The morphology and phonology of exponence

Phoevos Panagiotidis 31 Interfaces in Linguistics New Research Perspectives edited by Raffaella Folli and Christiane Ulbrich 32 Negative Indefinites by Doris Penka 33 Events, Phrases, an

Tai ngay!!! Ban co the xoa dong chu nay!!!

The Morphology and Phonology of Exponence

OX F O R D S T U D I E S I N T H E O R E T I C A L L I N G U I S T I C S

general editors: David Adger and Hagit Borer, Queen Mary, University of London;

advisory editors: Stephen Anderson, Yale University; Daniel Büring, University of California, Los Angeles; Nomi Erteschik-Shir, Ben-Gurion University; Donka Farkas, University of California, Santa Cruz; Angelika Kratzer, University of Massachusetts, Amherst; Andrew Nevins, University College London; Christopher Potts, Stanford University, Amherst; Barry Schein, University of Southern Cali- fornia; Peter Svenonius, University of Tromsø; Moira Yip, University College London.

recent titles

24 Quantification, Definiteness, and Nominalization

edited by Anastasia Giannakidou and Monika Rathert

25 The Syntax of Sentential Stress

by Arsalan Kahnemuyipour

26 Tense, Aspect, and Indexicality

by James Higginbotham

27 Lexical Semantics, Syntax, and Event Structure

edited by Malka Rappaport Hovav, Edit Doron, and Ivy Sichel

28 About the Speaker

Towards a Syntax of Indexicality

by Alessandra Giorgi

29 The Sound Patterns of Syntax

edited by Nomi Erteschik-Shir and Lisa Rochman

30 The Complementizer Phase

edited by E Phoevos Panagiotidis

31 Interfaces in Linguistics

New Research Perspectives

edited by Raffaella Folli and Christiane Ulbrich

32 Negative Indefinites

by Doris Penka

33 Events, Phrases, and Questions

by Robert Truswell

34 Dissolving Binding Theory

by Johan Rooryck and Guido Vanden Wyngaerd

35 The Logic of Pronominal Resumption

by Ash Asudeh

36 Modals and Conditionals

by Angelika Kratzer

37 The Theta System

Argument Structure at the Interface

edited by Martin Everaert, Marijana Marelj, and Tal Siloni

38 Sluicing

Cross-Linguistic Perspectives

edited by Jason Merchant and Andrew Simpson

39 Telicity, Change, and State

A Cross-Categorial View of Event Structure

edited by Violeta Demonte and Louise McNally

40 Ways of Structure Building

edited by Myriam Uribe-Etxebarria and Vidal Valmala

41 The Morphology and Phonology of Exponence

edited by Jochen Trommer

42 Count and Mass Across Languages

edited by Diane Massam

The Morphology and

Phonology of Exponence

Edited by

J O C H E N T R OM M E R

1

Great Clarendon Street, Oxford, OX2 6DP,

United Kingdom

Oxford University Press is a department of the University of Oxford.

If furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries

© editorial matter and organization Jochen Trommer 2012

© the chapters their several authors 2012

The moral rights of the authors have been asserted

First Edition published in 2012

Impression: 1

All rights reserved No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted

by law, by licence or under terms agreed with the appropriate reprographics rights organization Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above

You must not circulate this work in any other form

and you must impose this same condition on any acquirer

British Library Cataloguing in Publication Data

Printed in Great Britain by

MPG Books Group, Bodmin and King’s Lynn

Adam Albright and Eric Fuß

Birgit Alber and Sabine Arndt-Lappe

Laura J Downing and Barbara Stiebels

12 Non-concatenative morphology as epiphenomenon 427

Patrik Bye and Peter Svenonius

General Preface

The theoretical focus of this series is on the interfaces between subcomponents of thehuman grammatical system and the closely related area of the interfaces between thedifferent subdisciplines of linguistics The notion of ‘interface’ has become central

in grammatical theory (for instance, in Chomsky’s recent Minimalist Program) and

in linguistic practice: work on the interfaces between syntax and semantics, syntaxand morphology, phonology and phonetics, etc has led to a deeper understanding ofparticular linguistic phenomena and of the architecture of the linguistic component

of the mind/brain

The series covers interfaces between core components of grammar, ing syntax/morphology, syntax/semantics, syntax/phonology, syntax/pragmatics,morphology/phonology, phonology/phonetics, phonetics/speech processing, seman-tics/pragmatics, and intonation/discourse structure, as well as issues in the way thatthe systems of grammar involving these interface areas are acquired and deployed inuse (including language acquisition, language dysfunction, and language processing)

includ-It demonstrates, we hope, that proper understandings of particular linguistic nomena, languages, language groups, or inter-language variations all require reference

phe-to interfaces

The series is open to work by linguists of all theoretical persuasions and schools ofthought A main requirement is that authors should write so as to be understood bycolleagues in related subfields of linguistics and by scholars in cognate disciplines

In this volume, Jochen Trommer brings together internationally recognized ars to address fundamental issues concerning the mapping of morphosyntactic fea-tures to phonological forms (the question of exponence) The book is designed so thatimportant topics in the theory of exponence are simulataneously tackled from boththe morphosyntactic and morphophonological perspectives via chapters coauthored

schol-by experts in these respective domains The resulting synoptic view of exponence atonce highlights the theoretical importance of modularity in understanding interfacephenomena and the practical importance of collaboration between morphosyntacti-cians and morphophonologists

David Adger Hagit Borer

Notes on Contributors

Birgit Alberhas worked on various aspects of metrical theory and prosodic phology, and on the phonology, morphology, and syntax of German dialects Herrecent projects investigate the structure of linguistic typologies in the framework ofOptimality Theory She has taught at the Universities of Marburg and Trento and iscurrently teaching German linguistics at the University of Verona

mor-Adam Albright is an Associate Professor (Anshen-Chomsky Professorship inLanguage and Mind Career Development Chair) at MIT He is currently the man-

aging editor of the journal Morphology, and an associate editor of Language His

research interests include phonology, morphology, and learnability, with an sis on using computational modeling and experimental techniques to investigateissues in phonological theory His research focuses on the relation between gram-mar acquisition and diachronic change, and on modeling gradient phonologicalacceptability

empha-Sabine Arndt-Lappeis working in English linguistics at the University of Siegen,where she received her doctorate in 2005 Her main research interests lie in Englishmorphology and phonology, with a focus on both empirical work and linguistictheory (Optimality Theory, exemplar-based models of grammar) Recent publications

include English Prosodic Morphology (Springer, 2007) as well as articles on variation

in English compound stress assignment

Ricardo Bermúdez-Oterois Senior Lecturer in Linguistics and English Language

at the University of Manchester He previously held a Postdoctoral Fellowship ofthe British Academy, followed by a Lectureship in Linguistics at the University ofNewcastle upon Tyne His research focuses on the morphosyntax–phonology andphonology–phonetics interfaces, with particular attention to diachrony His recent

publications include chapters in The Cambridge Handbook of Phonology (2007), nency and Morphological Mismatches (Oxford University Press, 2007), and The Black- well Companion to Phonology (2011).

Depo-Eulàlia Bonetreceived her Ph.D in Linguistics from MIT in 1991 and since thenhas been a professor at the Universitat Autònoma de Barcelona; she is a member ofthe Centre de Lingüística Teòrica at the same university Her research has focused

on phonology and morphology, most often in relation to clitics in Catalan and otherRomance languages Her latest research, cast within Optimality Theory, deals withphonology–morphology conflicts in imperatives

Patrik Byeis currently a researcher with the Center for Advanced Study in retical Linguistics at the University of Tromsø He has published scholarly articles

Theo-on a variety of topics including the syllable structure, quantity, and stress systems ofthe Finno-Ugric Saami languages, North Germanic accentology, and phonologically

viii Notes on Contributors

conditioned allomorphy He is the editor, with Martin Krämer and Sylvia Blaho, of

Freedom of Analysis? (Mouton de Gruyter, 2007).

Paul de Lacyis an Associate Professor of Linguistics at Rutgers, the State University

of New Jersey He specializes in phonology and its interfaces with morphology andphonetics He has published work on a variety of topics, most notably markedness

He is also the editor of the Cambridge Handbook of Phonology (Cambridge University

Press, 2007)

Laura J Downinghas been a research fellow at the ZAS (Berlin) since 2001, whereshe co-leads a project on the formal marking of clause linkage Her research since herdissertation has concentrated on various aspects of the phonology of Bantu languages,especially the interface of phonology with morphology, syntax, and focus Her work

on reduplication and minimality in Bantu and other languages is synthesized in her

book, Canonical Forms in Prosodic Morphology (Oxford University Press, 2005).

Eric Fussis a professor in historical linguistics at the University of Leipzig Previously

he held positions at the University of Stuttgart and the Goethe-University Frankfurt,from which he obtained his doctoral degree in 2005 He is currently book review

editor of the journal Linguistische Berichte His research interests include syntax,

morphology, and theoretical approaches to language change

Daniel Harbour specializes in the morphosyntax and morphosemantics of lessfamiliar languages He received his Ph.D in Linguistics from MIT in 2003 and is nowReader in the Cognitive Science of Language at Queen Mary University of London.His latest research is on the foundations of phi-features and the theory of cognitiveevolution

Sharon Inkelasis Professor of Linguistics at the University of California, Berkeley,where she has taught since 1992 Inkelas received her Ph.D in Linguistics from Stan-ford University in 1989, and taught at UCLA and the University of Maryland beforemoving to Berkeley She is a veteran of three previous LSA summer institutes Inkelasconducts research into morphology, phonology, and their interface She is the author,

with Cheryl Zoll, of Reduplication: Doubling in Morphology (Cambridge University

Press, 2005)

Andrew Nevins is Reader in Linguistics at University College London He is the

author of Locality in Vowel Harmony (MIT Press 2010), co-author with Karlos Arregi

of Morphotactics (Springer, 2012), and the co-editor, with Asaf Bachrach, of tional Identity (Oxford University Press, 2008) and, with Bert Vaux, of Rules, Con- straints, and Phonological Phenomena (Oxford University Press, 2008) His overall

Inflec-research goal is to synthesize different strands of linguistic Inflec-research and seek sals and variation not only within but across the different submodules of the grammar:morphology, syntax, and phonology

univer-Barbara Stiebelsis Assistant Director at the ZAS (Berlin), where she leads a project

on clause-embedding predicates She has worked on the typology of argument ing and the typology of complement control, morphology (morpheme orders, word

link-Notes on Contributors ix

structure of particle verbs, stem allomorphy), and lexical semantics (the semanticcontribution of verbal particles/prefixes in German, the semantics of category-shiftingmorphology and German clause-embedding predicates)

Peter Svenoniusis a professor and senior researcher at the Center for AdvancedStudy in Theoretical Linguistics (CASTL) at the University of Tromsø His research ismainly in syntax and its interfaces with meaning and form Among other topics, hehas worked on adpositions, directional particles, and cross-linguistic expressions ofspace and motion, and also on microcomparison of Nordic languages and dialects

Jochen Trommeris Lecturer in Linguistics at the University of Leipzig and izes in theoretical phonology and morphology, with a special focus on micro- andmacro-variation in lesser-studied languages (Albanian, Uralic, Nilotic, Kiranti, andAlgonquian) Currently his main interests are the learning of morphological segmen-tation and meaning, the role of moras in phonology and morphology, and the residue

special-of non-concatenative morphology (polarity and subtraction)

Dieter Wunderlich, born 1937 in Rostock (Germany), studied physics in Jena,Leipzig, and Hamburg, and since 1965 linguistics in Berlin, where he wrote hisdoctoral dissertation about tense and time reference in German In 1970 he became aprofessor at the Free University of Berlin From 1973 to 2002, he was Chair in GeneralLinguistics at Düsseldorf Besides several other activities, he was the founding pres-ident of the German Society of Linguistics 1978–1980, Fellow of the Wissenschaft-skolleg Berlin 1991–2, and Speaker of the Research Unit on the Lexicon 1991–2002

List of Abbreviations and Symbols

Morphosyntactic features and glosses

acc.agr accusative-like agreement

d indirect object role (dative)/determiner

List of Abbreviations and Symbols xi

xii List of Abbreviations and Symbols

in/on interior/exterior region case

nom.agr nominative-like agreement

List of Abbreviations and Symbols xiii

wit.pst witnessed past

Phonological features and elements

xiv List of Abbreviations and Symbols

HPSG Head-Driven Phrase Structure Grammar

K oblique case/functional preposition

List of Abbreviations and Symbols xv

OCP Obligatory Contour Principle

OED The Oxford English Dictionary

RP British Received Pronunciation

SK Saba Kirchner (2010) (Bermúdez-Otero)

SLN Sign Language of the Netherlands

∗ diacritic marking syntactic word

xvi List of Abbreviations and Symbols

ˇσ light syllable

´σ primarily stressed syllable

`σ secondarily stressed syllable

< > spelling

informa-by which the presence of an abstract morpheme M (i.e a minimal bundle of

mor-phosyntactic/semantic features) might be reflected in phonological form:

1 M is realized by a string of phonemes (Halle and Marantz 1993, Stump 2001,

Trommer 2001, Wolf in press)

2 the morphosyntactic identity of M causes a neighboring abstract morpheme

Mto be realized by a special suppletive allomorph (Bobaljik 2000, McCarthy 2001, Harbour, Béjar, and Adger 2001, Ackema and Neeleman 2003)

Carstairs-3 the phonological shape of the exponent of M causes a neighboring abstract pheme Mto be realized by a special suppletive allomorph (Rubach and Booij

mor-2001, Paster 2005, Bonet, Lloret, and Mascaró 2007, Bye 2007)

4 the presence of M triggers a phonological process which is restricted to M or

a class of morphemes including M (McCarthy and Prince 1994a, 1995a, Stump

2001, Embick and Halle 2005, Inkelas and Zoll 2005, Downing 2006, Pater 2010)This volume presents a synopsis of the state of the art in research on exponence,based on a novel conception: every chapter systematically discusses a specific aspect

of exponence from the point of view of current theoretical morphology, but alsofrom a theoretical phonology perspective This is achieved either by means of joint(‘tandem’) papers by a phonologist and a morphologist (e.g Downing and Stiebels oniconicity), of single-authored chapters by experts being equally at home in both fields(e.g Nevins on dissimilation effects), and of competing chapters on the same problem,one extending the phonological perspective into the realm of morphology, and oneemploying the opposite line of attack (see the chapters by de Lacy and Wunderlich onpolarity)

The importance of this synoptic view of exponence including phonology and phology can maybe best be illustrated by a crucial set of data from Classical Ara-bic where specific lexical verb classes mark the imperfective by a high stem vowel

mor- Jochen Trommer

([i] or [u]) and the perfective by the low stem vowel [a] whereas the distribution isroughly the opposite in a further verb class (Chomsky and Halle 1968, McCarthy1979) One possible interpretation is that this is a case of morphophonological polar-ity, where phonological constraints on paradigmatic distinctness require that relatedinflectional forms differ along a fixed phonological dimension (here: vowel height)

This approach is pursued in this volume by Wunderlich’s chapter on morphological

polarity Another analysis which suggests itself is to treat these data as a case of trary morphological homophony—different verb classes accidentally choose differentmarker sets which, taken together, lead to the appearance of phonological polarity—

arbi-an account suggested in this volume by de Lacy’s chapter on morphophonological

polarity Crucially, assigning the Semitic case to pure morphology allows a morerestrictive statement of the possible instantiations for morphophonological polaritywhereas a morphophonological analysis allows us to discard a problematic case ofaccidental homophony from the agenda of theoretical morphology In the terms of

Bermúdez-Otero (this volume), one module of the grammar is in danger of becoming

the ‘waste bin’ of the other Consequently, a central topic of all chapters in this volume

is the division of labour between phonology and morphology or, put in a slightly more general way, the problem of how to restrict analytic underdetermination in a

modular architecture of the grammar

A second unique feature of this volume is that most of the chapters combine ahandbook-style survey of current approaches to exponence with substantial new and

original theoretical proposals This is maybe most obvious for the chapter by Bye and Svenonius which proposes a new general format for the structure of exponents

with far-reaching consequences for possible systems of non-concatenative ogy, but also shows up at many other points Thus the chapter by Nevins suggests

morphol-a novel substmorphol-antimorphol-ally theory-driven clmorphol-assificmorphol-ation of morphosyntmorphol-actic dissimilmorphol-ation

effects De Lacy’s analysis of apparently polar consonant mutation in DhoLuo breaks

new ground in identifying the phonological structure which triggers mutation not asfloating features as in standard analyses (Lieber 1992, Akinlabi 1996, Zoll 1996), but asdefective (featurally underspecified) segments which coalesce with consonants of thephonological base The main goal of the approach combining evaluation of the state-

of-the-art and innovative approaches is to unravel hidden trends of convergence in

current theoretical linguistics, i.e areas where different lines of research are reachingtacit or even unconscious points of consensus Let me just mention two especiallyclear examples:

The downfall of Item and Process: The position that non-concatenative morphology is the

result of genuinely morphological processes, a crucial claim in Matthews (1991) still heartedly defended in Anderson (1992), has generally been given up in the practice of currentresearch.1 Apart from mainstream theoretical phonology which has pursued the reduction

half-of non-concatenative exponence to phonological primitives at least since McCarthy (1979),paradigmatic approaches to morphology have silently abandoned this domain as an area of

1 Stump (2001: 9–10) seems to restate Matthews’s position, but does so in a way which is explicitly patible with the position of Lieber (1992) who maintains that non-concatenative morphology is basically a phonological phenomenon.

com-Introduction 

research, whereas recent work in Distributed Morphology seeks to eliminate (Siddiqi 2006) or

to restrict the role of readjustment rules—one of the technical residues of the morphologicalprocess tradition—to the insertion of diacritics which are interpreted by operations in thephonology module (Frampton 2010; see the chapter by Bye and Svenonius for more discussion)

The triumph of Realizational Morphology: All major approaches to morphology, including

applications of Optimality Theory (Grimshaw 1997, 2001, Wunderlich 2001a, Trommer 2001),

by now embrace the view that morphological spell-out is realizational, i.e involves the mappingfrom abstract features to morphosyntactically underspecified markers (Halle and Marantz1993) or spell-out rules (Stump 2001) mediated by general and genuinely morphological uni-versal and language-specific rules or constraints (e.g impoverishment rules in DM or rules ofreferral in Paradigm Function Morphology (see the chapters by Trommer and by Albright andFuß for detailed discussion)

The volume starts and ends with two large-scale programmatic texts which line general research programs for the morphology and phonology of exponence:

out-Bermúdez-Otero develops a Four-Hypothesis Program for restricting the interaction

of morphology and phonology in linguistic theory (the Morph Integrity sis, the Indirect Reference Hypothesis, the Phonetic Interpretability Hypothesis, andthe Cycle Hypothesis) This program is fleshed out by the Generalized NonlinearAffixation approach which maintains that all apparently non-concatenative mor-phology is the result of concatenating defective phonological material that provokes

Hypothe-adjustment in phonology The chapter by Bye and Svenonius—though developed

independently—can be seen as a specific implementation of this program In lar, the authors argue that, apart from introducing pieces of phonological structure asexponents, the only additional idiosyncratic information morphological markers mayspecify are requirements on their linearization The formalism is further restricted byconfining itself to a small set of potential linearization targets, structurally parallel

particu-to potential clitizization sites in morphosyntax Bermúdez-Otero also puts forward astrong hypothesis on the status of semiproductive and non-productive phonologicalalternations In particular, he argues that these are confined to the stem-level in

a cyclic architecture of morphology and phonology and follow from the effects ofmorphological listing which may interfere with morphology in providing input tophonological evaluation at the stem level, but not in later strata of phonology

The chapter by Downing and Stiebels addresses one of the most fundamental and

intricate aspects of exponence, iconicity Although, as has often been emphasized,modern linguistics is built on the assumption that language is essentially non-iconic,Downing and Stiebels show that a number of current approaches in formal grammarinvoke iconicity principles of a very abstract type Thus work in Generalized-TemplateTheory (McCarthy and Prince 1994a, 1995a, Downing 2006) employs general ways tolink the category of a morphological template (e.g stems vs affixes) to its (language-independent) canonical size, whereas work on affix order assumes that the linearorder of morphemes typically reflects semantic scope (cf e.g the Mirror Principle,Baker 1985) Downing and Stiebels give a critical evaluation of these approachesand the relevant phenomena, but argue also that recent attempts to reduce iconicityphenomena fully to frequency of use (see Haspelmath 2008) are not without problems

 Jochen Trommer

The remaining chapters discuss more specific theoretical areas of exponence, lowing roughly the following taxonomy which is based on the idea that virtually allmajor phenomena in exponence fall out of a cross-classification of basic ontological

fol-categories such as identity and zero with the syntagmatic and paradigmatic

dimen-sions of language

Syntagmatic Morphosyntactic Dissimilation NevinsNon-Identity of Exponence

de LacyWunderlichParadigmatic Polarity

Bonet &HarbourSyntagmatic Allomorphy

FußParadigmatic (Directional) Syncretism

Alber &Arndt-LappeTruncation

Syntagmatic

Paradigmatic Paradigmatic Gaps

Identity of Exponence

Albright &FußParadigmatic Syncretism

For example morphological dissimilation effects result if grammars require identity between syntagmatically related elements (identical morphemes may notcooccur adjacent to each other in the same word form), whereas polarity (say reversal

non-of tone from High to Low and vice versa in nominal plural formation) involvesnon-identity of paradigmatically related forms In directional syncretism, a paradigmcell ‘parasitizes’ another while allomorphy represents syntagmatic parasitic expo-nence since the trigger of the allomorphy parasitizes the exponence of another mor-pheme by conditioning the latter’s realization in a way that indirectly signals its own

Introduction 

presence Paradigmatic gaps (i.e., the nonexistence of forms predicted by matic structure) are the paradigmatic counterpart of syntagmatic zero exponence inzero affixation Truncation (including subtractive morphology) is treated here as aborderline case between zero exponence and paradigmatic parasitic exponence Notethat directional syncretism is discussed in the general chapter on syncretism, and thatparadigmatic gaps are not represented by a separate chapter

paradig-The chapter by Nevins argues, based on a detailed survey of empirical cases

dis-cussed in the literature, that identity avoidance in exponence (e.g the ban on twoadjacent identical clitics in many Romance languages) is by no means a unitaryphenomenon, but occurs in four markedly distinct varieties These varieties differnot only in their sensitivity to phonological and/or morphosyntactic features, and thelocality domains in which they apply, but also in the repair operations they trigger, andcorrespond to different derivational modules of the spell-out process Thus Nevins

makes a case for modular parallelism, a recurring topic in the entire volume.

The contributions by de Lacy and Wunderlich treat a topic which has only recently

become prominent in the literature: polarity (e.g Baerman 2007) De Lacy treatsmorphophonological polarity, i.e instances of a morphological category which isexpressed by the systematic reversal of a phonological feature, whereas Wunderlichfocuses on purely morphological polarity, i.e inflectional exponence where the featurecombinations [+F –G] and [–F +G] (for any morphosyntactic features F and G) areexpressed by the same marker and the marker for [+F +G] is also syncretic to theone for [–F –G], but distinct from the exponent of the reversed (‘polar’) feature com-binations De Lacy shows that none of the alleged examples of morphophonologicalpolarity discussed in the literature provides a stringent argument for the existence ofthe phenomenon since the only well-understood case of apparent polarity—voicingalternation in Dholuo—does not exhibit proper polarity, whereas other examples

of morphophonological polarity are ambiguous (i.e offer themselves to alternative

analyses) or not sufficiently documented to allow a decisive evaluation Wunderlich

comes to almost the same conclusion for morphological polarity, arguing that propermorphological polarity has only been found in one set of data, the number-casemorphology of Old French, where it has emerged as a last-resort device to maintain

a complex set of contrasts when sound changes had effectively reduced the marker

inventory to a single formative (-s).

The theoretical significance of morphological polarity is closely linked to theassumption that syncretism is in some sense systematic, hence not reducible to acci-dental homophony of exponents The evidence for this assumption is one of the

central topics of the chapter by Albright and Fuß on syncretism They show that there

is substantial morphology-internal, but also syntactic, psycholinguistic, and historicalevidence that many instances of syncretism are cognitively real Based on this premise,they develop a detailed overview of the crosslinguistic variations in syncretism andthe theoretical tools to capture them, concluding that simple underspecification isnot sufficient to capture all cases of systematic syncretism

The chapter by Bonet and Harbour addresses another central area of exponence, contextual allomorphy Bonet and Harbour show that the phenomenon is much

more widespread than usually assumed, and disentangle scrutinously morphological,

 Jochen Trommer

syntactic, and phonological factors which all play a crucial role in triggering ent types of allomorphy The chapter reaches two major conclusions on the state ofthe art in allomorphy: first, important restrictions on allomorphy which have beenproposed in the literature are empirically untenable—especially different versions ofParadigm Economy conditions (Carstairs-McCarthy 1991, 1994, Cameron-Faulknerand Carstairs-McCarthy 2000) and the claim that lexical root morphemes are uni-versally exempt to suppletive allomorphy (Harley and Noyer 1998, Embick and Halle

differ-2005, Embick 2010) Second, other theoretically crucial questions such as the possibledirectionality conditions for allomorphy and the possibility of long-distance allomor-phy are seriously understudied, and await more thorough empirical and theoreticalinvestigation

Similar conclusions are reached by Trommer in the chapter on Ø-exponence

which shows that similar forms of Ø-exponence are pervasively assumed in mostcurrent approaches to morphological exponence Thus even Word-and-Paradigmapproaches which decidedly reject the concept of Ø-affixation (Matthews 1991,Anderson 1992, Stump 2001) standardly invoke zero exponence rules (i.e rules notintroducing phonological material but blocking other exponence rules) which are inmany respects similar to Ø-affixes After developing a typology of different technicaldevices to achieve Ø-exponence, the chapter provides an extensive discussion of pos-sible restrictions on Ø-exponence which follow from different existing formalisms ormay be independently stipulated, concluding that most of these restrictions are eitherempirically problematic or have so far not undergone systematic evaluation

Alber and Arndt-Lappe’s chapter achieves both a rehabilitation and a unified

account of truncatory exponence They provide detailed evidence that the treatment

of truncation as an ‘extragrammatical’ and unpredictable process without propergrammatical meaning and function is ill-founded, and show that both subtractivetruncation (often called “subtractive morphology” in the literature) and templatictruncation share important formal properties Most crucially, in both cases, the trun-cation is carried out with respect to a prosodic size restriction, determining in tem-platic truncation the shape of the phonological material remaining after truncation,and defining in subtractive morphology the material which is ‘removed’ from the base.Alber and Arndt-Lappe give a detailed crosslinguistic survey of attested truncationphenomena identifying common and more marginal patterns, and a critical evalu-ation of current approaches to truncation in the Prosodic Morphology (McCarthyand Prince 1996) and Generalized Template (McCarthy and Prince 1994a, 1995a)tradition, considering especially the problem of patterns generable by current imple-mentations which are not actually attested empirically

The second big area of exponence which is in central respects governed by prosody

is reduplication, which has seen a rush of important new theoretical proposals in thelast fifteen years (McCarthy and Prince 1994a, 1995a, Raimy 2000a, b, Frampton 2010,

Struijke 2000, Inkelas and Zoll 2005) The chapter by Inkelas gives a compressed

sur-vey of the relevant data and the central current approaches, focusing especially on thedivision of labour between morphology and phonology in this area and the intricateways in which reduplication specifies the morphological basis which it targets forcopying

Introduction 

Thus the contributions to this volume converge on the need to divide the labour

of exponence between morphology and phonology in a highly modular way ever, this conclusion does not license the pursuit of morphological and phonologicalresearch in isolation from each other On the contrary, it requires researchers who mayself-identify primarily as morphologists or primarily as phonologists to work in closecollaboration The goal of this volume is to provide a further step in this direction.The conception of this volume has been born and fostered in the context of theinternational scientific network “Core Mechanisms of Exponence”, which has beengenerously funded by the DFG (German Research Foundation, project number TR521-2-1) I wish to thank all members of the network, especially Jonathan Bobaljik,Gereon Müller, Marc van Oostendorp, and Eva Zimmermann for invaluable inspira-tion and discussion

The architecture of grammar and

the division of labor in exponence

R IC A R D O BE R M Ú DE Z - OT E RO

Languages commonly exhibit alternations governed by complex combinations ofphonological, morphological, and lexical factors An alternation of this sort will oftenadmit a wide variety of analyses, each apportioning different roles to lexical storageand to morphological and phonological computation Such analytic underdetermi-nation poses a threat to falsifiability and to learnability: hypotheses can easily evadeempirical disconfirmation if potential counterexamples can be redescribed in manydifferent ways to suit the linguist’s convenience, and so theories risk losing empiricalcontent; by the same token, it becomes hard to explain how, among a plethora ofchoices, learners converge upon the target grammar (§2.2) To avert these dangers,the theory of grammar must set limits to the space of possible interactions betweenphonology, morphology, and the lexicon: in particular, it must ascertain the properdivision of labor between storage and computation (§2.3), and it must constrain theways in which morphological operations can manipulate phonological material and

in which phonological processes can refer to morphosyntactic information (§2.4).Concerning the question of storage vs computation, this chapter pursues thehypothesis that different types of alternation reflect different modes of interactionbetween the lexicon and the grammar This idea is fleshed out by means of a refineddual-route approach to exponence (§2.3.1, §2.3.5), in which the well-establisheddistinction between explicit symbolic generalization and implicit pattern associa-tion (§2.3.1, §2.3.4) is supplemented with a further distinction between two types

of lexical listing, analytic and nonanalytic (§2.3.1, §2.3.3.1), akin to Clahsen andNeubauer’s (2010: 2634) contrast between ‘combinatorial entries’ and ‘unanalyzed

1 I am specially indebted to Sarah Collie, Jesse Saba Kirchner, Andrew Koontz-Garboden, Tobias Scheer, and Jochen Trommer for inspiration in the writing of this chapter: their influence will be readily apparent

to all readers familiar with their work Section 2.3.3 draws on research previously presented at meetings in Manchester, Leipzig, Groningen, and Warsaw: I am grateful to the audiences on all these occasions for their comments and suggestions The chapter has benefited from careful scrutiny by Patrik Bye, Tobias Scheer, and Jochen Trommer; all errors and infelicities remain my own.

Architecture of grammar and division of labor 

entries’ (see also Stemberger and MacWhinney 1986, 1988) Assuming a stratal sion of Optimality Theory (OT), I show that the peculiar syndrome of propertiescharacteristic of stem-level morphophonology arises from the fact that stem-levelforms are stored nonanalytically but stem-level processes are nonetheless explicitlyrepresented in the grammar by means of symbolic generalizations, whose statusresembles that of Jackendoff ’s (1975) lexical redundancy rules (§2.3.3.1) The modelprovides a highly explanatory account of internal cyclic effects in stem-level domains,

ver-which I illustrate with classic examples such as English órigin ∼ oríginal ∼ orìginálity (§2.3.3.2) and còmp[ e ]nsátion vs cònd[`ε]nsátion (§2.3.3.3).

On the issue of the interaction between morphology and phonology, this chapterargues for the adoption of a restrictive stance based on general cognitive princi-ples of modularity and locality (§2.4.1) A program is proposed consisting of fourhypotheses (37): that morphology selects and concatenates morphs without everaltering their phonological content (§2.4.2); that phonological constraints other thanthose on prosodic alignment may not refer to morphosyntactic information (§2.4.3);that output phonological representations do not contain diacritics of morphosyn-tactic affiliation (§2.4.4); and that morphosyntactic conditioning in phonology issubject to cyclic locality (§2.4.4) These hypotheses will provide the guiding threadfor an evaluation of several mechanisms currently used to describe morphologicallyconditioned phonological processes, including construction-specific cophonologies(§2.4.2.3), indexed constraints (§2.4.3), and readjustment rules (§2.4.3) The balance

of argument supports a stratal-cyclic architecture for phonology—one, however, inwhich neither cyclicity nor stratification are innately stipulated, but both emerge fromfundamental storage and processing mechanisms (§2.3.3.2, §2.3.3.3) and from timingeffects in the child’s linguistic development (§2.4.2.3)

2.2 Analytic underdetermination

A morphologically conditioned phonological alternation of no more than nary complexity will often be compatible with several sharply different grammaticaldescriptions, all of which may succeed in covering at least the central facts As arepresentative example, this section examines a relatively straightforward alternationaffecting second-conjugation theme vowels in Spanish deverbal derivation We shallsee that, although the phenomena look simple, current linguistic theory offers a sur-prisingly wide range of alternative analyses The existence of such a large space ofconceivable grammars forces us to ask what analytic biases guide the learner’s choice

ordi-of generalizations

Spanish verbs are arbitrarily divided into three inflectional classes or ‘conjugations’,each associated with its own theme vowel: [-a-] for the first conjugation, [-e-] for thesecond, and [-i-] for the third; see (1a) In nouns and adjectives derived from verbstems, the theme vowel of the base undergoes deletion before vowel-initial suffixes,but remains if the following suffix begins with a consonant (Rainer 1993: 95, 96,Pena 1999: 4337, Bermúdez-Otero 2007b) For our purposes, the interesting fact isthat second-conjugation verb stems surface with [-i-], instead of the expected [-e-],

 Ricardo Bermúdez-Otero

before certain suffixes like [-B

fll-e], cognate with English -able/-ible, and [-"mjen”t”-o],

cognate with English-ment (Rainer 1993: 95, 1999: 4609–10) This substitution of [-i-]

for [-e-] also occurs in certain inflectional forms, such as the past participle.2(1) 1st -conjugation base 2nd -conjugation base 3rd -conjugation base

a aðflmi"R-a-R ‘admire.inf’ be"B

fl-e-R ‘drink.inf’ su"fR-i-R ‘suffer’

b aðflmiR-a-"ðfloR-Ø ‘admirer’ beBfl-e-"ðfloR-Ø ‘drinker’ sufR-i-"ðfloR-Ø ‘sufferer’

c aðflmi"R-a-B

fll-e ‘admirable’ be"Bfl-i-Bfll-e ‘drinkable’ su"fR-i-Bfll-e ‘sufferable’

d aðflmi" R-a-ðfl-o ‘admire.ptcp’ be"Bfl-i-ðfl-o ‘drink.ptcp’ su"fR-i-ðfl-o ‘suffer.ptcp’

The single fact of the alternation between [-e-] in bebedor ‘drinker’ and [-i-] in bebible

‘drinkable’ could conceivably be described in many different ways, surveyed in thefollowing paragraphs

One might begin by assuming that, at a relatively deep level of representation,

the stem of beber bears the theme vowel /-e-/ in both bebedor and bebible: i.e.

V√bebThe or, in “inferential” (Stump 2001: 1) or “amorphous” (Anderson 1992)theories, just /bebe/.3This representation could be the result of an operation of morphinsertion, possibly conditioned by a conjugation diacritic (2a), as in Distributed Mor-phology (Halle and Marantz 1993); or it could be generated by a realizational rule

of stem formation, again possibly referring to inflectional class features (2b), as inframeworks like Amorphous Morphology (Anderson 1992) or Paradigm Function

Morphology (Stump 2001); or it could simply be stored in the lexical entry of beber

(2c), as in lexicalist models like Lieber’s (1980: ch 1)

(2) a Th↔ -e- / [II] (see Embick 2010: 76)

b <[V, class II], (V → Ve)> (see Aronoff 1994: 68)

c beber↔ √beb The (see Bermúdez-Otero 2013)

On this assumption, the theme vowel of the verbal base of bebible must at some point

undergo a mapping that alters its phonological content along the lines of (3)—orits equivalent in the reader’s preferred theory of distinctive features for vow-els, adjusted according to the reader’s preferred assumptions concerning featural(under)specification

(3) [-high, –low, –back]→[+high, –low, –back]

Different frameworks would allow this process to take place in different places inthe grammar In inferential theories, where all exponence is driven by realizationrules, the transformation of the theme vowel could be included in the statement of

a realization rule for ble (4a); in this scenario, the environment for (3) is definedmorphologically Similarly, Distributed Morphology, though a lexical rather than an

2 Spanish verbs exhibit extensive theme-vowel allomorphy under inflection, especially in the second and third conjugations Roca (2010) provides comprehensive coverage of the inflectional facts, but does not address derivation.

3 I use hollow brackets to mark the edges of morphosyntactic constituents and of phonological cyclic domains I reserve solid square brackets for phonological surface representations, feature matrices, and the edges of prosodic units.

Architecture of grammar and division of labor 

inferential theory in Stump’s (2001: 1) terms, countenances readjustment rules (Halleand Marantz 1993: 127–9, Embick and Halle 2005), which directly alter the phonolog-ical content of morphs in morphosyntactically defined contexts, as in (4b); see further

§2.4.3 below

(4) a <[V, ble], (Ve →Vible)>

b /e/→ /i/ / Th,   a,-bl-

Alternatively, the effects of (3) could be achieved by phonological means One bility would be to muster the autosegmental machinery used to describe mutation

possi-in frameworks where exponence is strictly piece-based (e.g Lieber 1992: 165–71,Stonham 1994: ch 5, Wolf 2007, Hermans and van Oostendorp 2008, Trommer 2008c,

2011, Bye and Svenonius this volume, and see §2.4.2.2 below) For example, one couldposit an initial floating [+high] feature in the underlying representation of ble andmiento, docking onto the nearest vowel on the left.4Appropriate provisions in thephonology would prevent the floating autosegment from docking onto a low vowel,thus protecting first-conjugation stems; docking onto third-conjugation theme vowelswould be implemented vacuously

(6) a b e b - V - b l - e [be i le]

[+hi]

b a d m i - V - b l - e [a mi a le]

[+lo] [+hi] [+hi] docking blocked because ∗ [+high, +low]

Yet another option would be to assume that the floating [+high] feature is packagedwith the second-conjugation theme itself, rather than with ble or miento, so thatthese suffixes merely trigger an association rule that enables the floating feature todock onto its sponsor: see Wiese’s (1996) analysis of umlaut in present-day German

4 It is interesting to note that ble has an allomorph [-Bflil-], with a high vowel, which occurs before other

derivational suffixes: e.g [sost”e"n-e-R] ‘sustain.inf’ ∼ [sost”e"n-i-Bfll-e] ‘sustainable’∼ [sost”en-i-Bfli"l-iðflaðfl -Ø]

‘sustainability.’ Independent evidence for the presence of the feature [+high] is however lacking in other environments where thematic [-e-] is replaced with [-i-]: see note 2.

 Ricardo Bermúdez-Otero

In the various scenarios we have considered so far, the base of bebible is represented

as /bebe/ at a relatively deep level in the grammar and is mapped onto /bebi/ either

by a morphological realization rule or by a phonological process; but it is of course

possible to imagine derivations of bebible in which /bebe/ never occurs as either an

initial or an intermediate representation For example, we could assume that, from theoutset, the alternants /-e-/ and /-i-/ compete with each other in the morphology and

that, in the case of bebible, /-i-/ wins, blocking /-e-/ As we have already seen, different

theories of morphology will implement this competition in different ways, even if

we agree that /-i-/ has the more specific distribution and /-e-/ is the default: based theories in which lexical storage is restricted to roots and exponents of singlefunctional heads, like mainstream Distributed Morphology (cf Siddiqi 2009), willposit two underlying theme-vowel allomorphs, /-e-/ and /-i-/, competing for insertion

piece-in the Th position; theories allowpiece-ing stem storage (Lieber 1980: ch 1) might posit

two listed stems for beber, /bebe-/ and /bebi-/, again competing for insertion; andprocess-based theories of morphology would set up two realization rules of second-conjugation stem formation, competing for application Whether the contenders inthis morphological competition are pieces (of whatever size) or rules, the contextualrestrictions on /-i-/ could in principle be stated either morphologically or phonologi-cally, for it is a well-established fact that allomorphs can arbitrarily subcategorize notonly for morphological properties in their environment, but also for phonologicalones (e.g Kiparsky 1994: 19, Carstairs-McCarthy 1998b, Paster 2006, 2009, Bye 2007,Nevins 2011b, and see §2.4.2.1 below) In the case at hand, however, the option ofimposing a phonological subcategorization frame on /-i-/ may seem a little far-fetched(though not impossible): borrowing the autosegmental machinery of (5), /-i-/ could

be restricted to environments where a floating [+high] is present, but it would perhaps

be unsatisfactory if this autosegment then passed inertly through the phonology on itsway to Stray Erasure In contrast, an analysis relying on morphologically defined sub-categorization statements gives us yet another choice: whether to impose a contextualrestriction on /-i-/ or rather on ble and miento

The possibilities do not end there: notably, one could, if one wished to, find a way

to prop up the suggestion that two second-conjugation theme-vowel allomorphs,

/-e-/ and /-i-/, compete for selection and that, in the case of bebible, it is a floating

[+high] feature affiliated with ble that causes /-i-/ to win This could be done simply

by letting the choice of allomorph take place in the phonology, rather than in themorphology (e.g Tranel 1996, 1998, Kager 1996, 2009: 420ff., Mascaró 1996, 2007,Rubach and Booij 2001, Bonet and Harbour this volume) In this scenario, the input

to the phonology looks like the following:

(7) /beb-{e,i}-[+hi]b1-e/

If a high-ranking featural faithfulness constraint indexed to stems, [high], dominates the faithfulness constraint demanding the preservation of thefloating autosegment affiliated with ble, namely Max(affix)-[high], then the float-ing feature will dock only when it can do so without changing the height of thetheme vowel; cf the ranking metaconstraint Faith(root) Faith(affix) postulated

Ident(stem)-by McCarthy and Prince (1995a: 364) With a first-conjugation stem (8a), faithful

Architecture of grammar and division of labor 

docking is impossible If the stem belongs to the second conjugation (8b), however,the floating feature of ble can survive by coalescing with the input specification of

the /-i-/ allomorph of the theme vowel In the case of bebedor ‘drinker,’ we need to

assume that the agentive suffix dor does not carry a floating [+high] feature Thisgranted, /-e-/ could be made to win over /-i-/ in at least two different ways Since [e]

is the epenthetic vowel of Spanish, we can assume that it is the least marked underthe phonological constraint hierarchy of the language; if so, the selection of /-e-/ in

bebedor could be treated as an emergence-of-the-unmarked effect (McCarthy and

1

∗!

bebible

| [+hi]

2,3

∗

(c) [stembeb-{ e , i} ][affix do]

| |[ hi]

2

bebido

| [+hi]

2

∗!

bebedo

| [ hi]

1

Alternatively, the theory of phonologically driven allomorph selection developed inBonet (2004: 90ff.) and Bonet, Lloret, and Mascaró (2007) allows the input to thephonology to carry the stipulation that the /-e-/ allomorph is preferred to the /-i-/allomorph: cf (7) and (9a)

(9) a /beb-{e i}-[+hi]bl-e/

b /beb-{e i}-doR/

 Ricardo Bermúdez-Otero

In this approach, the right results follow from ranking Max(affix)-[high] above theconstraint Priority, which favors obedience to the allomorphic preferences stipu-lated in the input

This enumeration could easily be extended, but the point should by now be clear:morphologically conditioned phonological alternations pose a staggering problem

of analytic underdetermination I am of course not suggesting that all the analysessurveyed above are equally plausible, and I admit that, in at least a few cases, con-sidering a wider range of Spanish data would in all likelihood tilt the scales in onedirection or the other; but I shall not declare my own preferences here For my currentpurposes, it is rather more significant to note that all these different descriptions ofthe same facts rely on ideas advocated by current schools of generative grammarand that many of the relevant devices are regarded by their proponents as mutuallycompatible Indeed, looking at (1) from a nongenerativist viewpoint (e.g connec-tionist or exemplar-based)5would open up yet another cornucopia of descriptions.The problem of underdetermination becomes further compounded if, as seems likely,several of the grammars we have sketched turn out to have exactly the same weak gen-erative capacity (Chomsky 1963): presumably, each grammar will still have differentconsequences for acquisition and processing, and it will certainly instantiate a theorywith its own typological implications, but a linguist will not be able to choose betweenthese grammars by looking at the facts of Spanish alone

Going one step further, we can take our own predicament as linguists as indicative

of the immense obstacle that morphologically conditioned phonological alternationswould pose for the learner if she had to contend with an equally vast and shapelessspace of possible grammars.6 In this sense, we have come up against an instance ofthe logical problem of language acquisition (Baker and McCarthy 1981), which isarguably nothing other than the problem of induction (Hume 1748[2000]: section 4,Popper 1935[1959]: ch 1, §1) as faced by the language-learning child (Pinker 2004:

949, Bermúdez-Otero and McMahon 2006: 550) In this light it seems advisable

to impose a priori biases on the space of possible hypotheses about morphology–

phonology interactions: these are essential to guarantee learnability, for everyoneagrees that language learning is impossible without some analytic bias (Lappin andShieber 2007: 394–5); but, in addition, the postulation of analytic biases is one keysource of typological predictions in linguistics, models of channel bias providing theother key source (Moreton 2008)

Taking a broad view of the problem, one should expect an adequate set of analyticbiases to inform one’s answers to two basic questions First, in any instance of alterna-tion, which aspects (if any) reflect lexical storage of allomorphs, and which (if any)involve the generation of positional variants by computational processes? Second,which computations take place in the morphological component of the grammar,and which are carried out in the phonology? In this chapter, section 2.3 addressesthe first question; section 2.4, the second In section 2.3, I survey the debate on

5 For the distinction between connectionist and exemplar-based models, see Chandler (2010: 375–6).

6 Shapelessness, rather than size, is the major problem: when the space of possible grammars grows too

large for exhaustive search, learnability becomes dependent on the structure of the space, rather than on

its size (Tesar and Smolensky 2000: 2–3).

Architecture of grammar and division of labor 

the division of labor between storage and computation since SPE (Chomsky and

Halle 1968) In section 2.4, I consider a set of hypotheses concerning the ability ofmorphology to manipulate phonological exponents and the ability of phonology torefer to morphosyntactic information Throughout the discussion we shall see thatlinguists’ choices of analytic biases are typically guided by very general programmaticassumptions about human cognition on key questions such as the nature of memory,the modularity of mind, and the locality of computation

2.3 Storage vs computation

2.3.1 Preview: modes of interaction between the lexicon and the grammar

We begin by considering the relative roles of the lexicon and of the grammar larly the phonology) in determining the form of linguistic exponents If linguistics is

(particu-a br(particu-anch of cognitive science, then this inquiry need concern itself only with thosepatterns of exponence that show evidence of psychological reality To beg as fewquestions as possible, however, our operational definition of cognitive reality should

be appropriately liberal: we shall say, therefore, that a morphophonological pattern ispsychologically real if speakers extend it to new items—even if only sporadically—whether in language change or in experimental tasks such as wug tests (Berko 1958).The application of such tests has reached an impressive level of refinement in recentyears, enabling us to distinguish with growing precision between accidental patternsand linguistically significant generalizations (e.g Becker, Ketrez, and Nevins 2011).This research has thrown up some challenging results, which suggest that nativespeakers can acquire knowledge of remarkably subtle and apparently unmotivatedstatistical trends (e.g Hayes et al 2009)

By our operational definition of psychological reality, then, linguistic theory mustanswer for a highly heterogeneous set of morphophonological phenomena differingvery widely in productivity The well-studied case of Modern English past-tense mark-ing provides examples of generalizations lying far apart on this spectrum: notably, thealternation shown in (10a) sustains no exceptions (Albright and Hayes 2003: 151)and applies automatically to any new verb inflected by means of the suffix /-d/; incontrast, ablaut patterns like those in (10b) are seldom extended to new verbs, thoughtheir psychological reality cannot be denied precisely because novel strong forms do

arise from time to time in diachronic change (e.g snuck: see §2.3.2 below) and, with

varying frequencies, in nonce-probe experiments (Bybee and Moder 1983, Prasadaand Pinker 1993, Albright and Hayes 2003)

(10) a /-d/: [-id] after /t, d/ ["pæt-id, "æd-id]

[-t] after voiceless segmentsother than /t/

[tæp-t, pæk-t, k e "læps-t, pætS-t][-d] elsewhere [dæb-d, sæg-d, bæn-d]

b dôaIv∼dô e Uv, ôaIt∼ô e Ut, ôaIz∼ô e Uz

dôIŋk∼dôæŋk, swIm∼swæm, sIt∼sæt

bl e U∼blu:, gô e U∼gôu:

etc

 Ricardo Bermúdez-Otero

One is therefore led to ask: do such apparently disparate patterns call for essentiallydifferent accounts, each striking its own specific balance between the roles of thelexicon and of the grammar?

Chomsky and Halle (1968) answered this question largely in the negative: theyasserted that any pattern of alternation exhibiting any degree of psychological realitymust be expressed as a standard rule, i.e as a rule that applies whenever its inputsatisfies its structural description As we shall see in section 2.3.2, this aspect of

SPE’s doctrine reflected Chomsky and Halle’s adherence to an important strand of

structuralist thought: Bloomfield’s (1933: 274) view of the lexicon as a cognitively inertlist (Chomsky 1965: ch 2, endnote 16) This conception of the lexicon left grammaticalrules, and in particular the type of nonstochastic rule available at the time, as the onlydevice capable of expressing cognitively real patterns of alternation In consequence,

SPE and all the single-route theories of exponence inspired by it handle Modern

English strong verbs by means of mandatory ablaut rules (Chomsky and Halle 1968:

11, Halle and Mohanan 1985: 107–14, Halle and Marantz 1993: 127–9); insofar assuch rules operate upon phonological features but must refer to arbitrary diacritics orlexical lists, they are incompatible with a modular approach to the interfaces betweensyntax, morphology, and phonology, the subject of section 2.4 of this chapter (see

§2.4.3 and §2.4.4.) In section 2.3.2 I further recall how SPE’s assumptions favored

the opportunistic use of underlying specifications and of extrinsic rule ordering forthe purposes of subsuming lexically idiosyncratic patterns of alternation under inde-pendently motivated phonological rules This, in turn, forced Chomsky and Halle

to renounce the goal of devising unsupervised learning algorithms for phonologicalrule systems

In contrast with SPE, this chapter pursues the hypothesis that alternations of

differ-ent types reflect differdiffer-ent modes of interaction between the lexicon and the grammar.The specific proposal outlined below draws upon ideas from a variety of sources: mostimmediately,

(i) Jackendoff ’s (1975) theory of lexical redundancy rules,

(ii) the phonological architectures developed in Lexical Phonology (Kiparsky1982a, b) and Stratal OT (Bermúdez-Otero 1999, Kiparsky 2000), and

(iii) the dual-route approach to exponence (Prasada and Pinker 1993, Clahsen 1999,Pinker 1999, Ullman 2001, Pinker and Ullman 2002)

Of course, each of these sources has a rich and complex history of its own: both (ii)and (iii), for example, may be regarded as pursuing insights first expressed in thetaxonomies of alternation of the Kazan School (Kruszewski 1881[1995], Baudouin

de Courtenay 1895[1972]) This section provides an overview of my proposal; moredetailed arguments are laid out in sections 2.3.3, 2.3.4, and 2.3.5, where I have made

a sustained effort to emphasize the intellectual lineage of each idea

In line with dual-route frameworks, then, I reject single-mechanism connectionistand exemplar-based models of exponence (e.g Rumelhart et al 1986, Chandler 2010);

I assert instead that some morphophonological patterns are explicitly represented

in the grammar by means of rules in the broad sense, i.e by means of symbolic

Architecture of grammar and division of labor 

generalizations containing one or more typed variables (Marcus 1998, 2001: ch 3).7The criteria for regularity, in this specific technical sense, have been most fully workedout for inflection (Pinker 1999: 217–18, 222–4, etc, Pinker and Ullman 2002: 458–62)

By these criteria, the grammar of present-day English is found to contain symbolicgeneralizations driving the realization of past tense by means of the suffix /-d/, withits concomitant alternation between [-d], [-t], and [-Id]: see (10a) In contrast, otherpatterns of exponence such as strong verb ablaut (10b) satisfy our operational defini-tion of psychological reality but fail to pass the test of regularity: following the dual-route approach, I assume that such patterns are encoded implicitly in the connectionsbetween nodes in a distributed associative memory (although in §2.3.4 I note theimportant alternative offered by Albright and Hayes 2003)

Within the class of morphophonological patterns represented by symbolicgeneralizations, however, one can still observe wide differences in productivity andsusceptibility to lexical exceptions This fact suggests that dual-route theories of expo-nence require further elaboration: we need a ‘refined’ dual-route model (Clahsen,Sonnenstuhl, and Blevins 2003: 127, 149) As emphasized in the tradition of LexicalPhonology and Stratal OT, for example, present-day English exhibits a sharp contrastbetween ‘class-two’ or word-level constructions, such as /-d/ suffixation in the pasttense, and ‘class-one’ or stem-level constructions, like adjectival derivation with thesuffix /-Ik/.8 The alternation affecting the suffix /-d/ (10-a) is strictly exceptionless:notably, in the wug test conducted by Albright and Hayes (2003: 151) participantsfailed to produce the expected alternant [-d] in only one out of 937 responses to probesending in voiced segments other than /d/ (see also note 24 below) In contrast, therule whereby /-Ik/ attracts primary stress to the preceding syllable (11a) does sustainlexical exceptions (11b): see Fournier (2010: 28) for an exhaustive list

(11) a acrobát-ic, genét-ic, harmón-ic, idýll-ic, Miltón-ic, titán-ic, etc

b Cáthol-ic, Árab-ic

Ordinary dual-route models do not provide us with the means to account for thisdisparity between stem-level and word-level constructions On the one hand, thegeneralizations governing the assignment of stress to English stem-level forms drivediachronic processes of regularization and apply productively to novel forms, includ-ing phonotactically deviant items: accordingly, they cannot involve mere patternassociation, but must rather be explicitly represented in the grammar by means ofsymbolic generalizations (Hayes 1982: 236–7, and see §2.3.3.1 for further details) On

7 This chapter’s concerns are mainly architectural; in general, the implementation of symbolic alizations by means of rewrite rules, on-or-off parameters, inviolable constraints, or violable constraints will not be at the heart of the discussion, though optimality-theoretic constraint interaction will play

gener-an importgener-ant role in the execution of some gener-analyses (e.g §2.3.3.2, §2.4.2.2) gener-and in the formalization of some proposals (§2.4.3) The term ‘rule’ is therefore to be understood in the broad sense of ‘symbolic generalization’, unless a narrower meaning is made clear by the context.

8 Empirically, my distinction between stem-level and word-level constructions coincides roughly (but only roughly) with Siegel’s (1974: 111ff.) distinction between class-one and class-two affixation, and with Booij and Rubach’s (1987) distinction between cyclic and postcyclic affixation Conceptually, it is rather different: see §2.4.2.3 below, specially (62) and the immediately preceding paragraph Booij and Rubach’s (1987) postlexical stratum corresponds to my phrase level.

 Ricardo Bermúdez-Otero

the other hand, the absence of exceptions to the alternation shown in (10a) cannotmean that stored past-tense forms are unable to block regular /-d/ suffixation, forsuch blocking is essential to the survival of irregular strong forms

The solution lies in acknowledging two types of lexical listing English stem-levelderivatives, I shall argue, are mandatorily entered into the lexicon in the form assigned

to them by the stem-level morphology and phonology: i.e they are listed not as strings

of underliers, but as fully prosodified stem-level output structures This type of listing,which following Kaye (1995: 302ff.) I call nonanalytic, corresponds approximately

to Jackendoff ’s (1975: 643ff.) notion of ‘whole-form storage’ In contrast, word-levelconstructs, including regular past-tense forms, may be listed or unlisted; but, if theyhappen to be listed, they are crucially entered into the lexicon analytically, i.e asconcatenations of word-level input pieces Analytic listing thus resembles the concept

of a ‘combinatorial lexical entry’, which Clahsen and Neubauer (2010: 2634) postulate

on the basis of independent psycholinguistic findings (for further psycholinguisticsupport, see Stemberger and MacWhinney 1986, 1988) Because stem-level forms arelisted nonanalytically, i.e with their stem-level phonological properties fully specified,their lexical entries are able to block the on-line application not only of stem-levelaffixation rules, but also of the phonological processes (such as stress assignment)that target stem-level domains; the latter may consequently sustain lexical exceptions,

as we see in (11b) In contrast, the fact that word-level past-tense forms are eitherunlisted or listed analytically entails that the surface realization of the word-level suffix/-d/ is always computed on line and so strictly abides by the pattern shown in (10a)—excepting only cases of processing error, as shown by Albright and Hayes (2003: 151)

In the refined dual-route architecture proposed in this section, therefore, the tinction between explicit symbolic generalization and implicit pattern association,which has long been at the heart of dual-route models, is combined with a furtherdistinction between analytic and nonanalytic listing The result is a threefold tax-onomy of exponence mechanisms, where the idiosyncratic properties of stem-levelmorphophonology emerge from the interaction of nonanalytic listing with symbolicgeneralization In this account, stem-level processes operate in a way that closelyresembles Jackendoff ’s (1975) lexical redundancy rules, and indeed the exposition insection 2.3.3.1 will highlight the Jackendovian pedigree of the idea

dis-(12) distributed associative memory

nonanalytic listing

lexical redundancy rules

explicit symbolic generalization standard rules

In section 2.3.3.2, however, I will demonstrate that Stratal OT can model the logical effects of Jackendovian lexical redundancy rules at the stem level without anyaddition to its existing phonological technology For Stratal OT to handle a form like

phono-Árabic, three ingredients must come together:

(13) a nonanalytic lexical listing,

b morphosyntactic blocking,

c high-ranking faithfulness in the phonology

Architecture of grammar and division of labor 

First, it is necessary that stem-level constructs like -ic adjectives should be entered

into the lexicon with their stem-level phonological properties (including prosody)

fully specified Second, the presence of an entry for Árabic in the lexicon must be able

to prevent the morphosyntax from building the adjective on line through the addition

of the suffix -ic to the noun stem Arab, in just the same way that the lexical entry of drove blocks *drive-d Finally, metrical faithfulness must rank high in the stem-level

phonological constraint hierarchy, so that the proparoxytonic contour specified in the

lexical entry of Árabic can withstand neutralization to the default pattern exemplified

by idýll-ic, Miltón-ic, titán-ic, etc.

This way of capturing the phonological effects of lexical redundancy rules makestwo correct predictions (§2.3.3.2) First, the principle of free ranking requires us tocountenance the possibility that, in the stem-level hierarchy, some faithfulness con-straints may be superordinate whilst others may be crucially dominated by marked-ness But, in the model outlined in (13), superordinate faithfulness plays a crucialrole in protecting exceptional phonological properties stored in nonanalytic lexicalentries Stratal OT therefore predicts that the stem-level phonology will simultane-ously enact two types of generalization:

(i) default rules (like penultimate stress in -ic adjectives), driven by markedness

constraints subordinated to exception-protecting faithfulness; and

(ii) exceptionless well-formedness generalizations, enforced by top-ranked ness constraints

marked-This is a good result: exceptionless well-formedness generalizations at the stem

level are needed, inter alia, to express inviolable phonemic inventory restrictions

(Bermúdez-Otero 2007a)

Second, this account makes strikingly accurate predictions about cyclic cation within stem-level domains, which I shall again illustrate with evidence frompresent-day English Consider the familiar triads in (14), which involve two rounds

reappli-of stem-level derivation

(14) a órigin b orígin-al c orìgin-ál-ity

órigin orígin-ate orìgin-át-ion

The complex forms in (14c) violate a stem-level phonological generalization known

as the Abracadabra Rule (after Selkirk 1984: 117), which states that a sequence of

three pretonic light syllables will bear secondary stress on the initial syllable: cf cadábra, dèlicatéssen, Mèditerránean, càtamarán This violation of the Abracadabra

àbra-Rule is a cyclic effect: stress assignment reapplies after each round of affixation, and

the foot-head assigned to the second syllable of oríginal and oríginate in the second

cycle blocks the enforcement of the Abracadabra Rule in the third cycle

(15) originality

first cycle órigin

second cycle oríginal

third cycle orìginálity

 Ricardo Bermúdez-Otero

Several aspects of this phenomenon raise problems for Lexical Phonology in its sical rule-based versions (e.g Booij and Rubach 1987) First, the internal cyclicity ofthe stem level is stipulated rather than explained: Lexical Phonology simply states

clas-as an axiom that stem-level phonological rules reapply cyclically, whereclas-as word-leveland phrase-level rules apply once across the board within their respective domains.Second, Chung (1983: 63) observes that stem-level processes exhibiting cyclic reappli-cation in complex forms also sustain outright lexical exceptions in monomorphemic

items: the Abracadabra Rule fails in forms like Epàminóndas and apòtheósis, bearing

out Chung’s Generalization (22)—but Lexical Phonology offers no account for thiscorrelation Third, Lexical Phonology makes no provision for the fact that cyclicreapplication within stem-level domains is itself subject to exceptions: in (16a), forexample, we see the Abracadabra Rule unexpectedly displaying normal application(Collie 2007: 147); (16b) shows an instance of cyclic stress preservation failing in adifferent type of pretonic environment (Collie 2007: 289); and (16c) is an exception

to the cyclic transmission of trochaic vowel shortening (Collie 2007: 289–90, and seenote 10 below)

(16) a illégible ìllegibílity (alongside expected illègibílity)

b (tríangle) triángulate trìangulátion (alongside expected triàngulátion)

c c[aI]cle c[I]clic c[aI]clicity (alongside expected c[I]clicity)

Stratal OT solves these three problems of Lexical Phonology at a single stroke byadopting the recipe for lexical redundancy rules given in (13) In this view, the manda-tory nonanalytic listing of stem-level forms entails that, when a speaker of English

first produces or perceives the adjective oríginal, she immediately assigns it a lexical entry specifying a foot-head on the second syllable: /oríginal/ The existence of this lexical entry blocks the suffixation of -al to órigin on line, thereby preventing the derivation of *òrigin-ál-ity from the remote base órigin in one fell swoop And, in turn, faithfulness to the input foot-head in /oríginal-ity/ takes precedence over the

alignment constraints driving the Abracadabra Rule Thus, the sequence shown in(15) does not consist of steps in a single on-line derivation, but rather holds betweenhistorical events of stem formation and storage under a régime of nonanalytic listing,blocking, and high-ranking faithfulness (13) In this model, then, the word level doesnot show cyclic reapplication internally because word-level forms are either unlisted

or listed analytically: properties assigned in the output of the word-level phonologyare therefore not stored, and so cannot be fed again as input to the word-level phonol-ogy from the lexicon Chung’s Generalization holds because the same high-rankingfaithfulness constraints that protect the input foot-head on the second syllable of

complex orìginálity guarantee its survival in monomorphemic Epàminóndas And,

finally, cyclic reapplication effects can fail because they crucially depend on blocking

by lexical entries, but blocking is itself variable and depends on factors such as tokenfrequency in ways explained by parallel race models of processing (Schreuder andBaayen 1995, Baayen, Dijkstra, and Schreuder 1997, Hay 2003): this point will be

developed at length in §2.3.3.3, where I examine the famous case of còmp[ e ]nsátion

vs cònd[`ε]nsátion.

Architecture of grammar and division of labor 

In sum, the remainder of section 2.3 will show that we can make progress on thequestion of the division of labor between storage and computation of exponents byimplementing the refined dual-route framework outlined in (12) with the technology

of Stratal OT

2.3.2 Bloomfield’s lexicon and SPE’s evaluation measure

Let us return to the division of labor between storage and computation in SPE

Chom-sky and Halle’s theory imposed very strong biases on the analysis of alternations: itdiscouraged storing allomorphs in the lexicon, and it favored deriving alternations bymeans of phonological rules (cf also Bonet and Harbour this volume: §6.2.1) Thesebiases were implemented through two important postulates of the theory:

(i) First, Chomsky and Halle did not recognize morphology as a separate module

of the grammar: the surface structures generated by the syntax provided the

input to the phonology (1968: ch 1, §4) Nonetheless, SPE did in fact envisage

a number of processes mediating between syntactic surface structures andunderlying phonological representations (ch 1, §5; ch 8, §6) These processesincluded “readjustment rules” that, among other things, handled certain cases

of allomorphy: e.g a readjustment rule mappedVVsing past onto Vs∗ng,where /∗/ stood for an /i/ annotated with a diacritic feature that triggered alater phonological rule of ablaut converting /i/ into /æ/ (p 11) Significantly,the/∗/→/æ/ transformation itself was claimed to take place in the phonology.

The homonymous readjustment rules of Distributed Morphology operate in

a very similar way: see Embick and Halle (2005: 41) for a discussion of sang.

Embick and Halle (2005: 42) insist that the readjustment rules of Distributed

Morphology are phonological rules, just as the/∗/ →/æ/ transformation of SPE was supposed to be (cf §2.4.3 below).

(ii) Following Halle (1959), moreover, SPE posited an evaluation measure that

selected the system of lexical entries and phonological rules containing thesmallest number of symbols: this favored the adoption of any rule that dis-pensed with more symbols in lexical entries than it took to state the rule itself(pp 334, 381, 389)

The prevalent, though by no means universally shared, view today is that SPE’s

analytic biases were wrong First, they favored descriptively inadequate grammars,i.e grammars which misrepresented adult speaker competence The main problemwas the derivational remoteness of underlying representations, which could be vastlydifferent from surface forms and related to them through extremely long and opaquederivations.9A notorious example was Chomsky and Halle’s (1968: 233–4) postula-

tion of velar fricatives in the underlying representations of nightingale /nixtVng ¯æl/

9 In the controversy that ensued (e.g Kiparsky 1968, Hyman 1970, Crothers 1971, etc.), this problem was often characterized as one of excessive ‘abstractness.’ It must be noted, however, that underlying representations may be ‘abstract’ in very different ways, not all of which incur the problems that afflicted

SPE As we shall see in §2.4.2.2, for example, the underlying representation of a morpheme may consist of

a bare mora, which may be provided with segmental content by a variety of processes: e.g by lengthening a vowel or by reduplicating a syllable-sized string; see (53) Such an underlying representation is undeniably

 Ricardo Bermúdez-Otero

→ ["naItn

"­geIl] and dinghy /dinxi/→ ["dIŋi]: the former “explained” an exception to

trisyllabic shortening (p 52 and passim); the latter “accounted for” the unexpected

occurrence of [ŋ] before a tautomorphemic vowel (cf Bermúdez-Otero 2008) It isimportant to understand that these—for us—obviously misguided proposals were

more than a fanciful indulgence, but were highly favored by SPE’s analytic biases For example, given a relatively pervasive alternation like div¯ıne ∼ div˘ınity, ser¯ene

∼ ser˘enity, s¯ane ∼ sănity, etc., one can effect a net reduction in the overall symbol

count of one’s description by positing a single underlier for each root and setting

up a rule of trisyllabic shortening targeting long vowels in antepenultimate syllables

preceding a stressless penult (SPE, p 52).10Having done that, however, one needs to

exempt n¯ıghtingale from this rule If one regards rule diacritics of the format [ ±rule n]

(Chomsky and Halle 1968: ch 8, §7) as devices of last resort,11then SPE’s evaluation

measure forces one to look for the alternative that spends the fewest additional bols: this alternative may well involve adding a feature or segment to an underlier (in

sym-this case, a /x/ to /nixtVng ¯æl/) for the sole purpose of opportunistically triggering arule that happens to be already lying about (in this case, a battery of rules dealing with

the appearance of surface [tS] in question, bastion, righteous, etc.).

Second, the grammars favored by the analytic biases of SPE were unlearnable except

under extremely idealized and unrealistic conditions Chomsky (1957: ch 6) hadearlier disclaimed the ambition to provide a “discovery procedure for grammars”,i.e an unsupervised learning algorithm inferring a grammar from a corpus; he hadsettled instead for the lower goal of devising an “evaluation procedure”, i.e a criterionfor arranging a given set of grammars in an order of preference, given a corpus (seefurther Chomsky 1965: ch 1, §6–7) In this idealized scenario, an unspecified function

of Universal Grammar provides the learner with a set of grammars compatible withthe primary linguistic data, and the evaluation measure chooses the best In the 1950sthis was an expedient compromise: some structuralist linguists had arguably taken awrong turn by subjecting grammars to conditions which were designed to make a cer-tain sort of discovery procedure viable, but which in effect made it impossible to pro-vide adequate characterizations of fundamental aspects of linguistic structure, includ-ing morphology–phonology interactions (e.g Hockett 1942: 20–1, Moulton 1947:note 14; cf Pike 1947, 1953) Whatever the merits of Chomsky’s strategy in the 1950s,

however, it must remain a supreme goal of linguistic theory to devise unsupervised

learning algorithms that infer descriptively adequate grammars from the sort of inputavailable to children Yet it seems clear that there is no feasible discovery procedure

‘abstract’ (in the sense that it omits a great deal of concrete surface detail), but it need not require the

extremely opaque derivations that characterized SPE: indeed, the relevant mappings may be transparent

and monotonically structure-building.

10 Prince (1990: 368–70) demonstrates that, in fact, trisyllabic shortening is trochaic shortening

(Hayes 1995: 142–9) under final syllable extrametricality This explains why the suffix -ic, which normally

induces mere consonant extrametricality (18a), induces trochaic shortening of stressed penults, as in

c[aI]cle ∼ c[I]cl-ic (16c).

11 As Spencer (1991: 101) puts it, “Classically, generative phonologists have tried to use rule features as sparingly as possible.”

Architecture of grammar and division of labor 

for the grammars most highly valued by SPE’s evaluation measure:12automated rulelearners do of course exist (e.g Albright and Hayes 2003), but they do not return

SPE-style grammars OT provides a good term of comparison: the framework has

given rise to explicit learning algorithms tested in computer simulations of nontrivialacquisition tasks (e.g Tesar and Smolensky 2000, Boersma and Hayes 2001).These two problems are of course intimately related: if the grammars favored by the

analytic biases of SPE are unlearnable under realistic conditions (second problem),

then they cannot be descriptively adequate (first problem) But what went wrong?

The received wisdom, with which I agree, is that SPE struck the wrong balance

between storage and computation: see e.g the papers in Nooteboom, Weerman,and Wijnen (2002), specially Pinker and Ullman (2002), Jackendoff (2002a), and

Booij (2002) However, the usual assertion that SPE favored the minimization of

storage and the maximization of computation does no more than trivially restate

the effects of SPE’s evaluation measure; it fails to explain why Chomsky and Halle chose that measure over others Significantly, SPE did not overtly base its choice of

evaluation measure on a claim that the properties of the human brain put a premium

on storage in long-term memory—though this claim is made, and half-heartedlydefended, by Bromberger and Halle (1989: 56–7, especially footnote 7) Rather, theroot of the problem lay, as Jackendoff (2010: 37) correctly observes, in Chomsky andHalle’s espousal of Bloomfield’s (1933: 274) conception of the lexicon as an unstruc-tured, cognitively inert list: “The lexicon is really an appendix of the grammar, a list

of basic irregularities” (echoed in Chomsky 1965: ch 2, endnote 16) If the grammarcomprises only rules and lexical entries, and everything contained in lexical entries

is an accident, then it follows that every nonaccidental pattern, i.e every patternexhibiting any symptom of psychological reality (§2.3.1), must be expressed as a rule.Now consider two such patterns:

(i) The ablaut alternation present in English verbs like string ∼ strung, stick ∼ stuck, and dig ∼ dug since the sixteenth century was generalized to the orig- inally weak verb sneak in American dialects possibly as late as the nineteenth century: this yielded the innovative strong past-tense form snuck (Hogg 1988;

see further Wełna 1997) This analogical extension shows that the alternationbetween a high front vowel in the unmarked form and [2] in the past tense was

in some sense psychologically real for speakers; otherwise it could not have been

extended Therefore, if rules provide the only way to represent psychologically

real patterns, and if phonological rules are the preferred type of rule in this case,then one is forced to countenance phonological processes of ablaut for strongverbs in present-day English: see Halle and Mohanan (1985: 107–14) and Halleand Marantz (1993: 127–9)

(ii) The English word righteous appears to be semantically and phonologically related to the word right: someone is righteous if they do right, and the string

12 In fact, even the evaluation measure itself was never applied as defined, since no one ever attempted

to compare the global symbol counts of two complete packages of lexicon plus rules (Prince 2007: §2.1.2).