(English) Phrase Structure

Developing DCG's for English.

Consider the following complications

• competence - developing methods to recognize legal syntax
  vs
  performance - developing methods to generate legal syntax
• usually methods for recognition are not appropriate for language generation

  [Although Prolog programs may be reversible, they usually need to be designed for one of the two possibilities, especially for efficiency's sake.]

• most work has been done on recognition
• there is not universal agreement on the rules of grammar
• there is not universal agreement on the mechanism of language acquisition [Recall the arguments of whether the brain is born with special features for language or if it just has general features which can be exploited to rapidly acquire language.]

  Some arguments can be used to support either position

  - babies learn rapidly, so brain circuitry must be set for language formation;

  - babies learn many tasks rapidly, so brain circuitry must be adaptable

  - the structures of known languages are similar, possibly because of similar brain features geared for language

  - the structures of known languages are similar, possibly because of limitations consistent with evolution of human capabilities

• linguists search for a universal explanation of language mechanism
  • Implications - can bring heuristics to level of theoretical explanations
  • Modeling of human behaviors models can be explanatory or exhibitory. [AI aside: psychologists often complain about neural network models. How does an AI person counter these?]

Grammar School explanations:

Sentence is grouping of phrases.

Phrase is a grouping of words.

Any PS schema should (if possible) explain these groupings. E.g., a grammar's non-terminals should be the same syntactic entities which are considered to be recognized parts of speech.

An example exception is what is called a semantic grammar. Essentially semantic grammars behave like keyword and template systems. They may be almost "ad hoc" in trying to pick portions of the source language out as having special significance, regardless of their syntactic classification in the source language.

Example sentences

   * She land the space shuttle.
   * Her landed the space shuttle.

Note: an asterisk is normally used by linguists to denote sentence forms that are not syntactically correct in the language under discussion.

Each violates a "well-known" linguistic constraint.

To see what is wrong with any sentence, we are taught "tricks", one of which is to see how the sentence differs from an acceptable one with the same meaning. [Automatic parsers follow a similar device; for example, when the Java compiler "suggests" that a semicolon ';' is expected at a certain point.]

   * She land the space shuttle.
     She landed the space shuttle.

   * Her landed the space shuttle.
     She landed the space shuttle.

These "near-miss" pairs allow a parser (person or machine) to determine the parts of speech which are used incorrectly.