Modeling Characteristics (Triple Extraction - Part 5)
CraigTrim 110000G799 Visits (4583)
Linking verbs are about characteristics. Linking verbs do not express actions. Instead, they connect the subject of the verb to additional information about the subject.
In contrast to intranstives, linking verbs cannot end sentences, nor can they be followed immediately by adverbs. These must be followed by either nouns or adjectives, one or the other; those nouns and adjectives may be single words or multiple-word phrases. In addition, linking verbs constitute a small class of probably no more than a few dozen verbs.
Adjectives and nouns that follow linking verbs are closely associated with their subject noun phrases. An adjective that follows a linking verb generally summarizes some characteristic of the subject noun. In the example “the president looks weary”, weariness is a characteristic of the president; it's why other people perceive him to be weary (he looks weary). A noun that follows a linking verb refers to the same person or thing that a subject noun refers to. Given the example, “Silas remained an honest man”, both Silas and an honest man are the same person.
Linking vs Action Verbs
A linking verb joins the subject and predicate of a sentence. In contrast, an action verb expresses action or being.
An action verb tells the action of a sentence.
Linking verbs are sometimes harder to recognize than active verbs because they simply show that the subject exists. These verbs usually can't be demonstrated. In each sentence, the subject complement (the word which completes the meaning of the sentence) is either a noun or an adjective.
When linking verbs are used, the subject and complement are interchangeable:
This is not true for action verbs:
This extracted information is interesting. When modeling a domain, “Wildnerness” as an abstract concept (a characeteristic) might not be the first conclusion of the modeler. Nor would vinegar as a characteristic. The initial reaction of the modeler would be to characterize vinegar as a type of liquid. This is true. But vinegar is also a characteristic. Both are true.
Note that the use of “is a” has often been taken as a simple way to extract triple types:
This statement becomes:
Or in set theoretic language, given the set “Shopaholic”, Kella is a member. This is a valid approach, and there are many techniques for extracting “is-a” patterns from unstructured text
However, I believe we have more semantic precision if we use the approach I've described:
Further cases for consideration:
The white boxes represent additional classes and relationships that may be brought into the triple store. The linking between "Cat" and "Happy" is extracted from the text. The rest of the diagram is inferred from this relationship and / or additional knowledge already contained in the triple store.