A Personal Note on the Essence of HPSG and its Rôle in Computational Linguistics
In computational linguistics Head-Driven Phrase Structure Grammar (HPSG) by Pollard and Sag has become the single most influential framework for the declarative specification of grammar. This is not only my personal impression but an observation supported by figures showing that there are more people working on and with implemented head-driven phrase structure grammars than with any other linguistic grammar model.
What are the reasons for the great impact that HPSG has had on theoretical and computational linguistics? To answer this question,we need to identify the major contributions of this framework.
HPSG is a combination of different ingredients (approaches and modules) of different status within a theory of grammar. I see it as one of the shortcomings of the two excellent and influential volumes by Pollard and Sag that the different components of the theory and their respective status are not worked out and made explicit. The following incomplete list reflects my own classification of the main components of HPSG:
1. The metatheoretical-philosophical underpinnings:
2. The underlying mathematical formalism:
3. The approaches in the architecture of the grammar model, e.g.:
4. A number of approaches/theories of language-universal status, e.g.:
5. A number of language-specific analyses, e.g.:
Analyzing the components of HPSG in this way helps us to understand the role of HPSG in contemporary research. The originators of HPSG have always made clear that they adopted central ideas and approaches from many sources. But reviewing the structured list of components of the framework, it becomes obvious that HPSG is very different from GB, GPSG, LFG and other contemporary grammar models in architecture as well as formal and linguistic content.
More than in any other grammatical framework, linguistic generalizations can be isolated in individually stated, yet uniformily represented principles. Moreover, grammar model, universal grammar and particular grammars are expressed in a uniform powerful formalism. If one understands the built-in modular design of HPSG, one also understands why the framework can serve as the basis for theoretical and computational research even if certain language-particular or language-universal linguistic theories and analyses might not be adopted or later changed. In my opinion, HPSG should rather be viewed as a construction kit for formal linguistic theoriesnot just grammarsthat already comes with an initial set of serious theories and analyses.
The modular design of HPSG offers a large degree of flexibility for...
On the other hand, these multiple dimensions of freedom make it difficult to state what the essential properties of HPSG grammar models, implementations of HPSG formalisms, or concrete HPSG grammars for specific languages are. How many and which properties of the original HPSG must the model, formalism, or grammar possess to be called HPSG? It is obvious that there cannot be a formalizable answer to this question. There are many proposals now that involve considerable changes to the original framework and analyses. Even recent analyses and approaches of the originators of the theory divert from their earlier writings.We should view this constant development as an advantage. As long as the theory changes in several directions, it can evolve further. As long as it keeps evolving, it stays alive.