next up previous contents
Next: The formalism: () Up: Summary Previous: Summary


Constraint-based grammars are often used for natural language processing. One of the interesting properties of a constraint-based grammar is, that such a grammar is completely declarative: it does not enforce a specific processing regime, but allows various parsing and generation algorithms. The order of processing is independent on the result of the computation.

For that reason, such declarative grammars can, at least in principle, be used both for parsing and generation. However, from a somewhat more practical point of view, several problems must be solved once the same grammar is used for both parsing and generation. Some of these problems are the subject of this thesis. For example, it turns out that applying a `naive' processing strategy for the purpose of generation, gives rise to severe problems for grammars which are not written with the purpose of generation in mind.

The first chapter of the thesis provides the motivation for reversible grammars, and clarifies some of the fundamental issues involved. A grammar is seen as a device which defines a relation between form and meaning. Form and meaning are represented by phonological structures and semantic structures. A parsing algorithm computes for a given phonological structure its corresponding semantic structure. A generation algorithm computes the relation between form and meaning in the opposite direction.

An effectively reversible (or reversible for short) grammar is defined as a grammar which defines an effectively reversible relation (between form and meaning). A binary relation is effectively reversible if and only if it can be computed in both directions by a program which terminates for all inputs. Hence, a grammar is reversible if both parsing and generation is guaranteed to terminate (for all inputs).

Reversible grammars are interesting for linguistic, technological and psychological reasons. Linguistically it can be argued that a single language should be described by a single grammar. Furthermore, if such a single grammar is to be used as a `theory' about the language, it can be argued that it should be possible to check the predictions the theory makes about the language, and hence the grammar should be reversible.

From a language technological point of view, I argue in section 1.1 that reversible grammars make it easier to build good natural language processing systems. It is sometimes thought that grammars which are used for parsing only, can be very liberal in that such grammars may allow ungrammatical sentences (as these are not expected to occur in the input). I argue, on the contrary, that this type of overgeneration is bad, because it generally leads to false ambiguities. Hence, even if one is only interested in building a parsing system, it may be the case that a reversible grammar is a good method to obtain such a system. Moreover, it may be easier to build a single, reversible grammar, than two separate grammars.

In section 1.1 it is discussed whether humans base their language production and language understanding on a single body of grammatical knowledge. This claim would explain why humans speak the same language they understand and vice versa. It is argued that observed differences in language understanding and production may be due to differences at another level of cognitive behavior, rather than to differences in the grammatical component.

An important goal of the thesis is to improve upon existing parsing and generation techniques along the following two dimensions. Firstly, an important motivation is to extend the types of grammar for which the proposed parsing and generation techniques are applicable. Secondly, the parsing and generation techniques are motivated from a linguistic perspective. It is hoped that such a `linguistic' motivation for deduction improves the efficiency of parsing and generation as compared with non-linguistic deduction techniques.

next up previous contents
Next: The formalism: () Up: Summary Previous: Summary
Noord G.J.M. van