将复杂的句子翻译成一组SPO三元组（RDF）（可能带有具体化）

Question

Having just started out in RDF and ontology, mainly in converting unstructured text to RDF (maybe manually), I found many examples for converting simple sentences to RDF triples 1 2 3 , but not much about complex sentences (using adjectives or adverbs) 4 .

Example of complex sentence: Those kids who, with great pleasure, like chocolate ice-cream and those who like vanilla ice-cream, which of them is better?

Never mind the meaning of the sentence, what is remarkable is the use of adjectives for ice-cream (chocolate, vanilla) and liking (with pleasure), and the comparison between the two types of kids (who is better) and a question. How do we represent such sentences as a collection of SPO triple (RDF) statements - so that it can take care of adjectives, or adverbs or comparisons or other operations?

S1: Kids - Liking - (ice-cream, flavor, Chocolate)
S2: Kids - Liking - (ice-cream, flavor, Vanilla)
S3: liking with pleasure
S4: better(S1+S3, S2+S3)

Is there a general way to approach translating such statements?Where should one look for some exploration in this direction - unstructured text to n-tuples? Maybe I am not asking the right questions - in that case please let me know. Thanks!

Answer 1

Far from attempting to describe a gneric way for translating such statements I thought of how that concrete example sentence could be modelled in RDF triples.

That's my first (late night) draft:

@prefix : <https://example.org/> .

[] a :Question, :PreferenceQuestion;
   :option [ 
       a :GroupOfHumans;
       :ageGroup :young;
       :perfomingActivity [
          a :Activity, :LikingActivity;
          :mode :withPleasure;
          :object [
              :category :IceCream;
              :flavour :vanilla;
           ]
       ]
   ],
   [ 
       a :GroupOfHumans;
       :ageGroup :young;
       :perfomingActivity [
          a :Activity, :LikingActivity;
          :mode :withPleasure;
          :object [
              :category :IceCream;
              :flavour :chocolate;
           ]
       ]
   ].

This would be 23 triples using quite an expressive Question/Activity/IceCream ontology.

Answer 2

That is a very difficult question that you are asking, there is not a single way to represent "semantic" in a graphical or tree or linkage structure that is definitive standard.

Look for instance at Google syntax parse , Link Grammar , Spacy dependency parse trees or Stanford NLP parser .

See this question that is somewhat related, gives an example of link grammar parse result: How to translate syntatic parse to a dependency parse tree?

For more advanced approaches look at multinet and natural language to first-order-logic kind of approaches.

Where should one look for some exploration in this direction - unstructured text to n-tuples?

You can represent what ever you want with n-tuples or 3-tuples, the thing is to know what and how to represent.

Last thing, you can come up with you are annotations, they are tools to do it like https://prodi.gy/ . Have a look at this question that is seemingly unrelated but I give an example representation of questions that might help you achieve some task https://stackoverflow.com/a/32670572/140837

Good luck!!