RDF and related Vocabularies. Introduction to SPARQL Thierry Declerck, John P. McCrae

2 About Language Resources

3 What are Language Resources (LRs)? The LREC perspective A large conference is dedicated to Language Resources: LREC (Language Resources and Evaluation, see When submitting contributions, authors have to select between different areas (types of LRs): Evaluation Infrastructural Issues/Large Projects (for example Elexis ) Multimodality (corpus, lexicon or tools. Mono-, bi-, multi- or crosslingual, or language independent) Speech (corpus, lexicon or tools. Mono-, bi-, multi- or crosslingual, or language independent) Terminology (Mono-, bi-, multi- or crosslingual, or language independent) Written (corpus, lexicon or tools. Mono-, bi-, multi- or crosslingual, or language independent) For those different types of language resources, there is also the issue on how they are represented/encoded (raw or csv files, databases, XML, RDF, JSON etc.) And we should also be aware of legal aspects (licences, etc) realted to LRs 3

4 The Types of Language Resources listed on the Linguistic Linked Open Data (LLOD) cloud (State of the cloud in July 2017). Corpora; Terminologies, Thesauri and Knowledge Bases; Lexicons and Dictionaries; Linguistic Resource Metadata; Linguistic Data Categories; Typological Databases. More details on those areas and how to access the resources in the LLOD will be given during the lecture!

5 Examples of Repositories, Infrastructures and Metadata for Language Resources (not exhaustive listing) There are repositories and catalogues where one can access and download (or request for) languages resources. Some examples: ELRA (European Language Resources Association, also organizing the LREC conferences): LDC (Linguistic Data Consortium): META-SHARE: (European Initiative) CLARIN (Common Language Resources and Technology Infrastructure): ELEXIS (European Lexicographic Infrastructure): Meta-Share Metadata Model: IMDI Metadata: Linghub: (more details on this metadata infrastructure in this There are also language resources infrastructures, some examples: Importance of Metadata: for accessing the data and information included in those repositories and infrastructures, there is a need to manage and use appropriate metadata, some examples: lecture!) 5

6 LRE-Map: A special repository related to papers submitted to the LREC conferences The LRE-Map is a special type of repository, resulting from descriptions of language resources used or developed (and described) by the authors of submitted papers to the LREC Conferences. This effort is leading by the promoters of this repository to a (partial) standardisation of the names given to resources (for example WordNet which is written by users in many different ways, but also considering the used versions. LRE-Map has been opened to contributions from other conferences (see for a short introduction) Direct access to LRE-Map: Inviting all colleagues to add information and data to this repository! 6

7 About Web Resources

8 What is a web Resource? Looking first at Wikipedia ( The concept of a web resource is primitive in the web architecture, and is used in the definition of its fundamental elements. The term was first introduced to refer to targets of uniform resource locators (URLs), but its definition has been further extended to include the referent of any uniform resource identifier (RFC 3986), or internationalized resource identifier (RFC 3987). In the Semantic Web, abstract resources and their semantic properties are described using the family of languages based on Resource Description Framework (RDF). (Bold face from us. RDF to be described in detail in this lecture) In the early specifications of the web ( ), the term resource is barely used at all. The web is designed as a network of more or less static addressable objects, basically files and documents, linked using uniform resource locators (URLs). A web resource is implicitly defined as something which can be identified. The identification serves two distinct purposes: naming and addressing; the latter only depends on a protocol. It is notable that RFC 1630 does not attempt to define at all the notion of resource; actually it barely uses the term besides its occurrence in URI, URL and URN, and still speaks about "Objects of the Network". RFC 1738 (December 1994) further specifies URLs, the term 'Universal' being changed to 'Uniform'. The document is making a more systematic use of resource to refer to objects which are 'available', or 'can be located and accessed' through the internet. There again, the term resource itself is not explicitly defined. 8

9 From web resources to abstract resources ( The first explicit definition of resource is found in RFC 2396, in August 1998: A resource can be anything that has identity. Familiar examples include an electronic document, an image, a service (e.g., "today's weather report for Los Angeles"), and a collection of other resources. Not all resources are network "retrievable"; e.g., human beings, corporations, and bound books in a library can also be considered resources. The resource is the conceptual mapping to an entity or set of entities, not necessarily the entity which corresponds to that mapping at any particular instance in time. Thus, a resource can remain constant even when its content---the entities to which it currently corresponds---changes over time, provided that the conceptual mapping is not changed in the process. Although examples in this document were still limited to physical entities, the definition opened the door to more abstract resources. Providing a concept is given an identity, and this identity is expressed by a well-formed URI (uniform resource identifier, a superset of URLs), then a concept can be a resource as well. In January 2005, RFC 3986 makes this extension of the definition completely explicit: ' abstract concepts can be resources, such as the operators and operands of a mathematical equation, the types of a relationship (e.g., "parent" or "employee"), or numeric values (e.g., zero, one, and infinity).' The view of OntoLex-Lemon (more details later in this lecture): not only entities or concepts are web resources, but also language data. Not only corpora, lexicons, terminologies, but all types of lexical elements. 9

10 About RDF (Resource Description Framework)

11 Resources in RDF and the Semantic Web ( b) First released in 1999, RDF was first intended to describe resources, in other words to declare metadata of resources in a standard way. A RDF description of a resource is a set of triples (subject, predicate, object), where subject represents the resource to be described, predicate a type of property relevant to this resource, and object can be data or another resource. The predicate itself is considered as a resource and identified by a URI. Hence, properties like "title", "author" are represented in RDF as resources, which can be used, in a recursive way, as the subject of other triples. Building on this recursive principle, RDF vocabularies, such as RDFS, OWL, and SKOS will pile up definitions of abstract resources such as classes, properties, concepts, all identified by URIs. RDF also specifies the definition of anonymous resources or blank nodes, which are not absolutely identified by URIs. Now from Tim Berners-Lee: see (see next slide for a short quotation out of this page) 11

12 Tim Berners-Lee on RDF, in A Short History of Resource in web architecture. ( When RDF was first developed, it was motivated by the need for data about resources very much in the online information sense: data about documents, or 'metadata'. In fact it was designed to be able to describe anything, but many early users of RDF referred to it as metadata technology. RDF used the word "resource" rather awkwardly in fact as it turned out. In the beginning, many of the things being described were documents, and so the online information meaning of resource made sense. But in fact in RDF the resource was allowed to be anything at all. A class, rdf:resource even used the term as the universal class of all things. A little later, the Web Ontology Language decided to use Thing for that. (addition from us, also Nothing is a resource, for example an empty set) RDF came along in what I think was a neat way. It used completely existing web protocol extension devices to introduce a new system which was fundamentally different from the old HTTP+HTML one. The HTML web was a hypertext model, which pages and anchors. The RDF model was a knowledge representation one of arbitrary things. It did this by using the fact that a new language can define whatever it likes as what a local identifier denotes. A graphic language might use local identifier to denote lines and points. HTML used local identifiers to identify hypertext anchors. RDF used them to identify arbitrary concepts, people, whatever. So, in our view: if RDF can identify and represent arbitrary concepts, people, whatever, it should also be able to identify and represent any type of linguistic data! 12

13 Getting deeper in RDF (Resource Description Framework) The Resource Description Framework (RDF) is a family of World Wide Web Consortium (W3C) specifications originally designed as a metadata data model. It has come to be used as a general method for conceptual description or modeling of information that is implemented in web resources, using a variety of syntax notations and data serialization formats. It is also used in knowledge management applications. ( RDF was adopted as a W3C recommendation in The RDF 1.0 specification was published in 2004, the RDF 1.1 specification in ( In RDF one can write statements about something in the form of a triple: Subject + Predicate + Object. While this sounds like a basic syntax analysis, it is a bit more complicated, but still straightforward. An example (discussed next slide), showing a RDF Graph: Figure taken from Auer, Sören & Lehmann, Jens & Ngonga Ngomo, AxelCyrille. (2011). Introduction to Linked Data and Its Lifecycle on the Web. Reasoning on the Web in the Big Data Era, Athens, Greece / _1. 13

14 Getting deeper in RDF (Resource Description Framework): Why green nodes and red squares (in fact terminal nodes)? The figure introduced in the former slide contains green nodes, blue arrows and red squares. The green nodes are representing/denoting instances of classes ( Leipzig is an instance of the class City or similar). A subject in this figure can be recognized by the fact that it has an outgoing arrow. The reverse is valid for an object : it has an incoming arrow. An object is represented/denoted in the figure either by a green node (an object in our data model, a class or an instance of a class), or by a red square (a literal, thus not an object/ressource within our data model). The blue arrows represent the properties (or relations) between the subjects and the object (the predicate of the triple). A property pointing to a red square is called a datatype property (pointing to a literal data). Else we have object properties (pointing to an object). Subject and Predicate of a triple must be represented/encoded as an URI. The same is valid for an object pointed to by an object property. An object pointed to by a datatype property is represented/encoded as a data literal (can be classified by a XSD type*). Figure taken from Auer, Sören & Lehmann, Jens & Ngonga Ngomo, AxelCyrille. (2011). Introduction to Linked Data and Its Lifecycle on the Web. Reasoning on the Web in the Big Data Era, Athens, Greece / _1. * XSD types are: anyuri, base64binary, boolean, date, datetime, decimal, double, duration, float, hexbinary, gday, gmonth, gmonthday, gyear, gyearmonth, NOTATION, QName, string, and time 14

15 A Note on RDF as a data model vs serializations of RDF. RDF Model: At the core of RDF is a model for representing named properties and their values. These properties serve both to represent attributes of resources (and in this sense correspond to usual attribute-valuepairs) and to represent relationships between resources. The RDF data model is a syntax-independent way of representing RDF statements. ( RDF Serialization: The RDF data model provides an abstract, conceptual framework for defining and using metadata. A concrete syntax is also needed for the purposes of authoring and exchanging this metadata. The syntax does not add to the model; APIs may be provided to manipulate RDF without reference to a concrete syntax. RDF uses the Extensible Markup Language (XML) encoding as its syntax. However, RDF will not require (and conforming implementations must not require) an XML Document Type Declaration for the contents of assertions. In this respect RDF requires at most the XML wellformedness constraints. RDF schemas may but are not required to be XML DTDs. ( Examples of serialization of RDF are among others rdf/xml, turtle, json-ld and Ntriples 15

16 Now: What is the place/role of language data in RDF? If we represent now partially the graph discussed in the former slides into a serialization of the RDF data model (here in the N-Triples format), we can see that no language data is directly involved: (Example taken again from Auer, Sören & Lehmann, Jens & Ngonga Ngomo, Axel-Cyrille. (2011). Introduction to Linked Data and Its Lifecycle on the Web. Reasoning on the Web in the Big Data Era, Athens, Greece / _1 ). The following example Burkhard Jung is the mayor of Leipzig (subject) (predicate) (object) Can then be serialized as: < < < > (subject) (predicate) (object) What we observe: in this example, all the language data are included in the URIs. It is thus hard to detect it and to refer to it. While we deal in this example mainly with Named Entities, we would have the same issue with Substantives, so for example for Horse (taken from DBpedia), also in N-Triple notation: < < < 16

17 Now: What is the place/role of language data in RDF? (2) As we saw in the former slide, language data is included in the URI ( horse, mammal ). Nothing, in term of linguistic description, can be stated about the elements of a URI. And there is also the issue of multilinguality: how can I express that cheval is the French equivalent to horse? Creating a new URIs (< and stating that it has the same denotation as the English URI? Cumbersome, redundant and awkward, but a vocabulary built on the top of RDF, OWL, has a property called owl:sameas that would allow this kind of statement between two URIs, but this would not help at all to mark linguistic properties! As we also saw before, literal (also natural language data) can be used as values of a datatype property, like for example < <ex:hascolour> white. Here white is the literal object value of the triple. But as RDF states that only between URIs a relation (property) can be established, there is no way to mark white as the subject of a triple that would link by a property to an adjective object To sum up: within the basic RDF there is no way to straightforwardly add linguistic description to language data used in the conceptual world / data model represented by the RDF formal language. 17

18 About Extensions of RDF

19 Ways to better include, use and represent Language Data in the Context of RDF A series of vocabularies has been build in the context of RDF, expanding its expressivity and coverage. RDF-Schema (or RDFS or RDF(S) ) is one of those vocabularies. It is supporting the taxonomic or ontological modelling of RDF data (supporting the addition of more structural properties to the RDF Graph. An example: Picture By Karima Rafes - Own work, CC BY-SA 3.0, RDFS allows also to formulate restrictions on the types of resources that can serve as a subject (rdfs:domain) or an object (rdfs:range) of a property (see example next slide) 19

20 Annotation Properties in RDFS So-called annotation properties allow to add human readable information using natural language expressions. Examples of such properties are: rdfs:label rdfs:comment rdfs:seealso rdfs:isdefinedby An example for rdfs:label The use of rdfs:label supports multilingual descriptions of an object in the data model (in this example the URI is using an English word, but RDFs labels are in 2 languages) Taken from 20

21 Shortcomings of rdfs:label (rdfs:comment, etc.) The annotation properties are pointing to (or have as value of rdfs:range) literals. So that we can not express relations between two rdfs:label values (like stating for example that they are translations of each other). Nor can we add any linguistic information, like PoS, Gender, Number, etc. A reminder: A property (or relation) can only be established between resources in the data model, and which are necessarily encoded as a URI As a consequence, we can for example not (formally) state that certain types of concepts are expressed only by nouns (also with possible restrictions on their morphology). As another consequence, we can not with rdfs:label (formally) state which sense of a label for a certain class should be considered. So for example for bank : Taken from Kuekenhoff_Canal_002.jpg 21

22 Improvements with two Vocabularies built on the Top of RDF: SKOS and SKOS-XL Another vocabulary that is building on the top of RDF is SKOS (Simple Knowledge Organization System, see SKOS provides a model for expressing the basic structure and content of concept schemes such as thesauri, classification schemes, subject heading lists, taxonomies, folksonomies, and other similar types of controlled vocabulary ( Interesting from the point of view of the use of natural language in knowledge systems is the fact that a first step towards a differentiation of such expressions is done. SKOS foresees three different types of labels: skos:preflabel: Only one preflabel is allowed (for each language used) in the context of a knowledge object. This label should contain the authorized term for a concept for each langauge used) skos:altlabel: supports the use of terminological variants for denotating the concept (synonyms, acronyms, etc.). More than one altlable per language is allowed. skos:hiddenlabel: is a lexical label for a resource, where a KOS designer would like that character string to be accessible to applications performing text-based indexing and search operations, but would not like that label to be visible otherwise. Hidden labels may for instance be used to include misspelled variants of other lexical labels ( 22

23 A Graphical View on the Concepts and Labels organisation in SKOS Taken from: 23

24 Graphical Example of skos: < con: < con:c2 a skos:concept; skos:preflabel "Political violence"@en; skos:altlabel "Civil violence"@en; skos:altlabel "Violent protest"@en; skos:broader con:c4; skos:narrower con:c3. con:c3 a skos:concept; skos:preflabel "Terrorism"@en; skos:broader con:c2. con:c4 a skos:concept; skos:preflabel "Violence"@en; skos:narrower con:c2. Taken from 24

25 How to express Relation between multilingual Labels? Not possible in SKOS, but supported in SKOS-XL (next slide) Taken from: It is not possible to formally establish a translation relation between two labels in SKOS, as the value of the skos:label properties are still literals! 25

26 Labels as Resources/URI in SKOS-XL SKOS-XL defines an extension for the Simple Knowledge Organization System, providing additional support for describing and linking lexical entities ( Now labels are equipped with an URI and become a (Web) resource, and so relations (properties) can be established between them!, like for example: istranslationof hasisocode IsHypernymOf Etc In this also a better support of conceptual (multilingual) terminologies is supported as terms are the main building blocks of such knowledge systems, now with the same web resource dignity as the concepts. 26

27 Graphical Example of SKOS-XL (including the former Example of SKOS), slide skos: skosxl: con: term: < term:t1 a skosxl:label; skosxl:literalform "Civil violence"@en. con:c2 a skos:concept; skos:preflabel "Political violence"@en; skosxl:preflabel term:t2; skos:altlabel "Civil violence"@en; skosxl:altlabel term:t1; skos:altlabel "Violent protest"@en; skosxl:altlabel term:t5; skos:broader con:c4; skos:narrower con:c3. term:t2 a skosxl:label; skosxl:literalform "Political violence"@en. con:c3 a skos:concept; skos:preflabel "Terrorism"@en; skosxl:preflabel term:t3; skos:broader con:c2. term:t3 a skosxl:label; skosxl:literalform "Terrorism"@en. con:c4 a skos:concept; skos:preflabel "Violence"@en; skosxl:preflabel term:c4; skos:narrower con:c2. term:t4 a skosxl:label; skosxl:literalform "Violence"@en. term:t5 a skosxl:label; skosxl:literalform "Violent protest"@en. Taken from 27

28 A A Repository of SKOS and SKOS-XL Resources: is an interesting example on how terms (as skosxl:label) are now encoded in SKOSXL 28

29 Conclusions We have introduced the concept of Web Resources and shown how the Resource Description Framework (RDF) has been developed in the context of the Semantic Web for encoding (and linking) such resources. We have discussed the use of natural language data in the context of RDF data model, and seen that language data are used only as part of URIs or in the context of literal values of datatype properties We introduced vocabularies that were built on the top of RDF and that allow a better exploitation of natural language data. We presented in this session the vocabularies building on RDF: RDF(s), OWL, SKOS/SKOS-XL and we have shown how natural language data is increasingly better handled in such vocabularies, reached a status similar to those of (instances of) classes. But we consider that SKOS-XL as such is not well-suited for a detailed representation of lexical data, below the level of terms (expressed by skosxl:preflabel). Therefore a W3C Community Group was established to design a RDF based model that supports a full Semantic Web and Linked Data compliant description of lexical data. This model, OntoLex-Lemon will be presented in another lecture (but see 29

30 About SPARQL

31 RDF Data Models are representing Graphs! A data model in RDF is representing a graph. A subject node (URI) is related by a property (URI) to either an object node (URI) or a literal (other type of data). This type of statement is also called triple, as they have the form Subject Predicate Object. Graphically, this looks like: 31

32 Triple Stores RDF triples can be stored, either as files or in so-called triple stores. In such stores they can be straightforwardly queried, using the SPARQL query language. SPARQL allow to match a query expressed in the form of a graph (but using variables for either the subject, the predicate or the object, or all of them!) with the stored graph (set of triples). We can for example ask for returning all the triples included in the store, just querying for?s?p?o (give me all statements consisting of a subject, a predicate and an object). This can be for sure restricted (which is making much more sense!), and so we can ask for all instances of classes with a query like?s rdf:type?o (give me all triples in which we have the statement subject is of rdf:type object, or to query for all subclass relations?s rdfs:subclassof?o, etc. To sum up: SPARQL is a graph pattern matching engine. 32

33 This slide is slight modifications from SPARQL is a recursive acronym, which stands for SPARQL Protocol and RDF Query Language. SPARQL consists of two parts: query language and protocol. The query part of that is pretty straightforward. SQL is used to query relational data. XQuery is used to query XML data. SPARQL is used to query RDF data. Despite this similarity, SPARQL differs in that it was designed to operate over disconnected sources over a network in addition to a local database. In particular, the SPARQL protocol allows transmitting SPARQL queries and results between a client and a SPARQL engine via HTTP. A SPARQL endpoint is simply a server that exposes its data via the SPARQL protocol. At its most basic, a SPARQL query is an RDF graph with variables. For example, consider the following RDF graph: ex:juan foaf:name Juan Sequeda. ex:juan foaf:based_near ex:austin. Now consider a version of the previous RDF graph that has variables instead of values:?x foaf:name?y.?x foaf:based_near?z. 33

34 More (slightly modified) from At first blush, this is not very different from RDF itself, and that s intentional. SPARQL queries are based on the concept of graph pattern matching. A basic SPARQL query is simply a graph pattern with some variables. Data that is returned via a query is said to match the pattern. the vocabulary of SPARQL:: Graph pattern. Specifying a graph pattern, which is just RDF using some variables. Matching. When RDF data matches a specific graph pattern. Binding. When a specific value in RDF is bound to a variable in a graph pattern.## The following SPARQL query has all the major components from SPARQL: PREFIX foaf: < SELECT?name FROM < WHERE {?x foaf:name?name. } ORDER BY?name 34

35 More (slightly modified) from Let s look at each component of SPRQL in turn. The PREFIX keyword describes prefix declarations for abbreviating URIs. Without a prefix, you would have to use the entire URI in the query (< Create a prefix by using a string (foaf) to reference a part of the URI (< When you use the abbreviation (foaf:name), it appends the string after the colon (:) to the URI that is referenced by the prefix string. The SELECT keyword is the most popular of the 4 possible return clauses (more on the others later). If you ve used SQL, SELECT serves very much the same function in SPARQL, which is simply to return data matching some conditions. In particular, SELECT queries return data represented in a simple table, where each matching result is a row, and each column is the value for a specific variable. Using our SPARQL query above in which we SELECT?name, the result would be a table with one column and as many rows as match the query. The variable?x is not returned. The FROM keyword defines the RDF dataset which is being queried. There is an optional clause, FROM NAMED, which is used when you want to query a named graph. The WHERE clause specifies the query graph pattern to be matched. This is the heart of the query. A graph pattern, as mentioned above, is, in essense, RDF with variables. Finally, ORDER BY is one of the several possible solution modifiers, which are used to rearrange the query results. Other solution modifiers are LIMIT and OFFSET. 35

36 More (slightly modified) from Return Clauses In addition to SELECT, there are three other very important return clauses that you can use: ASK, DESCRIBE, and CONSTRUCT. ASK queries check if there is at least one result for a given query pattern. The result is true or false. DESCRIBE queries returns an RDF graph that describes a resource. The implementation of this return form is up to each query engine, so you won t see it used nearly as often as the other return clauses. CONSTRUCT queries returns an RDF graph that is created from a template specified as part of the query itself. That is, a new RDF graph is created by taking the results of a query pattern and filling in the values of variables that occur in the construct template. CONSTRUCT is used to transform RDF data (for example into a different graph structure and with a different vocabulary than the source data). CONSTRUCT queries are useful if you have RDF data that was automatically generated and would like to transform it using well-known vocabularies, or if you have RDF data using vocabulary from one ontology but need to translate it to another ontology. After SELECT this is the most common type of query in practice, and a major reason why agreeing on every aspect of an OWL ontology ahead of time is not necessary. Translation using CONSTRUCT is relatively cheap. 36

37 More (slightly modified) from SPARQL Protocol The SPARQL protocol enables SPARQL queries over simple HTTP requests. A SPARQL endpoint is simply a service that implements the SPARQL protocol. For example, if you do a curl on the following: curl I The response is the following: HTTP/ OK Date: Mon, 21 May :43:38 GMT Content-Type: application/sparql-results+xml; charset=utf-8 Connection: keep-alive Server: Virtuoso/ (Linux) x86_64-generic-linux-glibc25-64 VDB Content-Length: Accept-Ranges: bytes This means that SPARQL is basically an API!. 37

38 Examples of queries (using virtuoso dbpedia): please consult: Follow the instructions given at ql-nuts-bolts/ and play with the 15 examples of queries described in on this Webpage! 38

39 End of Session 1 Questions?

40