Excercises

Excercise A: Understanding and Modifying a NIF document

A1: Finding One Way Through a NIF Document

The file data/wilde.ttl contains a smalll NIF document representing a quote from Oscar Wilde. Open it in your text editor with syntax highlighting for Turtle and try to get an overall impression of the structure described by the triples for a minute.

This auto-generated documentation page for NIF Core vocabulary contains explanation on the semantice of the occurring classes and properties.

Next, ensure that the Fuseki SPARQL Server has been started and start browsing with the LodLive Web of Data Browser by either visiting the the URL file://$MATERIALS_ROOT/tools/lodlive/app_en.html?http://sd-llod-nif-session.nlp2rdf.org/text/ow/offset_0_49 [1] or start the naviation as described in the tools page. Then you start browsing from the nif:Context of that doument in the . Expand all outgoing nif:word relations from the context and arrange the nif:word circles in a way that the linear chain of words in the sentence can be followed easily. (This might look like includes/exA-lodlive-Wilde-initial.png).

A2: Modifying a NIF Document

Now think of a little adjustment of the statement achived by adding a short (up to four words) sentence or replacing/inserting a few words in the additional sentence. For example:

One can survive everything nowadays except death. What about Two?

(Please pardon the lowest form of humour [2] and feel invited to use a creative alternative.)

Start adjusting the document so that it represents the altered/extended version. For the above example, the following table provides the character offsets:

token	begin	end
What	50	54
about	55	60
Two	61	64
?	64	65

Note

In the original document the nif:Context and the single nif:Sentence were the same string individual. This combination of roles is not applicable any more if another sentence is added.

Since one can easily creates syntax mistakes when editing Turtle manually (unless one is routined doing so), use turtle / turtle.bat from the Apache Jena Command Line Tools by the following command line invocation (with the working directory at $MATERIALS_ROOT) ./tools/jena-cmds/bin/turtle --validate data/wilde.ttl (Linux/Mac) or .\\tools\\jena-cmds\\bat\\turtle.bat --validate data\\wilde.ttl (Windows).

When there are no messages about syntax errors/warnings from the validation tool, you can stop and restart the Fuseki Server. This will allow to browser the NIF graph that will reflect the changes you applied.

Further, you should use RDFUnit to check whether your modified document still adheres to fundamental constraints of NIF documents. To do so, invoke ./tools/run-rdfunit-on-wilde.sh or .\\tools\\run-rdfunit-on-wilde.bat at $MATERIALS_ROOT in a command prompt. A successful execution will terminate with lines similar to these:

[INFO  SimpleTestExecutorMonitor] Tests run: N, Failed: F, Timeout: 0, Error: E. Individual Errors: I [INFO  ValidateCLI] Results stored in: rdfunit/results/sd-llod-nif-session.nlp2rdf.org_text_ow.extendedTestCaseResult.*

(If the execution instead fails with a stack trace, you might have submitted a malformed Turtle document - please re-validate with the turtle command to exclude this error source. If the unsuccessful execution persists, let the tutor have a look.)

Opening the file rdfunit/results/sd-llod-nif-session.nlp2rdf.org_text_ow.extendedTestCaseResult.html in your browser will give a detailed listing of errors found in the RDFUnit run and the subject resource IRIs of the violating statements. Rectify all issues and repeat the validation & check procedure above.

A solution for the modification suggested earlier can be found in ./examples-solutions/wilde-extended.ttl

[1]	$MATERIALS_ROOT here is an insertion mark for the folder where you placed the extracted contents of the materials archive.

[2]	“A pun is the lowest form of humor—when you don’t think of it first.” - Oscar Levant

Excercise B: Querying NIF data with SPARQL

For this excercise, use the Fuseki dataset brown. It contains a subset of the documents from the traditional and eminent BROWN corpus of English, converted to NIF, with Part-of-Speech-Tags (POS-Tags) represented as OWL instances from the Ontology of Linguistic Annotations project.

To get a first feel how information on the POS annotation is represented, start again using LodLive Web of Data Browser navigate the graph, for example from either

http://brown.nlp2rdf.org/linkeddata.php?t=url&f=xml&i=http://brown.nlp2rdf.org/corpus/a01.xml#offset_0_155 or http://brown.nlp2rdf.org/linkeddata.php?t=url&f=xml&i=http://brown.nlp2rdf.org/corpus/a02.xml#offset_367_443.

First nagivate to one or two word resources, from them follow nif:oliaLink and then investigate the inheritance relations (rdfs:subClassOf) starting from classes that were assigned to the OLiA tag instances.

Tip

You can inspect the classes and individuals of the BROWN tag set representation of OLiA using an HTML page generated by LODE or by using Protege opening ontologies/brown.owl.

Now it’s time to compose the first SPARQL query over this graph. Ensure that the Fuseki SPARQL Server is running and then visit http://localhost:3030/dataset.html in your browser. You can enter queried directly in the text field overed in the query view, but it is recommended to save earlier attempts and query snippets in an editor in parallel via copy & paste, as you might want to re-use and adapt earlier attempts. (The Fuseki query does not offer a query history feature.)

Tip

To look up specific details about the SPARQL syntax, definded functions etc., you can consulted in the official SPARQL 1.1 Spec from W3C [3].

B1: Finding Adjectives

As first query, retrive a sample of eight words tagged as adjectives (olia_brown:JJ). The binding in the result set should contain both the IRIs of the word resources and their sub-strings they point at.

B2: Basic Prevalence Statistics for the POS Tags

Now, find out which are the four POS tags that were assigned most frequently to the tokens occuring in the graph. You can compare your query with a version optimized for brevity in examples-solutions/most-common-brown-tags.sparql.

B3: Searching for Usage of Passive Voice

For combined usage of NIF structure and OLiA, try now to compose a query of two adjacent words that are part of a verb phrase in passive voice, i.e. we look for two adjacent words where the first one is a form of to be (the tag should carry be an instance of any sub-class of olia_brown:Be) and the second is a past participle (the tag schould be an instance of olia_brown:LexicalVerbPastParticiple).

Tip

To check for whether ?classA is a (transitive) subclass of ?classB, the Property Path feature of SPARQL is helpful.

After completing a query with the constraints described above you will notice that the result bindings also contain phrases like ‘been charged’ that are rather part of the past perfect constructions. These cases can be excluded by disallowing that the first word described in the query is preceded by a form of have. Extend the query accordingly .

Tip

The FILTER NOT EXISTS { [...] } construct of SPARQL, will be useful for the exclusions.

You can compare your (extneded) query with examples-solutions/passive-voice.sparql.

B4: Searching for Zero-Conversions

In Englisch derivations involving word class change between adjective and verb without any change in the written representation is quite common. Formulate a query that looks for candidates of evidence of such conversions in the BROWN graph, i.e. word-forms that occur both as adjective and verb.

Tip

This means we are looking for a pair of word instances that share their nif:anchorOf value.

You can compare your query with examples-solutions/adj-verb-zeroconv.sparql

After having the query that collects individual points of evidence, modify the query so that for each these candiate word forms, the number of occurrences as adjective and as verb is calculated.

Tip

Achieving this will require joined sub-queries.

You can compare your query with examples-solutions/adj-verb-zeroconv-with-stats.sparql

[3]	Disregarding the section 18 completely, which is only relevant for creating implementations of the lanugage. Parts of the language described in sections 13, 14 and 16.2 to 16.4 will not be needed for these excercises as well.