Feature structure

A feature structure is an abstract hierarchical structure that is used to represent linguistic properties and dependencies . Feature structures are used in some linguistic formalisms, e.g. B. in the head-driven phrase structure grammar and the lexical-functional grammar .

Characteristic structures consist of a set of pairs, each of which contains a characteristic name and an assigned value. These values, in turn, can have two different manifestations:

• They can be atomic , i.e. they can not be broken down further and represent a single value.
• But they can also be complex. In this case they are represented by a further, subordinate feature structure.

Feature structures are usually represented by attribute-value matrices . An example of a feature structure that models a person is shown here as an attribute-value matrix:

${\displaystyle {\begin{bmatrix}\mathrm {NACHNAME} &Mustermann\\\mathrm {VORNAME} &Max\\\mathrm {ADRESSE} &{\begin{bmatrix}\mathrm {STRASSE} &Hauptstr.1\\\mathrm {PLZ} &12345\\\mathrm {ORT} &Neustadt\end{bmatrix}}\end{bmatrix}}}$

The person represented has the three characteristics "surname", "first name" and "address". The first two features are atomic, the last is complex and refers to a new feature structure that describes the address more precisely by specifying "street", "zip code" and "city".

Feature structures can be compared with special relations such as subsumption with regard to their information content. They can also be unified with each other. This means that the information contained in two feature structures is combined in a new feature structure. However, if the initial structures provide information that cannot be combined, then the unification ( unification grammar ) fails.

Linguistically motivated unification

In formal and applied linguistics, unification is often used to map congruence. Since the unification is associative and commutative, one can easily handle more complicated cases of congruence, such as B. doubled objects in Macedonian .

In the sentence ја читам книгата "(I) read the book", the proclitic pronoun in gender and number is congruent with the specific noun that becomes the direct object of the verb. The (simplified) feature structures of the individual words would look like this

${\displaystyle {\begin{bmatrix}\mathrm {LEMMA} &toj\\\mathrm {POS} &pron\\\mathrm {CASE} &acc\\\mathrm {NUMBER} &sg\\\mathrm {GENDER} &fem\end{bmatrix}}}$

${\displaystyle {\begin{bmatrix}\mathrm {LEMMA} &cita\\\mathrm {POS} &verb\\\mathrm {TENSE} &pres\\\mathrm {NUMBER} &sg\\\mathrm {PERSON} &1\end{bmatrix}}}$

${\displaystyle {\begin{bmatrix}\mathrm {LEMMA} &kniga\\\mathrm {POS} &noun\\\mathrm {DEF} &1\\\mathrm {NUMBER} &sg\\\mathrm {GENDER} &fem\end{bmatrix}}}$

the following unification schemes would create the syntactic structure of the sentence:

• P.case = acc & V.obj.gender = P.gender & V.obj.number = P.number & V.obj.def = true & V.objlc = P
• V.obj = N

If one were to interpret the above equations in Prolog , for example , all syntax trees of the sentence would arise. The objcl (clitic object) attribute is not linguistically motivated, but prevents two proclitic pronouns for a direct object from being appended to the verb.

Individual evidence

1. ↑ Jan Strunk: Basic Course Syntax - Congruence and Valence with Characteristic-Value Descriptions. April 26, 2007, Syntax Basic Course - Summer Semester 2007, Ruhr University Bochum, pp. 1–35 ( Memento of the original from January 8, 2016 in the Internet Archive ) Info: The archive link was automatically inserted and not yet checked. Please check the original and archive link according to the instructions and then remove this notice.