next up previous contents
Next: Semantic-head-driven Generation Up: Summary Previous: Introduction

The formalism: $ \cal {R}$($ \cal {L}$)

Chapter 2 defines a formalism, called $ \cal {R}$($ \cal {L}$), which is in several respects typical for formalisms in use in computational linguistics. It can be characterized as a constraint-based formalism, where the constraints are restricted to the path-equations known from PATR II. Unless PATR II, the formalism does not prescribe that phrases are built by concatenation. Hence the formalism is comparable to pure Prolog, but instead of first-order terms, the data-structures of the formalisms are feature structures (defined by path equations). Furthermore, the formalism is defined within a very general setting, provided by the work of [32]. This provides for more powerful constraints to be added to the formalism, without affecting some of the properties of the formalism, if the necessary constraint-solving techniques are available for these more powerful constraints.

The resulting formalism, $ \cal {R}$($ \cal {L}$), is used in the thesis as a language to define grammars with, but moreover as a language to define meta-interpreters in. A grammar of $ \cal {R}$($ \cal {L}$) is a definite clause specification of the relation sign/1. A simplified clause of a grammar might read as follows:

\pr\pred
\head{\mbox{\it sign}(\mbox{\rm M})\mbox{\tt :-}}
\body{\mbox{\it sign}...
...box{\rm M} ~\mbox{\it sem} \doteq \mbox{\rm D}_{1}~ \mbox{\it sem}.}
\epred\epr
Such a clause will generally be written in matrix notation as follows:

\pr\pred
\head{\mbox{\it sign}(\avm{\mbox{\it syn}: \mbox{\rm s}\\
\mbox{\it p...
... \mbox{\rm S}\\
\mbox{\it phon}: \mbox{\rm P}_{1}-\mbox{\rm P}}).}
\epred\epr

The procedural semantics of the formalism is comparable to Prolog's proof procedure. Thus, refutation of a goal proceeds in a top-down manner. A left-most computation rule is assumed, such that the leftmost atom of a goal is reduced first. Furthermore, the search tree is traversed in a depth-first backtrack manner. Alternative proof strategies for $ \cal {R}$($ \cal {L}$) grammars are defined later as meta-interpreters in $ \cal {R}$($ \cal {L}$).

In the case we use $ \cal {R}$($ \cal {L}$) for grammars, the p-parsing problem for a path p is defined as follows. Assume the input for parsing is a feature structure with phonological representation $ \phi$ as the value of the attribute phon. The answers to the phon-parsing problem will be those compatible signs in the grammar which have $ \phi$ as the value of their phon attribute. Similarly, assume the input for generation is some feature structure with semantic representation $ \sigma$ as the value of the attribute sem. The answers to the sem-parsing problem are those compatible signs in the grammar which have sem as their semantic structure.

It is shown that in general the p-parsing problem for $ \cal {R}$($ \cal {L}$) grammars is not solvable, by showing that an undecidable problem, Post's Correspondence Problem, can be defined in a $ \cal {R}$($ \cal {L}$) grammar. It can thus be concluded that in the general case, grammars of $ \cal {R}$($ \cal {L}$) are not reversible.

In practice however, the grammars computational linguists tend to write are reversible. For this reason an important task consists in the construction of proof procedures which solve the p-parsing problem for the grammars typically encountered. This is the goal of the third and fourth chapter of the thesis. In these chapters generation- and parsing techniques are presented, which are more generally applicable than some competing techniques. Moreover these techniques are motivated linguistically in that certain head-driven and lexicalistic aspects of most modern grammars are exploited in those techniques. It is hoped that the linguistic foundation of these processing techniques leads to an increased efficiency.


next up previous contents
Next: Semantic-head-driven Generation Up: Summary Previous: Introduction
Noord G.J.M. van
1998-09-30