Accounting for Underlying Forms in HPSG Phonology

The paper aims to present approach to HPSG phonology which would account for underlying forms of phonemes. It shows some of the issues arising in monostratal analyses of phonology, and proposes a solution based on a notion of underlying representations. The approach presented, partly inspired by Optimality Theory, resolves cases of neutralisation and opacity by formulating constraints which either restrict the surface representation or relate it to the underlying form. 1 Why are underlying forms desirable in HPSG phonology? Since most work in HPSG is focused on syntax or morphology, standard representations of phonology are reduced to supplying a word with a feature of type phon, a string of symbols equivalent to the orthographical spelling or pronunciation of the word (and its particular variants). Such strings are afterwards combined in higher-level objects to form phrases. While such a simplified approach is sufficient for solving problems based around syntax, morphology, and semantics, the HPSG does show a lot of potential for expanding the phonology within its framework. One notable attempt to do so was undertaken by Bird (1995), who introduced constraint-based phonology into HPSG. The framework outlined in Constraint Based Phonology: A Computational Approach is essentially a monostratal system, where well-formedness of a particular word or phrase is decided by phonological and morphological constraints. This system allowed for linking phonology and morphology, and resolved issues by ruling out ill-formed segments and word structures. The framework proposed was based around the principles of COMPOSITIONALITY and a requirement that a framework be MONOSTRATAL. The latter meant, in simplified terms, that any phonological representation has only one level, corresponding to forms actually appearing in the surface representations, and no abstract representation is stored: An even stronger constraint than those mentioned above is the requirement that a linguistic framework be MONOSTRATAL. This means that there is only a single level of linguistic description; descriptions pertain to occurring surface forms and not to artificially constructed abstract representations. As we shall see in section 1.5.1, the requirement that a linguistic framework be monostratal is equivalent to