On the need for a separate perception grammar Paul Boersma, University of Amsterdam October 4, 1999 This article adduces evidence for the existence of a separate grammar for phonological perception, by showing that this allows us to solve some old paradoxes and that it helps us create a more economical theory of phonology. The perception grammar is a language-specific processing system that converts acoustic forms to phonological surface structures, i.e. the perception grammar adds covert structure to the bare phonetic events. What we perceive in a speech utterance in a communicative setting tends to be quite different from what is acoustically there. The acoustic speech signal only contains phenomena that the peripheral auditory mechanism will identify as continua of loudness, periodicity, noise, and frequency spectra, and as temporal relations between these. The phonological surface structure that we perceive, however, is a much more structured representation: it may contain sequential tiers of discrete perceptual feature values (voicing, tone, vowel height, nasality, place, sonorance, frication) and their simultaneous and sequential combinations (segments), as well as hierarchical structures such as syllables and feet. Since all these structures are not directly observable, they can be called covert or hidden surface structures. It is the task of the language-specific perception grammar to construct them from the overt acoustic signal. In this article, I first give a summary of the theory of functional phonology (Boersma 1998). After a short review of the simplest activity of the perception grammar, namely categorization of continuous acoustic features into discrete perceptual feature values, I present a fundamental discussion of the role of perception in the production grammar, and show how distinguishing between articulation and perception allows us to solve the paradox between the structuralist and generative grammar models and Charles Reiss's tooth-loss paradox. I then briefly touch upon the formalization of the roles of perception in the recognition grammar and in learning. Then I address the issue of empiricism with respect to substantive content in phonology, which finally allows us to tackle a complicated activity of the perception grammar, namely sequential abstraction. In this last part, I first show that logical inconsistencies arise if the OCP is taken to reside in the production grammar; having thus established that the OCP and LCC are structure-building constraints whose appropriate place is in the perception grammar, I discuss their role in the example of the history of Dutch syllabification, in which coronals played a role opposite to that of labials and dorsals. Parts of this article were presented at GLOW (Potsdam, April 1, 1999) and at the ICPhS Satellite Meeting on Perception in Phonology (San Francisco, July 30, 1999).