UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The prosodic structure of Finnish and the theory of phonological government Millard, David 1991

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata


831-ubc_1992_spring_millard_david.pdf [ 2.86MB ]
JSON: 831-1.0086597.json
JSON-LD: 831-1.0086597-ld.json
RDF/XML (Pretty): 831-1.0086597-rdf.xml
RDF/JSON: 831-1.0086597-rdf.json
Turtle: 831-1.0086597-turtle.txt
N-Triples: 831-1.0086597-rdf-ntriples.txt
Original Record: 831-1.0086597-source.json
Full Text

Full Text

THE PROSODIC STRUCTURE OF FINNISH AND THE THEORY OF PHONOLOGICAL GOVERNMENT by DAVID MILLARD BMus, The University of Victoria, 1982 Diploma in Applied Linguistics, The University of British Columbia, 1985 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS in THE FACULTY OF GRADUATE STUDIES (Department of Linguistics)  We accept this thesis as conforming to the required standard  THE UNIVERSITY OF BRITISH COLUMBIA December 1991 © David Millard, 1991  In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission.  Department of  Linguistics  The University of British Columbia Vancouver, Canada  Date  DE-6 (2/88)  December 30, 1991  The Prosodic Structure of Finnish and the Theory of Phonological Government by David Millard Submitted September 1991 to the Department of Linguistics, UBC, in partial fulfillment of the requirements for the degree of Master of Arts.  ABSTRACT The goal of this thesis is to examine same of the claims of the Theory of Phonological Government in the light of data taken from the Finnish language. The Theory of Phonological Government describes the oraganization of segments into constituent groupings on the basis of government relations established at the level of Underlying Representation. These prosodic constituents are related to each other, again in terms of government. While these prosodic constituents are functionally similar to traditional notions of the syllable, division of a surface string into syllables does not necessarily yield the relations predicted by the theory. Government Theory, then, makes predictions which are different from those of other syllable theories and these predictions can be tested. The Finnish language exhibits a number of phonological processes that are sensitive to prosodic structure and thus offers an excellent test case for the theory. Chapter One presents an overview of the theory and the predictions it makes. Chapter Two examines same general phonological characteristics of Finnish. In Chapter Three I examine two prosodically-influenced segmental ii  deletion processes that interact with each other and shows how Government Theory accounts for than in a principled fashion with minimal appeal to specific 'rules' of deletion. In Chapter Four I examine the process of Consonant Gradation, a mutation process that is driven by syllable structure. In Chapter Five I demonstrate that certain surface strings of segments that appear to be CVVC and CVCC syllables are best treated as being other than syllables and show how the theory not only accounts for the data but also predicts the existence of such non-syllabic strings. Finally, in Chapter Six I resume the discussion of Consonant Gradation and examine the nature of some exceptional gradation forms, showing how the theory accounts for these more unusual forms.  Thesis supervisor: Professor P. A. Shaw  iii  TABLE OF CONTENTS ABSTRACT  ii  LIST OF ABBREVIATIONS ACKNOWLEDGEMENTS INTRODUCTION Notes to the Introduction  vi vii 1 3  CHAPTER ONE - The Theory of Phonological Government  4  1.1 The Structure of Segments  4  1.2 Representations of Phonological Structure  6  1.3 Prosodic Constituents  7  1.4 The Governing Properties of Segments  10  1.5 Government between Neutral Segments  11  1.6 The Relation of Segments to Prosodic Categories . . . 	 12 1.7 The Generation of Prosodic Structure Notes to Chapter One CHAPTER TWO - The Sound Pattern of Finnish 2.1 Consonant Sequences 2.2 Stress Notes to Chapter Two  14 16 18 18 22 24  iv  CHAPTER THREE - The Deletion of Segments  26  3.1 Two Segmental Deletion Processes  26  3.2 e-Deletion  27  3.3 Empty Categories and Phonological Government  28  3.4 t-Deletion  35  Notes to Chapter Three CHAPTER FOUR - Consonant Gradation  40 41  4.1 General Features of Consonant Gradation  41  4.2 The Nature of Consonant Gradation  42  Notes to Chapter Four CHAPTER FIVE - WC and VCC Groupings 5.1 CVVC and CVCC Non-syllables  48 50 50  5.2 CVCC Sequences  51  5.3 Monanoraic Sequences  53  5.4 Monomoraic Vowel-Sonorant Sequences  56  5.5 CVVC Sequences  59  5.6 The Derivation of Doubling Forms  63  Notes to Chapter Five  72  CHAPTER SIX - Further Aspects of Consonant Gradation 6.1 Same Exceptions to Consonant Gradation  76 76  6.2 Some Remarks on Glide Formation  79  Notes to Chapter Six  82  CONCLUSION  83  LIST OF ABBREVIATIONS While I have endeavoured to avoid the use of abbreviations it has been necessary for the sake of space to use some abbreviations in the examples and derivations. Standard linguistic abbreviations have been included here for the sake of completeness. ant ATR C  cont cor ECP ess.  F gen. ICG lab M m N nam. O part. pl. R s d SCG sg. son transl. ✓ V: voi w  anterior Advanced Tongue Root consonant continuant coronal Empty Category Principle essive case metrical foot genitive case Interconstituent Government illative case labial metrical word mora Nucleus node nominative case Onset node partitive case plural Rime node strong (branch of a metrical structure) syllable Subconstituent Government singular sonorant translative case vowel long vowel voiced weak (branch of a metrical structure)  vi  ACKNOWLEDGEMENTS  It is difficult to assess, after four years, which of the many people who have influenced the writing of this thesis made which particular contribution. I shall have to offer most of my friends and fellow students a 'global' statement of thanks--otherwise, the list of names would be prohibitively long. Among the instructors and faculty members outside my thesis committee, I must single out Drs. Glyne Piggott, Zita MbRobbie, Ewa CzaykowskaHiggins, and Charles Ulrich. To them I owe much of my grounding in phonology and not a few warnings about the pitfalls of working in a nonstandard framework. If I chose to ignore those warnings, it is no reflection on their powers of persuasion. I would also like to thank Dr. Guy Carden who more than once banished an administrative difficulty for me and who, incidentally, first introduced me to the intricacies of Finnish phonology. To the non-supervisory members of my committee, Drs. M. D. Kinkade and Juta Kitching, I offer my heartfelt thanks not only for their comments and suggestions, but also for their patience in putting up with my somewhat unorthodox work habits. Needless to say, they do not agree with everything I have to say, but this thesis would nonetheless be poorer were it not for their influence. To my supervisor, Dr. Patricia Shaw, I cannot offer enough thanks for her instruction, her encouragement, her patience and her energy on my behalf. I owe any right to call myself a phonologist to her. Again, while she does not agree with everything I say, this thesis simply could not have been written without her. Finally, I must acknowledge my family, whose lives have all been touched in one way or another by the writing of this thesis, and all of wham offered 'gentle reminders' to me to finish writing it. This thesis is dedicated to the menory of my mother, Peggy Millard, who never even knew I had taken up linguistics but who always supported me in my past endeavours.  vii  INTRODUCTION In a paper entitled "Generative Phonology vs. Finnish Phonology" Lyle Campbell states that "[t]heory and data are in a trading relation (the socalled hermeneutic circle or spiral). That is, the theory explains the data, but a particular set of data (a language) described in terms of the theory may reveal inadequacies in the theory which lead to revisions." (Campbell 1981; 147). He then goes on to argue that Finnish is a language rich in theoretical test cases and has a history of exposing inadequacies in generative phonological theory. Generative Phonology has seen many changes since the publication of that paper, but the somewhat jocular title still carries same weight. Finnish remains a good language for testing phonological theories and while advances in Generative Phonology, particularly in the areas of autosegmental representation and syllable structure, have made Finnish more amenable to phonological explanation, the problems set by Finnish have by no means all been solved. This thesis will focus on some of the problems associated with the prosodic structure of Finnish. Finnish has a number of processes that are sensitive to prosodic structure, ranging from a series of deletions and their various effects to the weakening process of Consonant Gradation which is sensitive to the structure of the Rime. I cannot claim to have resolved all the issues of all these processes, but I am able to sketch the form that such a solution must take. As these processes are sensitive to what may be termed informally 'syllable structure', any account of them must, therefore, be undertaken in the context of a well-articulated theory of syllable structure. I will investigate the contribution that one syllable theory, the theory of Phonological Government presented by Kaye, Lowenstamr and Vergnaud (1989) has to make toward the understanding of  1  a number of phonological problems in Finnishl. I shall also argue that the theory of Phonological Government makes a number of non-standard predictions and assumptions that account for the Finnish data in an elegant fashion. In the first chapter of this thesis I shall outline the theoretical assumptions I shall be using. The second chapter will present some of the phonological characteristics of Finnish and demonstrate how the theory can be applied to describe them. In these two chapters I shall assess some difficulties the theory faces in its standard form. Specifically, I shall demonstrate the need of a subtheory dealing with the erection of prosodic structure and demonstrate the inability of Charm Theory to account for all the observed licit and illicit consonant sequences in Finnish. In the remaining chapters I shall examine particular instances of prosodically sensitive processes. Chapter Three deals with two interacting deletion processes. Chapter Four presents the basic details of Consonant Gradation. Chapter Five examines the variety of observed 'syllable' types in Finnish and presents a Government-based account of them. Chapter Six returns to the subject of Consonant Gradation, focussing on sane apparently exceptional aspects of the process in the light of the account developed in preceding chapters.  2  Notes to the Introduction 1. Kaye, Lowenstanm and Vergnaud actually exclude the notion 'syllable' from their theory in favour of a sequence of alternating onsets and rimes without any additional hierarchy. However, as Consonant Gradation involves a relationship within easily defined Onset-Rime (OR) pairs, it will be convenient to speak of these OR pairs informally as syllables. This convention will also facilitate comparisons with alternative representations in models that include the syllable as a legitimate category within their framework. I shall therefore continue to use terms such as 'syllable structure', 'onset of a closed syllable' etc. as the most economical way of describing these notions in informal discussion. I shall further note here that Kaye, Lowenstamm and Vergnaud (1989) is a published version of an earlier manuscript (Kaye, Lowenstamm and Vergnaud 1987). The 1989 version is a translation into German of a paper originally written in English. As the German version shows minimal revision over the English version (and, indeed, maintains some uncorrected errors of the English version) I have chosen, when actually quoting the paper, to quote the English text rather than to retranslate the German. In these instances, I have cited the location of the material in both the English and German versions. Where I simply cite the location of an idea without direct quotation, I provide references to the published version only.  3  CHAPTER ONE - The Theory of Phonological Government In this chapter I shall outline the basic aspects of Government Theory (cf. Kaye, Lowenstamm and Vergnaud 1989) and sketch out an analysis of the problems to be addressed in this thesis. 1.1 The Structure of Segments' In the theory of Phonological Government, segments comprise neither the structureless feature matrices of Chamsky and Halle (1968) nor the hierarchy of features found in e.g. Clements (1985). Instead, the representation of segments involves a marriage of the theory of distinctive features with the unitary segmental components of Dependency Phonology (see Anderson, Ewen and Staun 1985). Each segment is composed of one or more 'elements', which are themselves bundles of a subset of the total number of features available (see Kaye, Lowenstamm and Vergnaud, (1985) for a detailed discussion). Each element represents a fully specified segmental matrix which is capable of functioning independently as a pronounceable speech sound. One feature of each element acts as a head for the element (in Kaye, Lowenstamm and Vergnaud's terminology, this is the 'hot' feature) and this feature alone can interact with other elements (by spreading, for example). Within a segment elements are organized in headoperator relations with the head contributing its entire matrix and the operators adding only their 'hot' features to change the value of the corresponding features within the head. To give a more concrete demonstration I shall use the elements A and I which, in the absence of any  4    	:  operators, define the sounds [a] and [I], respectively. The full representation of these elements is presented in (1): (1)  (fram Kaye, Lowenstanm and Vergnaud 1985: 309) A  I  - ROUND  - ROUND  + BACK  - BACK  :LILO_ - ATR + LCrel  1 + HIGH - ATR -LOW  (Hot features are underlined) When combined in a fusion operation (designated by the operator '.'), these elements define two new sounds the identity of which is determined by which element is the head: (2)  (from Kaye, Lowenstamm and Vergnaud 1985: 309-10) head  operator  : - ROUND : +  BACK  I 1 +- g?	 11 = []  1 - ROUND '  : - ROUND '  1 - ROUND  '  : + BACK :  1-  1 - HIGH. ATR	 '1	: : + LOW 1 A  : - HIGH ATR : - LOW = e]  •  I	:  	+  - HIGH : ATR ' '1 + LOW A  -  :  BACK  HIGH  : - ROUND  ' - BACK '  1 + HIGH : 1	I 1 - ATR : - LOW : I  	:  	:  : - ROUND :  '  .  . : - ATR 1	1 ' - LOW : I  I  i  -  BACK  : 1  [  By way of shorthand notation, the elemental structure of a segmental representation may be written as: (3)  X  X  A [m]  [E] 5  where, following the conventions of Dependency Phonology (see Anderson, Ewen and Staun 1985), the head is placed above the operator(s). However, where it is necessary only to identify a segment without referring to its constituent structure, I shall employ conventional phonetic symbols: (4)  X  X  Be  1.2 Representations of Phonological Structure The representations of phonological structures within this theory are based on the principles of Autosegmental Phonology. Segments (represented as clusters of elements) are grouped on a melodic tier and associated to positions (designated 'X') on a skeletal tier. Skeletal positions are further grouped into prosodic constituents ('syllables') which are in turn organized into metrical groupings. A basic schema is presented in (5): (5) Metrical Structure:  Word Foot  Prosodic Constituents:  Skeletal Tier: Melodic Tier:  / \ s w I II 0 R 0 R 1 1 : (see discussion NIN immediately below) III# X X X X III! [III s s s s (s=segment)  In practice, it will not usually be necessary to specify every level in each derivation.  6  1.3 Prosodic Constituents The notion 'prosodic constituent' is intimately tied to the notion of 'government' in this theory. Government is defined (Kaye, Lowenstamm and Vergnaud 1989, 37; 1987, 8) as "a binary, asymmetric relation holding... between two skeletal positions". The relation is strictly local: the governor and governee must be adjacent; and unidirectional: at a given prosodic level, governors govern in one direction only. In defining prosodic constituents two types of Government are required: government within a constituent (Subconstituent Government or SCG) and government between constituents (Interconstituent Government or ICG) 2 . Subconstituent Government holds between the head of a constituent and its complement, and is strictly head-initial. As a consequence of this, all prosodic constituents are maximally binary (a head cannot govern in two directions at once) and left-headed. Interconstituent Government holds (for the present discussion) between constituent heads and skeletal positions to their left (i.e. ICG is head-final). In the present theory, there are three constituents: the Onset, Nucleus and Rime (abbreviated 0, N and R). The head of the Onset is the skeletal position under the left branch of the Onset node (should it branch) while the head of the Rime is the skeletal position under the left branch of the Nucleus node. Onset and Rime are maximal projections of  7  their heads while the Nucleus is a medial projection of the head of the Rime. The camplete inventory of well-formed constituent types is therefore: (6)  0 0 I1\ 'I 1 \ 1 1	 \ X X X  R I N I X  R I	I N 1\ X X  R 1\ N \ 1	 \ X X  Because heads must always be adjacent to their complements, the head of a Rime may govern only the right-hand branch of the Nucleus (where the Nucleus branches and the Rime does not) or the right-hand branch of the Rime (where the Rime branches and the Nucleus does not). A structure where both the Nucleus and Rime branch is ill-formed, as the segment dominated by the right branch of the Rime would be separated from the head by the right branch of the Nucleus and would therefore be ungoverned. Such a structure is illustrated in (7): (7)  where z is ungovernable because it is not adjacent to the head x (Kaye, Lowenstamm and Vergnaud 1989, 39). Returning to the discussion of Interconstituent Government, it is now possible to define the ICG configurations available. Kaye, Lowenstamm and Vergnaud (1989, 53; 1987, 24) present two principles of Government that limit the possible IOG structures: (8) a. Only the head of a constituent may govern. b. Only the nucleus (or a projection thereof) may govern a constituent head.  8  (8a) is a general condition on Government. (8b) applies specifically to ICG, for while the head of an Onset may govern a preceding Rimal or Nuclear complement, it cannot govern the head of a non-branching Rime, whereas the head of a Rime can govern the head of a non-branching Onset. There are, therefore, five possible ICG configurations: (9) a. Government of a Rimal complement by an Onset  R 1\ N  0 1  \ X X <-- X b. Government of a Nuclear complement by an Onset 0  R  1\ X X <-- X  c. Government between contiguous Nuclei  X < d.  X  Government between a Rime and an Onset complement 0 \ \ X X <  e.  N \ i	 \ X (X)  Government of the head of an Onset by the head of a Rime 0  R 1\ N \  \ X <-- X (X) (9c) demonstrates one form of government between two Nuclei. Another possible government structure involving Nuclei is Projection 9  Government (Kaye 1987: 6). The theory allows for the projection of Nuclei onto a separate autosegmental tier. On the level of the nuclear projection, Nuclei are adjacent and may govern each other. The direction of Projection Government is subject to parametric variation and in Finnish the relevant direction is left-to-right 3 . Projection Government will become significant in the discussion of Pseudo-Campensatory Lengthening in Chapter Five. 1.4 The Governing Properties of Segments In any theory of syllabification, there must be same designation of what segments may occur in an Onset position or in the Nucleus or Rime. In the theory of Government this is handled largely by a property called 'Charm' (see Kaye, Lowenstamm and Vergnaud 1985 for a full discussion of Charm) 4 . Charm is the property of certain elements and may be positive (Xi), negative (X - ) or neutral--i.e. charmless (X0). Like the hot features, elements impart their charm value to the segment they are associated with. The distribution of charmed segments is determined by the following principles (Kaye, Lowenstamm and Vergnaud 1989, 42; 1987, 13): (10) a. Charmed segments may be associated with governing positions (may be governors); charmless segments may be associated with governed positions. Charmed segments may not be governed. [But see (8b)] b. Positively charmed segments may not occur in non-nuclear positions; negatively charmed segments may not occur in nuclear positions. As a working hypothesis (and one which I will adopt), Kaye, Lowenstamm and Vergnaud (1985, 311; 1989, 42) assign charm values in this fashion: vowels headed by the element W (which includes [+low] in its specification) or associated with the element ki ([+ATR]) are positively charmed; 10  stops, affricates and non-strident fricatives are negatively charmed; all other segments are neutral. It should be noted that although the terms 'consonant' and 'vowel' have no primal status in the theory (consonantality and vocalicity are considered byproducts of prosodic structure) I shall use them as a convenient designation for the class of sounds that occupy non-nuclear positions and the class of sounds that occupy nuclear positions respectively. This designation correlates quite directly with the traditional classification of consonants and vowels. 1.5 Government Between Neutral Segments Not all consonantal contacts are between negatively charmed and neutral segments. For example, in the English word armour, the [r] and the [m] are both neutral segments. Same neutral segments must be able to govern others. The governing properties of neutral segments are restricted by their complexity. A segment is said to be more complex than another if it is composed of a greater number of elements. Thus, given the structures for [m] and Er]: (11)  [r] No  11°  (The meaning of these symbols is discussed in Chapter Two)  ?o  Uo  it is perspicuous in the representation itself that [m] is more complex  11  than [r]. Kaye, Lowenstamn and Vergnaud (1989, 63; 1987, 34) stipulate that: (12)A neutral segment may govern if it has a complexity greater than its governee. The order of complexity among neutral segments is given as (Kaye, Lowenstamm and Vergnaud 1989, 63; 1987, 34): (13)  {glides, r} < 1 < nasals [< s]  (s added - -DM)  The theory also allows for the possibility of empty positions (skeletal positions with no melodic content). An empty position has a complexity of zero, of course, and may (indeed, must) be governed by any segment with phonetic content (Kaye, Lowenstamm and Vergnaud 1989, 62). The existence of empty positions and their distribution will be argued for in Chapter Three. 1.6 The Relation of Segments to Prosodic Categories As presented in Kaye, Lowenstamm and Vergnaud (1989), the theory lacks a principled means of erecting prosodic structure. It is implied that there is a pre-existent prosodic tier consisting of a string of Onset and Rime nodes which associate via the skeletal tier to the charm-bearing elements in a top-down fashion (the Nucleus appears to be a projection from positively charmed elements). This position is not tenable, as shall soon be made clear. A better approach would be to project prosodic structure, bottom-up, according to the charm properties of the elements, but even this approach has its problems. In the strongest version of the theory (Kaye, Lowenstamn and Vergnaud 1989, 42 ff.) vowels are positively charmed, obstruents negatively charmed  12  and sonorants charmless. These charm assignments make it easy to stipulate that the distinction between consonants and vowels that allowed one to assign vowels to Nuclei and consonants to Onsets is their charm value. This idyllic scenario cannot be maintained, however, for Kaye, Lowenstamm and Vergnaud (1989, 47-48; 59-62) are forced to allow for the possibility of both charmless vowels and obstruents 5 . In the theory, vowel tenseness and positive charm are properties of the ATR element I+. In a language such as English which exhibits Closed Syllable Laxing, the element I* is not lexically significant. Vowels are underlyingly charmless and receive a charm value (became tensed) in nonbranching Rimes later in the derivation. Charmless obstruents must be postulated in order to allow for obstruent sequences (e.g. English cactus). Government of one negatively charmed segment by another is not permitted (10a), so the obstruent sequence must have the form [kot - ]. Given that there are both charmless vowels and obstruents at the level of Underlying Representation, charm is no longer sufficient to distinguish consonants fran vowels. Given the further fact that vowels can remain charmless through the course of a derivation and still be associated with a Nucleus one must conclude that it is not only the charm value of a segment that determines the prosodic categories with which it may be associated. I see no easy way out of this dilemma. I feel canpelled to conclude that vowels are distinguished from consonants on the basis of sane property P (perhaps the feature [vocalic] or some other feature) which vowels possess and consonants do not. This same property also defines vowels as able to project Nuclei and ensures that when they receive a charm value, it is positive. This solution is admittedly ad hoc but will serve for the 13  time being 6 . 1.7 The Generation of Prosodic Structure We are now in a position to derive prosodic structure fram underlying representations. To take a simple case, that of ranta 'shore' nam. sg. the lexical representation consists of the segments of the lexeme and the skeletal positions to which they may associate: (14)  X X X X X  r° e no t e -  Upon mapping the segments to the skeletal positions, the P-bearing elements (which in this case are also charmed) project Nuclei:  (15)  N  N  X X X X X 1 11111  r° e n° t  -  Next, the following right-headed (Interconstituent) Government configurations are established (Nuclei govern non-nuclei; charmed segments govern charmless segments): (16) XX XX X 11111 11111  r° a+ n° t - a+ 1_1 I t  14    Left-headed (Subconstituent) Government configurations are established: I__I I I  (17)  N '1 N I	 4r	11 XXXXX 1I		 1I		 I	I	 II		 II ✓ anta t  I  t  I  I  T I  I I  A Rime node is now projected from each Nucleus and with each Rime there is an ancillary Onset that precedes it 7 : (18)  0 R N  1  0 R I 1  *  XXX  XX  ✓ anta TIT! I- I -... I T I  I I  Finally, the Onset and Rime nodes are maximally filled as their government configurations allows: (19)  0 R 0 R II	 II \ I	I	 1I 1 N 1 N \ 1I	 I		 n\ II		 I1 XX XXX 1	I	 1	I I I I I I I ✓ a n t a  The erection of prosodic structure is not always so simple as this example implies, but I will treat the more problematic cases as they arise.  15  Notes to Chapter One 1. I present the structure of phonological elements in the manner of Kaye, Lowenstamm and Vergnaud (1989) primarily because I am utilizing their theory of prosodic structure. Analogues to Charm and other governing properties could be proposed for representations of segments using, for example, feature hierarchies. For the most part, my discussion of segemental interactions will have little to say about the nature of the segments involved other than their governing properties. 2. I use the term 'Subconstituent Government' rather than 'Constituent Government' to stress the fact that the Goverment domain in question is contained within a constituent. 3. The direction of projection Government in Finnish is revealed by the direction of Vowel Harmony which is left-to-right. 4. The term 'Charm' has been borrowed from the physical sciences where it is a property attributed to the hypothetical subatomic particles quarks. It is not a wholly appropriate borrowing, for Charm and Government theory is distinct from Particle Phonology (see Schane (1987) and references therein). The usage is a manifestation of what Dr. Jay Powell (personal communication) calls 'Physics Envy' in linguistics. 5. Both versions of the Kaye, Lowenstamm and Vergnaud paper (1987 and 1989) in the discussion of neutral segments within Nuclei state that "negatively charmed segments may appear in such positions [i.e. a governed position within N]" (Kaye, Lowenstamm and Vergnaud 1989, 48; 1987, 18). Clearly, as negatively charmed segments are absolutely prohibited within the Nucleus (Kaye, Lowenstamm and Vergnaud 1989, 42), this is a misreading for 'charmless segments'. Possibly, this is an uncorrected reference to their earlier version of the theory (Kaye, Lowenstamm and Vergnaud 1985) where there were only two charm values--positive and negative. In the 1989 version the category 'neutral' or 'charmless' has been introduced and applies to all the elements designated 'negative' in the 1985 paper. The term 'negative' has been given a new meaning in the 1989 paper (see Kaye, Lowenstamm and Vergnaud 1989, 42). 6. Difficulty in providing a principled distinction between consonants and vowels is hardly restricted to Government Theory. See Selkirk (1984) for a arguments against basing the distinction on major class features. Selkirk bases the distinction between potential heads of a Nucleus and non-heads of a Nucleus in a given language on the basis of sonority. Under this view, segments of a certain degree of sonority or higher could be nuclear heads and project an N node. Sonority correlates roughly with Kaye, Lowenstamm and Vergnaud's notion of 'complexity'. Unfortunately, complexity could not be used in this fashion, for some neutral segments, such as to are less complex than other, more sonorous segments, such as s° (t° contains two elements, so has three). This is an area that requires further research. 7. I assume that the generation of ancillary Onsets is a universal principle of phonology. Glyne Piggott (UBC lectures, 1988) has argued that there are no truly vowel-initial syllables. In the present framework, Onsets and Rimes are in an Interconstituent Government 16  configuration. One could formulate a principle (analogous to the 8criterion of syntax) that all Onset nodes must be governed by a Rime node and all Rime nodes must govern an Onset node. Where these conditions are not met, the missing constituent must be inserted to render the expression well-formed. 8. It will be seen in Chapters Three and Five that failure to satisfy the available Onset and Rime nodes or conditions on govetueent will result in syncope or epenthesis of segments.  17    CHAPTER TWO - The Sound Pattern of Finnish This chapter will examine the sound inventory of Finnish (with particular reference to the consonants) in terms of the Theory of Phonological Government. The theory predicts, on the basis of a particular inventory, what clusters are permissible in a language. It will be seen that for Finnish, the fit between theory and fact is close, but there are some surprising discrepancies, not all of which are easily explained. 2.1 Consonant Sequences Finnish has the following inventory of phonemes (after Lehtinen 1962, xv)i: p t k v s h 1 r j m n  (1)  i y e 8 o a a  In the theory of segmental representations the available elements and their identifying 'hot' features include (Kaye, Lowenstamm and Vergnaud 1989, 74; 1985, passim.): [+ coronal] L 	[+ slack vocal cords] H 	[+ stiff vocal cords] [+ constricted] ?a 	CP [+ continuant] [+ round] U° [- back] Io - high] v° the cold vowel--no hot feature [+ ATR] 1+  (2) RP -  -  [  18  The segmental structures of the Finnish sounds are thus: 10  ro  jo  mo  ?o  ?°  R°  I°  ?o	R°  I I  I I  (3) p -	t -	k -	v-	so ?° I	I  R°  v°  ?o  II I I I  no  ho 2  CD°  I  U° ?o ?° I I II I I I 1 I  u° R° R°  U° I I  I  Ii-	H-	H-	I.. -	0°  No  No  ';()  CD°  i+  e+  yf  o+  a+  e  0+  u+  I° I I  io I I  I° I I  io I I  A+ I I  A+  uo  uo  I+  A+  U°  U°  I°  I I  I I  I I  I I  I+  I+  I+  I+  I  A+  i+  Of the consonants (which form the subject of this section), [p,t,k,v] are negatively charmed, the remainder are neutral, having the order of complexity [n,m,s] > [1] > [r,h] 3 . Finnish syllables have no branching onsets (Karlsson 1983, 19; Whitney 1956, 15). Consonant sequences in Finnish words are thus restricted to two types: CPC - (Interconstituent Government of a neutral segment by a negatively charmed segment) and CPC0 (Interconstituent Government of a neutral segment by a neutral segment of greater complexity). The third remaining possibility, C - 00 would define the Subconstituent Government relationship of a branching Onset which is  19  unavailable in Finnish. Based on this inventory, Government Theory predicts the following distribution of sequences: (4)  permitted sequences  CPC-	hp 1p rp sp mp 4 *C- Co ht It rt st nt hk lk rk sk 1)k hv lv ry sv my nv CP<CP  hn rm hm in rn hn is rs hs rl hl  non-permitted sequences *ph *th *kh *vh  *pl *tl *kl *v1  *pr *tr *kr *vr  *ps *ts *ks *vs  *pm *tm *km *vm  *pn *tn *kn *vn  *pv *tv *kv *vp *vt *vk  *CP>cP *ml #mr *Ins *Inh *nl *nr *ns *nh *sl *sr *sm *sn *sh *lr *lh *hr	*rh *C- C -	*pt *pk *pv *tp *tk *tv *kp *kt *kv *vp *vt *vk  These predictions accord fairly well with the facts of Finnish, but not completelys. Incorrect predictions are listed below: (5)attested sequences non-attested sequences predicted impermissible predicted permissible  C- Caps ts ks CPC- *hp *sp #mv . *nv  C° >C° nh rh ns ms C"<C' *ln *rn *rl *hs C- C-  tk 6  The distribution of these unexpectedly good sequences varies. The C-s sequences ps, ts, ks, ns are quite common, (e.g. lapsi 'child', katsoa 'to look', maksa 'liver', kansa 'people'), as is the sequence tk, (e.g. jatko 'continuation'). The sequences nh and rh are rare, but they do occur in e.g. vanha 'old', harha 'delusion'. Interestingly, the unexpectedly good clusters involve either [s] in the second position or another coronal sound in the first position. [s] is well-attested for clustering in languages where no other sounds may and in ways contrary to its position 20  in sonority hierarchies. In English, for example, CCCC sequences are possible in words such as demonstrate. If another sound is substituted for the s, the result is invariably bad (*demonktrate, *demomptrate). Similar effects are observed in Finnish (e.g. pyrstO 'tail of a bird', but *pyrtk(5). To my knowledge, this special status of [s] remains unexplained not and I willA pursue the subject here beyond noting the problem. The sequence as is rare (but occurs in e.g. jamsi 'rasp'). This comparative rarity is not so remarkable as one might first think. In Finnish, [m] appears in syllable-final position only before [p] and is often derived by assimilation of /n/. Word-final Intl surfaces as [n] (e.g. koditon 'homeless' nam. sg . versus kodittomat 'homeless' nam. pl.) 7 . The sequence tk is not so easy to deal with, as C - C- sequences are not permitted by the theory. Kaye, Lowenstamm and Vergnaud (1989, 59-62; 1987 30-32) allow for the possibility of a series of charmless (and therefore governable) stops since stop-stop sequences obviously do occur in many languages. I hesitate to suggest that Finnish is such a language for the presence of a charmless stop series which included, e.g., the sound to would predict the sequences *tp, *tv in addition to the acceptable sequence tk. I shall leave the question of the occurrence of tk in Finnish unresolved, pending further research. In addition to the CsC clusters mentioned above, Finnish exhibits another type of CCC cluster in which a sonorant is followed by a geminate obstruent. Examples include helppo 'easy', kynttilä 'candle' and poltta 'burn'. I shall deal with the structure of these forms in Chapter Five, for they offer a problem for the theory. However, the segmental content of these clusters never violates the sequencing restrictions discussed above.  21    The unexpectedly bad clusters present some curiosities. The lack of clusters ending in p is a puzzling but well-known idiosyncracy of Finnish (Prince 1984, 239). The absence of *nv, Amy, *ln, *rn, *rl may be a sonority effects. As my purpose in this section is to provide a list of possible consonant sequences in Finnish rather than to explain the distribution of observed sequences, I will not pursue the issue further. 2.2 Stress Government Theory does not deal directly with stress but it will be seen that stress interacts with other prosodic processes in Finnish. I will deal with these cases as they arise and present the basic facts of Finnish stress here in terms of the metrical theory of Hayes (1980). In Finnish, primary stress falls on the first syllable of a word; secondary stress (which is optional in nouns) falls on subsequent oddnumbered syllables (Whitney (1956, 16), Lehtinen (1962, xvii), Keyser and Kiparsky (1984, 23-25)). In terms of Metrical Phonology, a left-headed binary Foot is erected at the left margin of a word (the application is reiterated if secondary stress is applied). The stress patterns for the two pronunciations of sammalena 'moss' essive singular are represented: (6) without secondary stress  w w w sammalena  with secondary stress  w s w sammal ena  Both Whitney and Lehtinen (op. and loc. cit. above) mention that in words of five or more syllables, secondary stress is shifted to the fourth  22  syllable if the third syllable is short. The example Lehtinen gives is Oppettamatomon (gloss not given). I will not be dealing with any examples of this type which are, in any case, more a consideration for Metrical Theory than for Government Theory.  23  Notes to Chapter Two 1. As Finnish spelling is very nearly identical to phonetic transcription, I shall be citing Finnish forms in standard Finnish orthography. I must, therefore, note some differences. Among vowels, the IPA equivalents of a and 6 are m and m respectively. The finnish alphabet also includes the letters <d> and <g>. Outside of their occurrence in loanwords, <d> represents a voiced dental stop, but <g>, where it occurs in the sequence <ng> represents part of a geminate velar nasal. Thus, the weak grade form Helsingin (gen. sg . of Helsinki is pronounced [helsiuuin]. Lehtinen includes /d/ and /u/ among the phonemes of Finnish, but these two sounds are derived by Consonant Gradation and are of limited distribution.  2. The representation for [h] is not included among the consonantal representations of Kaye, Lowenstamm and Vergnaud (1989). On the basis of its continuancy being its most salient feature, I have chosen to represent it as O. The sound of [h] varies according to position. Intervocalically it can be quite lightly articulated (Karlsson 1983, 17). Before consonants I have heard it came close in sound to a velar fricative. Acoustic studies reported in Shuken (1984, 124) suggest that in the Finnish [h] the glottis is open wider than for whispered vowels. See the last section of Chapter Five for same interesting hC interactions. 3. I have excluded j from the consonant clusters discussed for two reasons. First, j does not occur before other consonants (one finds i instead) and second, j is often derived from i by glide formation. Glide formation which occurs 1) where the plural morpheme /-i-/ occurs intervocalically (talo 'house' + -i- pl. + -(e)n gen. ---> taloja) and 2) where a non-low front vowel spreads onto an Onset left empty by deletion (e.g. arki 'weekday' + -n gen. ---> arjen--see Chapter Six). Not all Cj sequences are derived by glide formation. Sane examples include kirja 'book', kalju 'bald' and tyhjyys 'emptiness'. With the exception of loanwords such as anjovis 'anchovy', j is restricted to appearing after 1, r and h--the least camplex sounds in Finnish. 4. Nasals regularly assimilate to following stops, so I have not included sequences such as mk, mt, np etc. 5. See Suomi (1985, 220-1) and Prince (1984, 239) for partial lists of illicit Finnish consonant sequences. A dictionary search has failed to turn up evidence for certain clusters and I have provisionally included them among the list of non-permitted sequences. These include *hr, *hl, *hs, *mh, *sh, *lh, *sl, *tv, *kv, *vt, *vk. The theory predicts that the absence of these clusters is a systematic gap, but I do not have enough evidence to evaluate this claim fully. Certain illicit sequences do appear in compounds (where government relations do not hold) and in unassimilated loanwords (cf. Suomi 1985, 220-1). 6. The sequence kt also occurs, but only in underlying forms and in a fairly limited distribution. The underlying /kt/ sequence is always prevented from surfacing either by lenition of the /k/ or palatalization of the /t/. For example, the form /kakte/ 'two' appears as kaksi in the nominative singular, as kanden in the genitive singular and as kahtena in 24  the essive singular. 7. Word-final consonants must be coronal in Finnish (Whitney 1956, 15). In the modern language, only the nim alternation is attested; word-final non-coronals simply do not occur. An approach such as ItO's Coda Condition (It8 1986, 26 and passim) might be co-opted, pending a satisfactory resolution of the question of non-coronal stop+s clusters. In Government Theory, it might take the form of a prohibition against non-coronals (segments lacking the element RP) occupying the right branch of a Rime when they are not Interconstituent Governed by some segment. 8. Sonority and complexity are very nearly analogous. In the present framework, a difference in complexity is all that is required to permit a sequence. In some sonority analyses (e.g. Selkirk 1984) a sufficient difference may be required to permit a sequence. Sonority differences leave unexplained the impermissibility of of the nasal+v sequences since v, being negatively charmed, does not require a complexity difference to govern a charmless segment. The labiality of v may be involved in this impermissibility in same, as yet, unclear fashion. If this is so, p and v are most likely patterning together.  25  CHAPTER THREE - The Deletion of Segments This chapter examines the constituent structure of Finnish lexical forms and provides a Government-based account of certain deletion processes that are sensitive to constituent structure. 3.1 Two Segmental Deletion Processes Keyser and Kiparsky (1984) present an interesting and insightful account of some camplex facts of Finnish syllable structure. Two processes that receive particular attention are what the authors term e-Deletion and t-Deletion. The set of data in (1) presents the processes of e Deletion -  and t-Deletion (Keyser and Kiparsky 1984, 9 & 11): (1)  'house"glass"fire"water"clay " snow " moss " root nom. sg . part. sg . nam.pl. part. pl. gen. sg. ess. sg .  /talo/ talo taloa talot taloja talon talona  /lasi/ lasi lasia lasit laseja lasin lasina  /tule/ tuli tulta tulet tulia tulen tulena  /vete/ vesi vett& vedet vesid veden veten&  /save/ savi savea savet savia saven savena  /lure/ lumi lunta lumet lumia lumen lumena  /sammale/ sarrmal sammalta sammalet sammalia samnalen sammalena  The effects of t-Deletion may be seen by camparing the partitive singular with the partitive plural. The partitive suffix is /-tA/, but when it is affixed to a polysyllabic vowel-final stem, the t does not surface (e.g. talo + -tA --> taloa) 1 . The rule of e-Deletion syncopates syllable-final /e/ at morphological boundaries (except in word-final position in disyllables). Where eDeletion has created a consonant-final allomorph, t-Deletion does not apply and the /t/ surfaces, as in the partitive singular forms tulta, vettd, lunta and samnalta. 2 .  26  In Keyser and Kiparsky's analysis, t-Deletion and e Deletion are -  formulated as in (2): (2) a.  t-Deletion oi# \  C  	/  --> 0 / c v  b.  e-Deletion  V I  v  1--> 0 /  .  :  t  e  cr  \  \  Where 0;1 denotes an unstressed syllable. As the rules are applied cyclically, intervocalic t and syllable-final e  are not deleted within stems. Furthermore, it must be stipulated that eDeletion is blocked in the nominative singular of disyllabic e-stems, and where an illicit cluster would result. For example, /sammale + -nA/ ---> sammalena 'moss' ess. sg . (*sammalna); /save + -tA/ ---> savea 'clay' part. sg . (*savta). The rules must be ordered so that the application of e-Deletion bleeds the potential application of t-Deletion (without this ordering, the derivation would proceed: /saveta/ ---> /savea/ ---> *sava). 3.2 e-Deletion In Keyser and Kiparsky's analysis, the rule of e-Deletion is interpreted as the breaking of an association line from the segment /e/ when it is found in syllable-final position, even when doubly linked 3 : (3)  a) V 4 T  e  b)  V C V \ e  27  Such rule applications were explicitly argued against by Hayes (1986) where the Linking Constraint was introduced: (4) Linking Constraint [Hayes 1986, 331]. Association lines in structural descriptions are interpreted as exhaustive. If we accept the validity of the Linking Constraint (and Hayes presents a strong case) then the e-Deletion rule cannot be made to do double duty in this fashion. Furthermore, the number of conditions on the e-Deletion rule (it cannot apply word-finally in disyllables or before the essive suffix /-nA/, it may not create illicit clusters) make it a suspicious sort of rule. I will therefore argue in the following section that the effects of e-Deletion are achieved through the action of a specific deletion rule, but fall out from the interaction of the prosodic theory and the structure of these roots. 3.3 EMpty Categories and Phonological Government In the theory of Phonological Government, deletions of segmental material do not affect the government relations within the skeleton. If a melody element (segment) is deleted, its skeletal slot remains. The distribution of empty skeletal positions is restricted by the Empty Category Principle: (5)  Empty Category Principle (ECP) (Kaye 1987, 6) Empty skeletal positions are subject to a special form of government (they must be properly governed).  28    Proper government is defined: (6)  Proper Government (adapted from Kaye 1987, 11) A skeletal position A is properly governed by another skeletal position B iff: i. B governs A ii. B has phonetic content iii. No other government relation intervenes between A and B.  The proper government relation that is relevant to the deletion of vowels is that which holds between adjacent Nuclei (Projection Government). If in a language L, the direction of projection government is right to left, then the Nucleus dominating the skeletal position of a deleted vowel must be preceded immediately on the Nucleus projection by a Nucleus that dominates a phonetically realized vowel: (7)  N ----> N 1	o 1 1  N a1  XXXXX 1 1 C  1 V  1 I C  X 1 1 C  I V  The underlying form of this hypothetical example would be /CVCVCV/ and the surface form would be [CVCCV]. In just these circumstances, the two consonants that came together as a result of the vowel deletion are not in a government relation. It would therefore be possible to create a cluster by vowel deletion that would not otherwise be tolerated in the language 4 . This cannot be the strategy of Finnish e-Deletion. If the final e of e-stems remains unrealized because its skeletal position is properly  29  governed, then the representation of e.g. tulta 'fire', part. sg . would be: (8) (Projection Government)  N---->N N 1 1 : XXXXXX 1 1  1 1  1 1  1 1  1 1  1 1  tulOta Notice that the 1 of the root and the t of the suffix are separated by an empty skeletal position; therefore, no government relation holds between them. In exactly this case, it is to be expected that superficially illicit clusters may occur--that *savta (from /save+tA/) should be wellformed. Since *savta is in fact ill-formed, it must be the case that there is no empty Nucleus separating the two consonants. The structure of tulta must be: 0 R 1 1\  (9)  N \ 1 : \ X X X II : 1 t u 1  0 R 1I 1  : N : II X X II II t a  In order to generate this structure, I propose that the root-final e is not associated with a skeletal position at the level of Underlying Representation: (10)  X X X II		 II		 II tule  30  The affixation of the partitive suffix /-tA/ takes the form: (11)  X X X  N  affixation  XX  II 1	 I tule  II	 I t a  I  N  projection of N and establishment of ICG relations  X X X X X 	 I	 I I I II II tuleta tI 1_1 I +1_1  0 R 0 R 1	 : \ 1	1 1 1 N \ , N I1 ,	 .	 :	 \ 1	 I X X X X X iI		 II I	I II		 1I t u 1 e t a  projection of 0 and R (first R picks up governed segment)  ,  tulta  The e being excluded from the prosodic structure fails to surface and is ultimately deleted by stray erasure. All things being equal, it might be expected that the surface form of (10) (the nominative singular) would be *tul rather than tuli. However, Finnish nouns have a minimal word length requirement of a metrical foot (see McCarthy and Prince, 1986, 2). In terms of the present theory, this means that the prosodic structure of nouns must conform (minimally) to the template OROR5 . In erecting prosodic structure according to this template, the skeletal slots will map onto the OR tier one to one (according to the charm values of their segments): (12)  0 R 0 R 1 1 1 1 1  N 1 1  X X X 1 1  1  tule  31  In order for the noun to be well-formed, the final Rime must came to dominate a segment. The necessary skeletal slot is inserted will be inserted to satisfy the OR template, the floating e associates to it, projects a Nucleus and is incorporated into the prosodic structure: (13)  O  epenthesis of skeletal slot  R OR  X XI X X < I  tule O  association of /e/ and projection of N  R OR ,  ININ  XXXX  tule O  incorporation of N into R  ROR  X X X X  tule  > tu l i  Forms to which the essive suffix /-nA/ is attached offer a peculiar problem. The stem-final /e/ always surfaces. Thus, the essive form of /tule/ is tulena and not *tulna. This comes as no surprise, for it has been established that Finnish does not allow any Cn clusters (see Chapter Two). However, the behaviour of the essive suffix cannot simply be ascribed to an aversion to generating Cn sequences. Finnish has a class of nominal roots such as perhe 'family' with the underlying form: (14)  X X X X X X I	 II		 II		 II		 1I perhe  32    When affixed to these roots (which are discussed in Chapter Five), CV suffixes such as the partitive /-tA/ normally geminate. For example, the partitive form of perhe is perhetta. The essive form is not *perhenna but perheeng. The /n/ of the essive suffix fails to spread backwards to fill the empty skeletal position and the root vowel lengthens instead: (15)XXXXXXXX :	 1 1/ 1	 1 perhe+na  *XXXXXXXX 1 : \I p e r h e+ n a  If the suffixal /n/ were in a government relation with the root-final empty position, it would be a case of proper government and the empty position should remain empty. As the vowel spreads onto it, the empty position must be governed by the vowel. Normally, head-final government relations are established before head-initial relations (see the Chapter One discussion of the generation of prosodic structure). The /n/ of the essive suffix therefore systematically fails to govern a preceding position6 . The derivation of tulena 'fire' ess. sg . proceeds as in (16). There is a crucial difference between this derivation and that in (11). In (11) the /1/ of the root was Interconstituent Governed by the suffix-initial consonant. That government relation established the /1/ as a non-head and therefore not within an Onset. In order to satisfy the OR template, the non-head had to be incorporated into a Rime. In the derivation in (16), the /1/ of the root is not Interconstituent Governed and so is eligible to become a head. It therefore associates to an Onset node. Additional Rime  33    and Onset nodes must be inserted to accommodate the governed /n/ of the suffix. (16)  X X X I II I t u 1 e t	 I  X X I n a t  affixation and establishment of ICG relations  N N 11 11 X X X X X I	I	 1I :	 II I	 I t u l ena  projection of N  O R 0 R II : I	 I	 I 1 ,	N N II I		 I	I 1I X X X X X II	 I	 I	I :	 II tulena  mapping (one to one) onto OR template  O R 0 R 0 R 	 11	 i	1 11 11  1 N 1 I N 1	 1	 i 11 X X X X X 11		 1I	 I 	 I tulena  IC governed /n/ mapped into an Onset; additional R inserted to satisfy wellformedness constraint on OR tier  O R 0 R 0 R 1	1	 1	 1 1 I I I 1 N 1, N N 1 	 1 	 1	 1 1	II	 1Jim XX X X XX IIII IIII'1 I t u l ena  floating /e/ incorporated into free Rime (by insertion of X and projection of N)  The derivation of forms such as savea 'clay' part. sg . (underlyingly /saveta/) proceeds in similar fashion. The v - t - sequence is not a possible government configuration, so the /v/ associates to an Onset as the /1/ of tulena does. The /t/ undergoes 't-Deletion'. Stems longer than two syllables which end in /e/ (e.g. /sammale/ 'moss') corroborate the assumptions I have made above. In the nominative sammal, the minimal word restriction is met without having to realize the 34  /e/. In the partitive sammalta, the government relation between the /t/ and the /1/ defines the /1/ as a non-head. The /1/ thus becomes part of a Rime. All government conditions are satisfied without realizing the /e/. In the essive form sammalena the inability of the /n/ to govern the /1/ allows the /1/ to associate to an Onset. The /e/ surfaces as it does in the derivation of tulena, above: as a Rime separating two Onsets. 3.4 t -Deletion The above analysis eliminates the need of a rule of e-Deletion in the grammar of Finnish and attributes the observed effects to the structure of the roots involved. In the case of t Deletion we have a genuine rule of Finnish that -  deletes intervocalic t. Keyser and Kiparsky's (1984) rule of t-Deletion is repeated in (17): (17)=(2a)  t-Deletion / \  --> 0 /CVV t  Where alq, denotes an unstressed syllable. This rule makes crucial reference to the notion 'unstressed syllable'. This weak syllable does not actually dominate the target segment of the rule but dominates the segment adjacent to it. This is a curious condition for a rule. What the condition really states is that the syllable of  35  which the t is in the onset must not be dominated by the weak member of a metrical Foot--i.e. it must not be in this configuration: (18)  *  F /\ / \  s w OR OR I	 11		 11 CV tV Viewed this way, the condition makes more sense, for the target segment is now within an identifiable domain. The course of investigation must now be to determine the process whereby t deletes and what there is about the configuration in (18) that blocks that process. Recall the discussion of the Empty Category Principle and Proper Government above ((5) and (6), repeated here): (19)=(5) Empty Category Principle (ECP) (Kaye 1987, 6) Empty skeletal positions are subject to a special form of government (they must be properly governed). (20)=(6) Proper Government (adapted from Kaye 1987, 11) A skeletal position A is properly governed by another skeletal position B iff: i. B governs A ii. B has phonetic content iii. No other government relation intervenes between A and B. Formulated another way, these principles may be regarded as stating that properly governed skeletal positions may be left phonetically unrealized. Obviously it cannot be that any properly governed skeletal position will be left empty, but it does imply that if a language specific rule does delink segmental information froth a skeletal position, the rule will not apply if the position is not properly governed.  36    Given the ECP, the formulation of the t-Deletion rule is quite simple: (21)  t-Deletion 0 X 4  t  condition: t is not dominated by the weak branch of a metrical Foot  The ECP and Proper Government define most conditions on this rule. Only intervocalic t will be affected. In (22a) the government relations of an intervocalic t are shown. In (22b), the Interconstituent Government relations of t preceded by a Rime-final sonorant are shown: (22)a. R II N 1I X II V  0 R 1I		 II 1 N 11 I	 I X X II		 I t V t	 I '—'  b. R 0 R I	I	 II :\ N \ 1 N II		 iI :	 \ X X X X II II		 II		 11 V r t V .1.	1	 I1		 11 t	I —,I  In (22b) the government relation between the t and the r violates condition (iii) of the definition of Proper Government. The Interconstituent Government relation between the t, and the r precludes Proper Government of the t by the following vowel. The condition on the stress configuration must, unfortunately, be stipulated as a condition on this rule alone. While it is tempting to suggest that the Foot defines a government domain so that consonants falling within that domain are not Properly Governed and therefore cannot be deleted, this cannot be the case. If consonants within metrical Feet were never Properly Governed, or if this condition were generalized beyond  37  the context of this rule, the deletion of k in the weak grade forms of disyllabic nouns such as joen ('river' gen. sg . from /joke+n/--see the discussion of Consonant Gradation in the next chapter) would not be possible. Keyser and Kiparsky (1984, 23-25) demonstrate an interesting optionality to t-Deletion that makes it clear that it is stress (rather, say , than the number of syllables in the word) that interacts with t-Deletion. The Finnish nominal system has two options for stress assignment. Primary stress may be assigned to the first syllable of a word with weak stress on the remaining syllables or primary stress may fall on the first syllable and secondary stress on the third. If the optional secondary stress precedes a t-initial syllable, the t is not deleted. Thus, the partitive singular of e.g. lokero 'pigeonhole' nay be either 16keroa or 16kerata: (23) a.  18keroa 'pigeonhole' part. sg .  / \  /  \  \  \  \  b. 16kerOta  idem  / \ \  /  / \ / \  \  S w w w OROROROR  s w s w OROROROR  11111111 11111111  11111111 1111111  X X X X X X X X 11111111 11111111  X X X X X X X X 11111111 11111111  1  ok  e  r  o  f t  a  1  r  o  t  a  In both (23a) and (23b), the position occupied by the suffixal t is properly governed by the following a, but in (23b) it is in the domain of a Foot and may not delete whereas in (23a) it is not within a Foot and may delete. The interaction of stress and Government is an area where little  38  research has been done. Further investigation may reveal the details of this interaction, but for the time being, the stress pattern must remain a stipulation on the t-Deletion rule.  39  Notes to Chapter Three 1. Finnish has a vowel harmony system in which suffix vowels must harmonize with root vowels for backness. Accordingly, I have used an archisegment in representations of suffixes that include harmonizing vowels. 2. The raising of final /e/ to [i] in non-doubling roots is a regular feature of Finnish. There is only one non-doubling root ending in e (nukke 'doll') that I have found not to do this. 3. In an interesting parallel to the behaviour of e-final roots (see preceding note), virtually all of what Keyser and Kiparsky call 'contracted stems' (in my Chapter Five discussion, I call them 'doubling roots') surface with a word-final short [e]. The behaviour of these forms reveals a different underlying structure, however. In Keyser and Kiparsky's analysis, e-raising roots end in a short /e/ while contracted stems end in a long /e/ which is geminated around an empty C-slot (Chapter Three, (3b)--the C-slot is required for certain consonantal doublings associated with these roots). They therefore attempt to unify the behaviour of all e-final roots by having both undergo e-Deletion. They are certainly right in seeing a connection between these two types of root and it must be no coincidence that virtually all e-final roots in Finnish behave in one of these two ways. 4. Kaye (1987, 17 ff.) demonstrates that vowel deletions of this sort do create otherwise impermissible clusters in Moroccan Arabic. 5. Monosyllables with long vowels (e.g. luu 'bone') vowels comprise a foot as they are bimoraic. It may also be that all superficially long vowels in Finnish are heteronucleic having the form (see Chapter Five): 0 R 0 R II I X X X \ / 1 The nouns with light diphthongs such as tie 'road' present more of a problem as they are arguably monomoraic and may constitute a genuine exception. Campbell (1981, 175), however, demonstrates that these diphthongs (which are derived historically from long vowels) pattern with long vowels in the language game kontti kieli. For example, the disguised form of the kieli 'language' is kooli kientti rather than *koli kientti. But, see my discussion of CVVC sequences in Chapter Five for evidence of a monamoraic interpretation. 6. Whether the non-governing behaviour of this suffix represents a morphological aspect of the suffix or a phonological aspect of the Finnish /n/, I cannot say. Geminate n does arise through the gradation of /nt/ sequences (e.g. ranta 'shore' nam. sg . rannan gen. sg .) but these geminates are derived by left to right assimilation and not by right to left spreading.  40  CHAPTER FOUR - Consonant Gradation In this chapter, I shall examine the phenomenon of Consonant Gradation in Finnish and present arguments for analyzing Gradation as a prosodically determined process. 4.1 General Features of Consonant Gradation Consonant Gradation is a process of what is traditionally described as 'weakening'. Because the effects of weakening processes have a wide variety of surface manifestations it is impossible to characterize Consonant Gradation as the addition of any specific set of elements (or features) to the phonological representation of the affected segments. In Finnish, these weakenings range from degemination, to voicing (with or without spirantization), to total assimilation, to deletion. In short, weakening affects both the internal structure of segments and their prosodic relationships with other segments or positions. Examples (fram Karlsson 1983,38) of gradation in the nominal system triggered by case suffixes of the form -C or -CCV are given in (1): (1)  pp ---> p tt ---> t kk ---> k p ---> v t ---> d k ---> 0 mp ---> mm nt ---> nn t)k ---> 4n It ---> 11 rt ---> rr  strong grade kauppa katto takki kipu aiti joki kampa ranta kenka kulta kerta  weak grade kaupassa katolla takissa kivussa aidille joesta kammalla rannalla kengasta kullaksi kerran  gloss 'shop' 'roof' 'coat' 'pain' 'mother' 'river' 'comb' 'shore' 'shoe' 'gold' 'time'  nam. nan. nan. nan. nan. nam. nam. nan. nam. nan. nan.  sg./iness. sg sg./adess. sg . sg./iness. sg . sg./iness. sg . sg./allat. sg sg./elat. sg . sg./adess. sg . sg./adess. sg . sg./elat. sg . sg./transl. sg . sg./gen. sg .  41  4.2 The Nature of Consonant Gradation Finnish Consonant Gradation is superficially similar to a number of processes attested in other languages and generally subsumed under the name 'mutation'. Lieber (1987; 72) defines mutations as "phenomena in which lexical stems exhibit two or more allomorphs that differ in only a single marginal segment (for example, an initial or final C or a vowel closest to the end of a word) and which appear in distinct morphological, syntactic, or phonological environments." One well-known mutation that shares strong similarities with Finnish Consonant Gradation is the Welsh lenition mutation. Same of the alternations of these two processes are compared below: (2) Finnish Gradation  Welsh Lenition  PP - P tt t kk k  p b t d k g  p v t d k 0  b v d g - 0  In (2), a series of 'strong' stops (geminate in Finnish, voiceless in Welsh) alternate with a series of 'weaker' stops (single in Finnish, voiced in Welsh). A second series comprising the weaker segments of the first series of alternations alternates with what must historically have been a series of fricatives'. In both languages the modern reflex of the velar fricative is 0. Lieber (1987) analyzes mutation as the addition of a floating autosegment to the specification of the affected consonant.  42  Thus, in a Welsh example such as  barged  [basged] 'a basket' -  y fasged  [@  vasged] 'the basket' (the symbol '@' represents schwa), Lieber's analysis would (using her notation) take the form: (3)  V  Mutation Tier  C  11 :+antl -cor +lab -cont:  VCCVC a s g e d  :+voi :+cont:  The above account will derive voiced fricatives fram their corresponding stops (allowing, of course, for the overwriting of the prespecified [-cont] feature of the stem-initial consonant) 2 . Such an account cannot, however, be transferred to Finnish. Consider the following examples of Finnish Consonant Gradation: (4) strong grade  weak grade  ratti 'wheel' rattina rattiin  rata 'track' ratana rataan  nominative sg. essive sg. illative sg.  ratin ratiksi  radan radaksi  genitive sg. translative sg.  The first difficulty encountered is that the target consonant is separated from the case suffix by a vowel. A Lieber-style analysis would assume that the first component of the suffix is the mutation autosegment. All of the consonantal mutations Lieber discusses involve consonants adjacent to the mutating morpheme. She allows (p. 96) for the possiblity of non-local mutations by placing the mutating autosegment on its own tier. Such a tactic is no longer considered viable in light of the development of the feature-hierarchy model (see, for example, Clements 1985). 43  The degeminating paradigm (ratti - ratin etc.) offers more serious problems. Degemination is not a process normally triggered by the addition of a feature to a segmental representation. Indeed, in the light of standard autosegmental representations of geminates, it is unprincipled to posit a feature that deletes one of the association lines of a doubly linked segment: (5)  X X  	/  X X    	A  /    A [ 4-7 ]  A third problem with the Lieberian approach is that if one assumes  that the putative autosegment can spread across short vowels, why can it not also spread across long vowels. The illative case morphololgy generates long vowels and Gradation fails to apply (see (4)). This failure of Gradation in illative forms is not simply a property of the illative case suffix, for Gradation systematically fails before long vowels. Compare rata 'track' with taide 'art' 3 : (6) rata 'track' radan radaksi ratana  taide 'art' taiteen taiteeksi taiteena  nominative sg. genitive sg. translative sg. essive sg.  In taide and in all nouns that lengthen their final vowels, Gradation is not observed even when the suffix involved triggers Gradation after short vowels. Since in autosegmental representations there is no distinction between long and short vowels on the melodic tier, it cannot be the case that a putative mutating autosegment is blocked from association by a long vowel but not by a short one. If the trigger of Consonant Gradation is not to be found in the  44  segmental structure, then where is it? A descriptive answer may be found in almost any discussion of Finnish Gradation, whether technical or pedagogical. Here are sane examples: (7)  i) In polysyllabic stems z t, k are subject to Consonant Gradation if they are followed by an ending which a) consists of only one consonant or b) begins with two consonants, and also on condition that c) between Et, t, k and the ending there is only a short vowel or a diphthong (not consonants or a syllable boundary) (Karlsson 1983: 31) ii) If a short syllable beginning with k, p or t is closed by the addition of a consonant, then the k, p or t undergoes certain changes called 'mutation' or 'softening'. (Whitney 1956: 21-22) iii) The strong consonant grade occurs in open and/or long syllables; the weak consonant grade occurs in short closed syllables. (Lehtinen 1962: 523)  iv) . . . consonant gradation . . . voices t (and weakens p, k in various ways) and degeminates double stops before a branching rime, that is, in the onset of a closed syllable. (Keyser and Kiparsky 1984: 15) The generalization common to all four of these descriptions is that Consonant Gradation applies to a consonant in the onset of a short closed syllable. In other words, what is relevant to the action of Consonant, Gradation is not the concatenation of certain morphemes, but the syllable structure that results fram the addition of suffixes to a stem 4 . If this generalization is valid (and I believe it is), then any analysis of Consonant Gradation must be expressed (at least in part) in terms of syllable structure. Finnish Consonant Gradation may thus be defined as a process 45  of phonological weakening conditioned by certain relationships among syllabic constituents. 'Phonological weakening' is an elusive concept. I have argued above that the kinds of weakening observed in Finnish are not easily accounted for in a model (such as Lieber's) that treats mutations as the addition of a set of features to the represenation of segments. The elemental theory of Kaye, Lowenstamm and Vergnaud (1989) fares little better for much the same reasons. Several proposals for dealing with weakening have been made, including, for example, Foley (1977), Vennemann (1988). These theories have not received widespread acceptance, due partly to their preference for treating segments as holistic units rather than as feature complexes. I am not prepared at this time to propose a theory of weakening, but I will make a few remarks on how I see weakening being handled within the framework I am using. Weakening occurs when a susceptible segment appears in a weakening environment. In Finnish, such an environment is (informally) the onset of a closed syllable. More properly, weakening will affect the segment dominated by an Onset which is Interconstituent Governed by a heavy (branching) Rime: (8) 0 :\ N \ X X X y z An appropriate segment in this position is subject to a number of restrictions. If it is doubly linked (i.e. geminate) it will delink one of its association lines (e.g. ratti ratin) 6 . If it is singly linked to a  46  skeletal position it becomes transparent to spreading of voicing and continuancy from the vowel which heads the Rime that governs it (e.g. kipu kivun). If it is linked to another segment (e.g. if the preceding segment shares elements with it by way of assimilation) then it will assimilate to that segment totally (e.g. ranta rannan). The preceding is admittedly a tentative proposal which awaits the elaboration of a fully developed theory of weakening. It is sufficient for present purposes to say that certain consonants in Finnish weaken in a given context (before a branching Rime) and when they do weaken, they weaken in the ways described. Of more significance is the distribution of this context for Gradation and the question of what constitutes a branching Rime in Finnish. It is to these questions that I turn in the following chapter.  47  Notes to Chapter Four 1. The Finnish /d/ does not pattern completely with the other weak grade stops. It would be expected (in Foley's (1977) strength hierarchy, for example), that if the /p/ had spirantized, so would the /t/. In standard Finnish, as described by Karlsson (1983) Whitney (1956) and Lehtinen (1962) the /t/ is voiced but not spirantized. However, Whitney (1956; 13) reports that [d] is "sometimes so soft that it is hardly heard". 2. A word must be said about velar fricatives. Lieber treats this subject in a footnote (Lieber 1987; 129, n. 36): "The segment [g] must be treated as utterly exceptional in any framework. Here we might assume a special rule which deletes the [y] that has been created [by the lenition process] " I confess to being suspicious of special rules that delete derived sounds that do not otherwise occur in a language, particularly as velar fricatives seem to be the most frequent victims of such rules. Foley (1977; 30-33) discusses exactly this propensity of velars across languages to be the weakest of segments in strength hierarchies, so it would appear that more is happening than the generation of a segment that violates the phonotactic constraints of a particular language. Just what is happening is not clear and Foley's calculus of strength values does not translate into terms of distinctive features or phonological elements (in Foley's framework, numerical values are derived for the various segments and rewrite rules assign than phonetic values). Foley does have same useful insights, however, for the concept of 'weakening' implies a change in the relative strength of a sound, not in its makeup. The difference between the strong and weak reflex of a sound is expressible in terms of their difference in distinctive features, but as Finnish demonstrates, the difference between all the strong reflexes and all the weak reflexes in a language do not necessarily boil down to a difference in the same set of features. Since Finnish deletes derived [y] I cannot ignore the problem entirely, particularly as the site of deletion has significant phonological effects. However, my discussions will be confined to the licensing of these deletions and their phonological effects. I will leave it to another to develop a full analysis of phonological weakening. 3. taide is a member of a class of noun roots that lengthen their final vowel in certain instances which will be discussed in Chapter Five. At the risk of getting ahead of my presentation, I must point out that the graded form taide conforms to the general pattern of Finnish Gradation. Such nouns derive historically from consonant-final forms and it is the remnant of that lost consonant that accounts for both the vowel lengthening and the unusual gradation pattern, as will be seen. 4. This generalization provides a fourth argument against a Lieberian approach (and I, in all fairness, must point out that Lieber makes no attempt to account for Finnish data in this fashion and almost certainly recognizes that this is not the way to go about it). It is curious, indeed, that the set of suffixes which have the floating autosegment is  48    exactly the set of suffixes that generate stem-final short closed syllables. There is, however, a set of exceptions to this generalization: the possessive suffixes -nsA (3 sg.), -mme (1 pl.), -nne (2 pl.) and -nsA (3 pl.). These suffixes do not trigger Gradation: rata 'track' radaksi radalla radassa  nom. sg . transl. sg . adess. sg . iness. sg .  ratansa nam. 3 sg./pl. poss. ratanne nam. 1 pl. poss. ratanne nam. 2 pl. poss.  The likeliest explanation for these exceptional suffixes comes from Lexical Phonology. If Consonant Gradation is restricted to level I (where case suffixation can be expected to occur) then possessive suffixes (which go outside case and number suffixes) belong to level II phonology where Gradation does not apply. 5. This wrong one provide a the Welsh  is not to say that I think Lieber's approach is necessarily the to take in the cases she discusses. It would be useful to unified account of mutations that included the Finnish type with type, however.  6. The singly linked segment thus created would not be subject to further weakening for there is a crucial difference between a singly linked segment derived form UR and one that has undergone degemination. In the latter case, the position of the segment remains a governor for the empty position created by degemination. This government relation will resist further weakening: a) singly linked R 0 R N  [  \ N \  \ X X X X I1 1	I I I VCVC  b) degeminated R  0 R i i\ N \ , \ 1	 \ i i	 \ X X X X X 1 I II II V C V C t 1 government domain \  49    CHAPTER FIVE - WC and VCC Groupings In this chapter I will examine some putative syllable types in Finnish and demonstrate that superficial structure in Finnish is not representative of underlying structure. 5.1 CVVC and CVCC Non-syllables In the discussion of the theory of Phonological Government (Chapter One, above) I presented a list of possible constituents from which the following inventory of 'syllables' (Onset-Rime sequences) can be derived: (1) Inventory of OR pairs  a . 0 iI : II X  R I	I N II X  S'OG ICG t__: CV  b.  c.  d.  e.  f.  0 I	I : I X  0 : : 1 X  0 R 1	I :\ N : \ 11 :	 \ X x X t1 11 t	I_I	1  0 R :\ I1 N : \ :	 \ :\ X xXx 11 t	1	 11_1 	 t t	1._..1	1  0 R 1\ 1\ N \ I \ :	 \ 1	 \ XxXx 11 t	1	 11 t1 t	1_11  R I	I N 1\ X x I1_1	t t	i_i	1 CV: CVV  R :\ N \ :	 \ X x t1 1 t	1_1	1 CVC  CCV  CCV: CCVV  CCVC  As Finnish lacks branching Onsets (Karlsson 1983, 19; Whitney 1956, 15), a'H only (la-c) are relevant to the present discussion. Nevertheless, the schemata CV, CVV and CVC do not at first glance appear to suffice for an  50  analysis of Finnish. Specifically, Finnish exhibits what seem to be CVCC and CVVC syllables': (2) a.  b.  CVCC 'syllables' salskea Pyrstii konsti  'slender' 'fish- or bird-tail' 'trick'  Polttaa kynttila monarkki lammu helm) kanssa  'to burn' 'candle' 'monarch' 'lamp' 'easy' 'with'  CVVC 'syllables' muuttaa muuntaa kiille kiilto saasta  'to move' 'to transform' 'mica' 'lustre' 'filth'  taloon kaupunkiin kuntaan  'house' illative sg. ill. sg . 'town' 'commune' ill. sg .  ranteeksi ranteiksi rannoiksi taiteet terveen perkeleettá  'wrist' translative sg. 'wrist' transl. pl. 'shore' transl. pl. 'art' nom. pl. 'healthy' gen. sg . 'devil' partitive pl.  5.2 CVCC Sequences The motivation for claiming the existence of CVCC syllables in Finnish is the presence (on the surface) of CCC clusters (Prince 1984, 239; Ito 1986, 40-47). Both of these authors recognize the restrictions on clusters of this type: they are always either sonorant+s+stop or sonorant+geminate stop. Prince and Ito both propose syllable structure  constraints that effectively preclude sequences of C+[-cont]+C, relying on (i) sonority sequencing constraints to guarantee that the first C is a 51  sonorant and (ii) the double linking of geminates to allow them where other [-cont]+C sequences are disallowed. While their analyses do account for the data in terms of descriptive adequacy, they fail to make some significant observations. A great many (if not most) of the sonorant+geminate forms in Finnish  are loanwords. Examples in addition to the forms for 'candle', 'lamp' and 'monarch' in (2a) include: polkka 'polka', korkki 'cork', =Aid 'monk' and temppeli 'temple'. None of these forms has a geminate in the source language; gemination is therefore a Finnish development perhaps reflecting a strategy particularly favoured in borrowing 2 . This observation does not, of course, explain why Finnish should develop a series of CCC clusters when they are otherwise disfavoured, but it does shed light on why so many of the CCC clusters that do occur involve geminates. As for CCC sequences where the middle member is s and only s, the  government relations which determine segmental sequencing allow s as the only non-sonorant consonant to be governed by a following obstruent (see Chapter Two, (4), above). It follows that there could be no form such as *kalpka since there could not be a form *kapka in the first place. It is less clear in the face of forms such as halm6 'fool' and kampi 'crank' where there cannot be, for example, a word such as *heilmpli as no sequencing constraints are violated per se. I will thus concede the operation of a constraint that limits the middle member of a CCC cluster to s.  52  5.3 Monomoraic Sequences Prince and Ito assume the possibility of Rimes of the form: (3) * R 1\\ N \ \ \ X y z  Or  It is here that the problem of CCC clusters is revealed, for such structures are precluded by the Government framework. If the claims of the theory are to be maintained, the obvious fact of CCC clusters in Finnish (regardless of their limited distribution) must be accommodated within the framework. Herein lies the true reason why the first member of these sequences is a sonorant. I propose that the sonorant in a Finnish CCC cluster does not depend directly from either the Rime node or the Nucleus node at all, but shares the skeletal position of the nuclear vowel creating a monamoraic sequence 3 : (4) :\ N \ X X / \ V son C Thus, it is not simply a preference for left to right decreasing of sonority in Rimes that assures that the first segment of a CCC cluster is a sonorant, but rather the fact that in Finnish (as in many other languages) only vowels and sonorants may appear in the Nucleus of a Rime. The monamoraic structure is not wholly unmotivated for Finnish. There is a set of light or 'rising' diphthongs (ie, uo,  yo) in Finnish (Karlsson  1983, 19; Whitney 1956, 14). They consist of a charmless vowel preceding  53  a charmed vowel (which acts as a head) attached to the same skeletal slot: (5)  / \ Vo Vt  The direction of government is not fixed and monomoraic vowel sequences of the reverse order do occur in Finnish. Such diphthongs arise, for example, in the suffixation of the plural morpheme -i-, which, having no skeletal position of its own, associates parasitically to a position in the root. If affixed to a root ending in a long vowel, the root vowel is shortened. If affixed to a root ending in a short vowel, the root vowel may mutate or syncopate 4 . If affixed to a root ending in a light diphthong, the first (non-head) element of the diphthong is deleted: (6) a. affixation of  [—i—]plural+[-11A]ess.  maa + i + nA ---> main luu + i + nA ---> luina b. X X X  X X  to a long vowel  'country' 'bone' XXX  XX  11111  ma  +i+na  (7) a. affixation of  maina  [ — i — ]plura1 4 [ -11A]ess.  to a short vowel  mykka + i + nA ---> mykkina 'mute' kala + i + nA ---> kaloina 'fish' lasi + i + nA ---> laseina 'glass' save + i + nA ---> savina 'clay' b. XX XX 1111 1111  X X  XXX  k a 1 a+i+na  X /  11  kalo  XX \11 11  ina  54  (8) a. affixation of [ i ]plura141-nMess. to a light diphthong -  -  tie + i + nA ---> teina suo + i + nA ---> soina yo + i + nA ---> Oina b. X X  'road' 'swamp' 'night'  X X  XXXX / \ so ina  > su o+i+na 0  There is a further piece of evidence for the monanoraic nature of the diphthongs created by plural affixation. Disyllabic roots which double their final vowel behave like the forms with underlying long vowels as in (6). The behaviour of disyllabic roots is demonstrated in (7). I shall motivate the structure of these forms in the next section, but a distinction may be drawn on the basis of their surface behaviour. If one canpares, for example, lasi 'glass' with perhe 'family' the main difference between them is the presence of the long vowel in certain forms of perhe: (9)   lasi lasing laseina  perhe perheena perheina  nan. sg . ess. sg . ess. pl .  The true distinction between long and short-vowel roots appears in forms that undergo Consonant Gradation: (10)  synti 'sin'  syntina synniksi synneiksi    ranne 'wrist'  ranteena ranteeksi ranteiksi    nan. sg . ess. sg . transl. sg . transl. pl.  In the translative singular forms in (11) the basic gradation pattern is established (weak grade with short vowels before a CC cluster; strong grade with long vowels before a CC cluster). In the translative plural forms, the diphthongs pattern with the vowel lengths--a diphthong derived from a short vowel is accompanied by the weak grade. A diphthong derived 55    	X  from a long vowel is accompanied by the strong grade. The difference in representation between these two forms (which I shall argue for in the next section) is: synneiksi  (11)  0 R I	 :\ N \  0 : : :  \ X  X  X  X I I S  R :\ N \ 1	 \  0 R  X  X  X  I I  \ /  y  ranteiksi  / \  ne  I I  I I  I N I	 : I I  I I  I I  iksi  0 : : 1I  R I\ N \ :	 \  0 II : II  R 0 II N II  R 1\ N \ :	 \  0 R :	 I , N I I  I I  X X X X X X X X X I I  I I  I I  I I  rante  I  I I  i  I I  I I  I I  ksi  5.4 Monomoraic Vowel-Sonorant Sequences Having established the presence of monamoraic vowel sequences in Finnish, I will suggest that there are also monamoraic vowel-sonorant sequences. Thus, for example, the representation of helppo 'easy' is: (12)  0 R  0 R  N I X  N \  X XXX / \ \/ helpo There is further evidence of the nuclear nature of these sonorants. A few forms (e.g. /lapse/ 'child', /kiitokse/ 'gratitude') which undergo e-Deletion (see Chapter Three, above) simplify their clusters in syllablefinal positions before consonants: (13)  lapse + tA ---> kiitokse + tA ---> lapse + n ---> kiitokse + n --->  lasta kiitosta lapsen kiitoksen  'child' part. sg. 'gratitude' part. sg • 'child' gen. sg. 'gratitude' gen. sg.  56    Forms with nasals behave the same way. Thus tuhat (/tuhante/) 'thousand' patterns with /lapse/ and not with /helppo/ 5 : (14)  'thousand' part. sg .  Although the geminate in tuhatta is a false geminate, being derived by concatenation of the /t/ of the root and the /t/ of the suffix, from the point of view of prosodic structure, there is essentially no difference between a true and a false geminate. If (as in other frameworks) one can syllabify e.g. kentta 'field' as: 0  (15)  R  t I 1 1 I I  R  0  N  \ \  :  \  i I  : 1 I I I  I\\  \  N I t  X  X XX X X  1  1\ e n  k  1  /  t  a  there should be no reason why one cannot derive *tuhantta in the same way, whether the tt collapses to form a true geminate (16a) or remains a false geminate (16b): (16) a. N 	i  	I  tuhante + to ---> tuhatta *tuhantta  0 11  R I    0 1 1  R  0  R  \\ :   1 1  1 1  1  N \ \ 1 N i 1	 \	 \ : X X X X X X X X I 1  1 t  1I 1 I  t b.    I I  I I  1  :  uhant  0 R 0 R 1\\ 11 1 1 1\\ 1 N 1 N\\  I 1 1	 \	 \ X X X X X X 1 1  I t  \ i  1 1  1  1 1  1 1  1 1  a 0 R 1	1 11 :N I  X 1 1  tuhantt  X 1 1  a  However, as this kind of structure is not available in the present framework, there is no way to derive the hypothetical forms in (16). Assuming 57    that the underlying forms of kentt& and tuhat map onto skeletal slots as: (17)  X X X X X i / \ \ / a kent  X X X X X X  II 1I 1I II I I	I tuhant e I  and  then one expects a different prosodic structure. Finnish will not tolerate the triconsonantal cluster that would result from syllabifying /tuhante+tA/ as *tuhantta so same other tactic must prevail. Interconstiuent Government relations are established between the root and the affix (the fate of the /e/ is discussed in Chapter Three, above) 6 : (18)  1  I 1  *---I 1 I  X X X X X X X X I  I  I  tuhantet  a  The mapping onto 0 and R nodes can be satisfied by the segments in government relations, so the ungoverned /n/ remains unexpressed 7 : (19)  0 R 0 R 0 R ,I I 1 : : :\ 1 N N \ N : 1 1I 1I : I 1 I \ X X X XXX X X t  u  h  anteta 1 : $ * 0 0   > tuhatta  58  5.5 CVVC Sequences Government Theory excludes the representation of WC Rimes: (20)  * R :\ N \ I\ \ X X X V C  Nevertheless, Finnish seems to attest CVVC syllables (cf. (2b) repeated here): (21)=(2b) CVVC 'syllables' a. muuttaa muuntaa kiille kiilto saasta  'to move' 'to transform' 'mica' 'lustre' 'filth'  b.  taloon 	'house' illative sg. ill. sg . kaupunkiin 'town' kuntaan	'commune' ill. sg.  c.  ranteeksi ranteiksi rannoiksi taiteet terveen perkeleett&  'wrist' 'wrist' 'shore' 'art' 'healthy' 'devil'  translative sg. transl. pl. transl. pl. nom. pl. gen. sg . partitive pl.  There is evidence, however, that most of the above V:C and WC sequences do not define constituents. Consider first the illative forms in (21b).  59    The presentation of more illative forms will reveal the nature of this suffix: (22)  nam. sg  ill. sg .  ill. pl .  maa tie luu talo kala  maahan tiehen luuhun taloon kalaan  maihin teihin luihin taloihin kaloihin  'country' 'road' 'bone' 'house' 'fish'  After monosyllables and diphthongs, the illative suffix takes the form -hVn with the suffix vowel being a copy of the last stem vowel. The spreading of the stem-final vowel is probably related to the regular process of Vowel Harmony in Finnish. In Vowel Harmony, suffix vowels gain their expression of backness fram the value for backness in the root or stem to which they are affixed. In the case of the illative, the entire identity of the vowel is so determined. The spreading of a Harmony process is different fram that which creates long vowels. Vowel Harmony spreads elements on the Nuclear projection and so is not blocked by the presence of intervening consonants: (23)  /kapu/ + /-nA/ e ssi v .  'cone'  I°  -,-  _ U	o A* A+ I	I II II N N N 1	I 1I 1I X X X X X X I	I II n vo ?o No 1I : 1 ?o uo Ro ..  .  I  H 	H- -  k  a  .  I  ?o  p y n  a  60    A linearization or 'Tier Conflation' operation then produces the representation: (24)  X X Xi X X X 1 1 I v° A'' I I I I ?o II, I 1  I 1 : I ?° U° R° Al II 1 1 I I uo 10 ?o 10 1 I  1  No  H-  kapynd The illative morpheme has the form: (25)  0 R 11	:\ N \ 1 1 1	 \ X X X 1  h  n  and when affixed to e.g. /talo/ 'house' creates the following structure: (26) OROR 1I 1	I	 1	I I :aN loN :	 \1	 1	 \:  X x x x ..  : 1 t  +  0 I	I ' 11  R 0 R 0 R 0 R I	I	 I	I 1	 1	 1\ 1\	11 N\ laNloN:N\ 1	 \i 	\ T.-.„1....._..	 \ 1	 \  x x x I  h  n  x x x x x x x  I I I  t  1  h  n  After tier conflation this structure is derived: (27)  O R O R O R I ,' ,' 1 1 I 1\ N 1 N 1 N \ I I I 1 I 1 I  \ XXXX XX X I I I 1 1 1 : I I I  talohon  61  The intervocalic h is deleted by a rule: 0  (28)  h  condition: /h/ may not be preceded by a vowel sequence or long vowel  The condition precludes the derivation of e.g. *tieen rather than tiehen 'road' ill. sg . or *maaan rather than maahan 'country' ill. sg. 8 . A second indication that V:C and WC sequences in Finnish do not form  a constituent is the failure of Consonant Gradation before these structures: (29)nam. sg . ranne taide aie laite  transl. sg .  transl. pl .  ranteeksi taiteeksi aikeiksi laitteeksi  ranteiksi taiteiksi aikeiksi laitteiksi  'wrist' 'art' 'intention' 'appliance'  In Chapter Four, I argued that the trigger for Gradation is a branching Rime node which governs the target for Gradation. Under the assumption that V:C and WC sequences constitute branching Rimes (as in (30)), the expectation would be that these branching Rimes also trigger Gradation. The data in (29) show that this is not so. (30) V:C and WC as branching Rimes R  R  \ N \ \ X X X  :\ N \ :\ X X X  :/  1  V C  I I  I I  I I  VVC  One must therefore conclude either that the specification of 'branching Rime' is not sufficient to define the trigger for Gradation (for example, Gradation may be sensitive to the number of morae in the Rime) or that the 62  	\  sequences in question are not branching Rimes. The Theory of Phonological Government explicitly disallows the structures in (30) (see Chapter One, (6)). On the basis of the structures derived in the discussion of illative case morphology, above, I conclude that V:C and VVC sequences are represented: (31)  R 0 R  R 0 R :\ N N \\  N N \\ and  I  X X X \/ V C  X X X 1 I  1 1  1 I  Vi V2 C  A form such as ranteeksi 'wrist' transl. sg . is thus represented: (32)  0 R I I  0 R OR I 1  1	IX  N\  I I  0 R I  I\  INN\   \ X X XX X X XX 1 1  1  I I  I 1  1 1\  rant  a  1 1  k s i  5.6 The Derivation of Doubling Forms The doubling of root-final vowels is not the only interesting property of these noun (and adjective) roots. The term for these forms that Keyser and Kiparsky (1984) use is 'contracted stems'. The reason for this appellation is the fact that they behave as though they had a final consonant in the nominative and either lengthen their final vowel or double a following consonant is other cases. Thus, ranne 'wrist' exhibits the weak grade in the nominative even though the final syllble seems to be  63  open and thus not a trigger for Consonant Gradation. Keyser and Kiparsky (1984, 19) assign these roots the form: (33)  C V C C V C V I	 I	 II	 I rante  and rely on their rule of e-Deletion to degeminate the /e/ and leave the empty C to trigger Consonant Gradation: (34) / \ /  a / \ / /\  CVCCVC rane I have argued above against e-Deletion and the use of a rule that affects geminates as well as single segments. Furthermore, in a theory that treats the skeleton as undifferentiated timing slots, there is no need to postulate two extra positions in these roots. Keyser and Kiparsky are correct in their interpretation of the data, that an empty position in these roots affects their surface behaviour, but I shall propose an alternative analysis that does not require two additional skeletal positions. I postulate that after association to the skeleton, these roots have the form: (35)  X XX X XX I	I	 II		 1I	 :	 1I rante  64  The empty position triggers several effects on the surface. (36) a. Consonant Gradation. Forms such as ranne 'wrist' exhibit gradation before an 'open' syllable. b.  Phrase and sentence level sandhi: terve lapsi 'healthy child' is pronounced [tervellapsi] terve poika 'healthy boy' is pronounced [terveppoika]  c.  affixation of partitive case suffix /-tA/ shows gemination: tervett4 'healthy' part. sg . rannetta 'wrist' part. sg . compare: tulta <--- /tule/ 'fire' + /-tA/ ('e-Deletion') taloa <--- /talo/ 'house' + /-ta/ ('t-Deletion')  d.  Gemination in ca pounds: ranne 'wrist' + kello 'clock' ---> [rannekkello] 'wristwatch'  e.  Vowel lengthening where the following consonant cannot geminate: i)  final consonants:  terve 'healthy' + -t (nom. pl.) ---> terveet perhe 'family' + -n (gen. sg .) ---> perheen ii) geminates and clusters: paine 'pressure' + -ssA (inessive sg.) ---> paineessa terve 'healthy' + -1tA (ablative sg.) ---> terveelta iii) the essive suffix /-nA/ (see above, section 5.2): perkele 'devil' + -nA (essive sg.) ---> perkeleen5 The gemination of consonants either in affixation or sandhi I will refer to simply as 'doubling'. I have designated this process of vowel lengthening 'Pseudo-Compensatory Lengthening' because the lengthening observed, while similar to true compensatory lengthening, does not fill a position voided by a morphological or phonological process but fills a position present, but unrealized, in the underlying structure of the morpheme. If compensatory lengthening were a general feature of Finnish  65  phonology, one might expect lengthening to occur where Consonant Gradation has resulted in degemination (in Government Theory, the skeletal slot is not deleted, but remains empty). This is not the case; for example, the genitive sg. form of kukka 'flower' is kukan not *kuukan. As I suggested above, the underlying form of a root like /terve/ 'healthy' is: (37)  XXXXXX 1 t e r v e  As a convenient shorthand, I will represent these roots in text as e.g. /tervex/. The status of the empty position at the end of the root is ambiguous. It is in a position that is potentially properly governed by the preceding vowel, in which case the vowel could spread onto it (i.e. Pseudo-Compensatory Lengthening would apply). It is also potentially properly governed by a following consonant, in which case the consonant could spread onto it (doubling). The correct government configuration (as determined by suffixation) will determine the prosodic structure realized.  66    In the unmarked case (derivation of the nominative) there is no overt suffix attached to the root. The derivation would proceed as follows: (38)  X X X X X X t e r v e TIt I I I  II I I 1• I  I  I  I__I 1._...1 I	 I I	 I II	II	II 1 N N 1 1 1 r 1	 4r X X X Xi X X I I  I I  mapping onto skeleton and establishment of ICG configurations  I I  projection of N and establishment of SCG configurations  I  t e r v e 0 R 0 R I	I	 :\ I	I :\ I	I N \ II N \ II	 I	 \ I	I 1	 \ XX XXX X I I  I I  I I  I I  mapping onto OR tier  I I  t e r v e The question that immediately comes to mind is: 'Why is the empty slot, Subconstituent Governed by a Nucleus, incorporated into the Rime rather than into the Nucleus?' It is clear from the behaviour of these words (doubling of following consonants) that this is so. If the Nucleus came to dominate the position, the result would be a long vowel (*tervee). To further complicate matters, the empty position does sometimes came to be part of a Nucleus (in the cases where Pseudo-Compensatory Lengthening applies). Keyser and Kiparsky (1984, 22) suggest that "the Finnish lexicon permits no V sequences, whether long vowels or diphthongs, in unstressed syllables". This assertion is corroborated by Whitney (1956, 14): "Diphthongs, with the exception of those in which -i- is the second element occur only in the first syllable of a word [the locus of primary stress]. Elsewhere, such combinations of vowels are not diphthongs , but 67  belong to separate syllables, and this is noticeable in the pronunciation" 9 . The import of these observations is that the structure of Finnish W sequences outside of stressed (or initial) syllables is either: (39)  X VI V2  or (40)  ROR I I  I I  N N I I  I I  X X I I  I I  V1 V2  ROR I I  I I  and N N I I  I I  X X \  /  V  I have already observed that Finnish has no branching Onsets. I shall therefore further assert that there is a restriction on the branching of Nuclei: outside of stressed syllables N may not branch in Finnish. This being the case, an empty position governed by a Nucleus must be dominated by a Rime and not by a Nucleuslo.  68    In a derivation where doubling occurs, such as painetta 'pressure' part. sg ., the empty position is Interconstituent Governed by the suffixinital consonant: (41)  X X X X X X 1t 1	 1	 1	 [	 1	 + p a' io n e 1	t t	1_1	1 t i_iI	 I  	I__I   I I I N  X X 11	 : a :	 1 t	I—II  I__I  I I I  I I 1 i  projection of N and SCG  I I I  N :  N    	II I  II  association and ICG  I I  S  X X X X X XXX  II	 I	I II paine :  1  O R 0 R I I  I I  I I  I	I	 II t a  0 R  ;\  I I 1  N 1 1 N 1	 : \ 1	1 I 1 XXXX X 1I I	 I	 I I	 I I	 1 paine  \ 1 N \	 1 :	 1 XXX \I I1 t a  mapping onto OR and gemination of t onto governed empty position  Here, the empty position is identified as non-nuclear by virtue of Interconstituent Government by a non-vocalic segment. It therefore is incorporated into R rather than N. The spreading of the /t/ does not necessarily follow automatically, but may be governed by rule--the evidence is as yet unclear on this point.  69    The final case involves the derivation of a long vowel as in perheeksi 'family' transi. sg.: (4 2)    X X X X X X I  :  1  I  I  	perhe t	1_1	I t	1_1	I  X X X 4.  1 1  1 1  k s i t 1 1 1_1 1 I t I-- 1  1_1 1 1  1 _1 I	 I 1 I	 : 1	 1I N : N N : N I1	 :	 * 1I  1 XXXXXXXXX II	 I	I II : II	 II	 II	 1 P e r h e k s i  O R 0 R 0 R 0 R 1	1	 : I	, 1, 1\ :\ 1 N 1  N \ 1	1 N \ 1 1 N I	I	 :	 \ I	I	 I I : \ I		 II XX XXX X XXX 1 1  1 1  1 I  association and ICG  1 1  projection of N and SCG  mapping onto OR tier and spreading of vowel  11 1 1 1 	 / I  perhe  k s i  Here, the /k/ of the suffix is not in an ICG relationship and must enter into a government relationship with the position preceding it. In order for this to happen, the position must project a Nucleus so as to become a governing category. The vowel must spread into the empty position so that it can govern (for empty positions cannot govern). It may be that the spreading of the vowel precedes and facilitates the projection of N, the prohibition on branching Nuclei preventing the two positions being dominated by the same N node. That the structure derived in (42) is the correct one is supported by the fact (mentioned above) that forms such as ranteeksi 'wrist' transl. sg . do not gradate in spite of the fact that superficially (and in some theories of syllable structure) they appear to have a susceptible segment in the onset of a closed syllable. Further corroboration of this 70  structure can be found in the behaviour of the plural morpheme /-i-/. When affixed to a doubling root (e.g. /rantex/ 'wrist'), the plural morpheme (which is a segment without a skeletal position) associates to the root-final position. When affixed to a non-doubling root (e.g. /ranta/ 'shore') the plural morpheme associates parasitically to the root-final position (and incidentally triggers a vowel mutation). With the further addition of a CCV case affix, the surface forms are superficially similar: ranteiksi 'wrist' transl. sg . and rannoiksi 'shore' transl. sg .. The difference between them is that /ranta/ undergoes Consonant Gradation while /rantex/ does not. In the one case, the ViC sequence constitutes a branching Rime and in the other it does not. The structure of these forms must therefore be: (43)  a. rannoiksi /ranta + i + ksi/ gradation OR  OR  1  1  1\ N\  II  1\ N\  I  \	 I  X  X XX  1  1  1/  r a	 n	 o  0R I  N  I\  X /  X X X 1  1  b. ranteiksi /rantex + i + ksi/ no gradation 1	 1 OROR i i :\ N\  OR 1 1\ IN\  11\11 II II  X  X  OR IN  1\11 I II  X  X  X  X  X  X  X  r a n  t  e  i  k s  i  1  iksi  71  Notes to Chapter Five 1. The Finnish lexicon contains a number of unassimilated loans that appear to have branching onsets: trigonametria 'trigonametry', mikrofilmi 'microfilm', platina 'platinum', Skotlanti, 'Scotland', strategia 'strategy'. Word-initial clusters most likely involve extrametricality (see Kaye, Lowenstamm and Vergnaud 1989, 44-47). Word-internal clusters that violate Government sequencing constraints may involve empty positions (see above, Chapter Three) that separate the offending segments, e.g. mikOrofilOmi. This second approach might account for the CSC clusters, but to appeal to empty positions in this case would allow for the possibility of CCC clusters in which the second sound is an obstruent. Such forms are not attested (*pyrt0k45). On the whole, it seems safer to postulate that CsC forms pattern with sonorant+geminate sequences like polttaa 'to burn'--especially as the first segment of CsC clusters are always sonorants. 2. Dr. Zita McRobbie (personal communication) informs me that the gemination of native forms such as polttaa 'to burn' and kanssa 'with' is a relatively recent development in Finnish. It is not, however, an across the board phenomenon. Finnish has many words of the form CV[+son]CV: jalka 'foot', pentu 'cub', kumpu 'hill'. One byproduct of this gemination in loanwords is that the the resemblance of the word to its form in the lending language is not lost when the word appears in a weak grade form. A comparison (using the genitive suffix -n) of the effects of Consonant Gradation on these forms and the effects Gradation would have in non-geminated versions is given below: a) actual Gradation pattern b) Gradation pattern without geminate  polkka polkan *polka polan 'polka' lamppu lampun *lampu lammun 'lamp'  monarkki monarkin *monarki monarjin 'monarch'  munkki munkin *munki mungin 'monk' -  -  -  -  -  -  -  -  This may be just a fortuitous coincidence of the borrowing of these forms at the same time Finnish was undergoing the historical change mentioned above, but in support of the idea is the fact that while 'monarch' was borrowed as monarkki, 'monarch' became monarkia. The lack of gemination may result from from a metrical effect of the extra syllable, but it is tempting to suggest that the extra syllable protects the k from Gradation (the genitive is monarkian) and so the k did not need to double. Older borrowings such as ranta 'shore' (from German Strand) undergo Gradation normally (i.e. ranta rannan). Newer borrowings fail to undergo Gradation even without gemination to preserve the original sound. Thus auto auton 'car' (although Dr. McRobbie informs me that she has heard *audon colloquially). On the question of quantity in Finnish, see Lehiste (1965). -  -  I owe this suggestion to Glyne Piggott (personal communication). 3. Much of the ensuing discussion might be more clearly presented in the  72    notation of Moraic Phonology (see Hayes, 1988). However, as long as it is understood that the representation: 0 R I I\ 1 N \\  0R I I  I I N   X X X XX \  u  la m p is equivalent to:  there should be no difficulty in following the X-slot notation. 4. The details of these mutations and syncopations are too complex and numerous to present here. See Karlsson (1983, 39-43) for discussion. 5. That the underlying form is /tuhante/ may be seen from the illative and essive forms tuhanteen and tuhantena. Forms with CC-initial suffixes exhibit the expected weak grade pattern; e.g. tuhanneksi translative sg. 6. The fact that tuhat derives fran the same Germanic form as does the English 'thousand' does not settle the question of whether the [e] that surfaces in longer forms which exhibit 'e-deletion' is underlying or epenthetic. Much depends on when this word was borrowed fram Germanic. If the borrowing was long enough ago, the borrowed form may have been *thusundl rather than *thusund. 7. In contrast, the underlying /n/ will be picked up if the OR mapping cannot be satisfied on the basis of government relations alone. The derivation of the essive form tuhantena wherein the /nj of the suffix cannot Interconstituent Govern the root-final /t/ proceeds: I.,..I I I  X I I  t  N  t I  N  I  ___ II  I I  *  I I  *  I I I  X  X  X  X  X  I I  I I  I I  I I  I I  I  I  I  I  N  *  I I  X  X  I I  I I  u h a n ten a (derivation continues)  73  OR OR 0*OR I I I III 11 I\ 1 i 1 N N N\ I I I I 1 I I I I II I I I I I I I\ X X X X XX X X IIIIIIII  1 $ 1 11 t u h a n ten a  OR OR  OR OR  1111\ 1111\ :  N  :  I I  :  I I  X I I  X I I  X I I  t  u  h  1111 1111  N\  1 X I  1 a  \ X I  1 n  ,  N  I  I  1  I I  I  I  X I  X 1  X I  X I  1 t  1 e  ,  1 n  N 1  I a  8. The effect of the condition on the rule likely falls out of a general constraint on vowel sequences. In these cases, the [h] is required to break the hiatus between vowels (see Schane 1987 for a discussion of hiatus-breaking). 9. Lehtinen (1962, xvii) treats Vu and Vy sequences outside of the initial syllable as tautosyllabic diphthongs as well. She also asserts that long vowels are always tautosyllabic. As she is dealing with pronunciation and not with phonological structure, her assertions are not too damning a criticism of the present proposal. In fact, where long vowels derived by morphological processes such as the affixation of the illative case suffix have the structure: R 0 R N N  XXX \  it is to be expected that in pronunciation the heterosyllabic geminate vowel will be surface as a single (phonetically) long vowel. 10. The stipulation that N may not branch in unstressed Rimes raises the possibility that N may branch where secondary stress is assigned. This possibility will not generate counterexamples to doubling where long vowels surface instead of an empty Rime-final consonant. Finnish has no nominal roots of sufficient length to create this situation. A three syllable form such as perkele 'devil' is not subject to secondary stress assignment. A four syllable form (if such exist) would have the Nucleus dominating the last vowel in an unstressed position. The possibility of secondary stress assignment in an affixed form exists, e.g. prkelettA, but the ICG relation between the suffixal /t/ and the root-final empty position precludes the interpretation of that position as Nuclear (ICG has precedence over SCG). Evidence for branching of N is found in the stress pattern of some 74  trisyllabic nouns. Keyser and Kiparsky (1984, 24-25) offer the example mellakka If secondary stress is not applied, the essive plural form is mellakkoina. If secondary stress is applied, then the essive plural is ngllakOina. One possible interpretation of these data is that the application of secondary stress allows the structure: 0 R 1	 I N\  0 R 0 R 0 R I1\ 1 N\ I N N \	 1	 i	 \	 i	 :\ X X XXX XXXXXX II \:	 I 1	 1	 1	I	 1	I	 I1 m e 1 a k oin a where the branching of N triggers Consonant Gradation (in this case, degemination of k). This does not seem likely to me for the behaviour of the plural marker /-i-/ (for example, the mutation of /a/ to [o] here and the cases where /-i-/ shortens or deletes root-final vowels--see above section 5.2) indicate that the plural marker does not have its own skeletal slot in UR. It is not clear how assigning secondary stress could insert a position to which the plural marker could associate. Another possibility is that the stress pattern itself influences gradation, but this cannot be the whole story, for one would expect to see non-plural forms with secondary stress in the weak grade and this is not the case (*mellakana). A third possibilty is that branching of a Rime always triggers Gradation and branching below the Rime in conjunction with stress will trigger Gradation. In this case, the branching of the Nucleus or of the skeletal position itself would count as the trigger for Gradation:  0 R  1	 1\ N \  0 R 0 R OR I	 :\ 1	 1 I 1 N \ N N \	 I	 1	 \ X X XXX XX X X X \: / \	 I	 I m e 1 a k o i n a Clearly, further investigation is required to resolve this issue.  75  CHAPTER SIX - Further Aspects of Consonant Gradation In this chapter I shall return to the subject of Consonant Gradation and examine same exceptional and unusual manifestations of this process. 6.1 Same Exceptions to Consonant Gradation In my earlier discussion of Consonant Gradation I presented the basic facts, repeated here: (1) pp tt kk p t k mp nt nk It rt  ---> ---> ---> ---> ---> ---> ---> ---> ---> ---> --->  p t k v d 0 mm nn qg 11 rr  strong grade  weak grade  kauppa katto takki kipu aiti joki kampa ranta kenkd kulta kerta  kaupassa katolla takissa kivussa Aldine joesta kammalla rannalla kengast& kullaksi kerran  'shop' 'roof' 'coat' 'pain' 'mother' 'river' 'comb' 'shore' 'shoe' 'gold' 'time'  nam. nan. nam. nam. nam. nam. nam. nam. nam. nam. nan.  sg./iness. sg sg./adess. sg. sg./iness. sg . sg./iness. sg . sg./allat. sg sg./elat. sg . sg./adess. sg . sg./adess. sg . sg./elat. sg . sg./transl. sg . sg./gen. sg .  There are certain exceptions to the above list and a number of effects which interact with the deletion of k. Gradation fails when t and k are preceded by non-sonorantsl: (2) posti postin *posdin *rasdin rasti rastin poskin *posin poski matka matkan *matan karahka karahkan *karahan but:  Lahti Turku jalka nahka  landen Turun jalan nahan  *lahten *Turkun *jalkan *nahkan  'post' nam. sg./gen. sg. ,, 'checkmark' ., 'cheek' ,, 'journey' 'stick' 'gulf' city name 'foot' 'skin'  It II II  ,,  The failure of t to voice after s is not surprising--proximity to the voiceless sibilant is sufficient reason for voicing not to occur. The failure of k to be deleted after s is another matter. Deletion of the  76  segment would leave an empty skeletal position (the glide-formation evidence below substantiates this). Empty positions must be properly governed, that is, they must be governed by a phonetically expressed segment and no other government relation may overlap with this first relation. In the structure: (3)    X X X X X X 11 1 1 1 1  1  1  1  1  1  poskin 1  1  1 I 1  1  1 _1  the k is governed by the i, but is not properly governed as the k is also in a government relation with the s. Deletion of the k would also leave the s to be governed by an empty position. The k cannot, therefore, be deleted. In the hC cases, the t is allowed to voice--proximity to the glottal fricative is not a sufficient deterrant to the spreading of voice to the t from the following vowel. Again, the case with h is not so tidy. In the majority of instances, k will not be deleted after h. Again this would be because the k governs the h and is itself not properly governed. This leads one to suppose that in the few examples like nahka/nahan 'skin' the k does not govern the h. Such non-government is possible if the h is (as were the sonorants in e.g. helppo 'easy') linked to the same skeletal  77  position as the vowel. The representations of karahka 'stick' and nahka 'skin' would be: (4)  a.OROR I	 '\ N N  0 R II II I N  I  I  I  I  I I  I  X  X  X  X  X  i I  I I  I  I  I  k  a  I r  1  I a  \  I  \  b.  0  R  I  I  I  0 R  N  I  I I  I I  I I  I I  I I  X  X  X  X  X  X  I  I II h k  I	 / \	 1I n a h k  I a  1  I a  In (4a) the k governs the h and so is not itself properly governed and cannot be deleted. In (4b) the h and k are not in a government relation (Onsets cannot govern segments within the Nucleus--KLV 1989, 53) and so the k is properly governed by the following vowel and can be deleted. This analysis is strained by examples such as jalka jalan 'foot' nam. sg ./gen. sg .. In order to maintain that the deleted k is not in a government relation with the preceding sonorant and is therefore properly governed, I must posit the structure: (5)  0 I	I : I X I j  R 0 R II :	 I\ N : N \  : 	 \ X XX X / \	 :	 : : a 1 k a n   >  0 R 0 R I : :	 I\ 1 N 1 N \ 	 I	 \ X X XX X :	 / \ I	I 11 j a 1 a n  I have not found any instances of a root with the form CV[son]kV in which the k is not deleted in the weak grade. Accordingly,  every  root of  this form must involve a monamoraic vowel-sonorant sequence. This is an unorthodox conclusion, to say the least. However, given the existence of roots like /helppo/ 'easy' on the basis of which I originally argued for monamoraic vowel-sonorant sequences, it is not unreasonable to expect such sequences to turn up elsewhere. Indeed, the simplest assumption a language learner would make is that once forms like helppo established the 78  existence of monamoraic vowel-sonorant sequences, all such sequences are monamoraic unless there is evidence to the contrary. 6.2 Same Remarks on Glide-Formation The deletion of k also leads to other interesting effects. Of particular interest is a process I have designated Glide Formation. In certain circumstances, where k is deleted fran a position preceding a nonlow non-round front vowel or a high round vowel, the vowel spreads to fill the Onset of the syllable.  The distribution of Glide Formation is demon-  strated in (6): (6) /joke/ /arke/ /polkime/ /jalka/ /turku/ /luku/ /kyky/ /huoku/ /ruoko/  strong grade  weak grade  joki arki polkimen jalka Turku luku kyky huoku ruoko  joen arjen poljin jalan Torun luvun kyvyn huoun [ruo?on]  'river' 'weekday' 'treadle' 'foot' city name 'number' 'Ability' 'breath' 'reed'  nan. sg ./gen. sg . gen. sg./nam. sg . nam. sg ./gen. sg. VT  IV VI VI  The effects of Glide Formation (e.g. arki arjen, luku - luvun make it -  clear that only the segmental structure of the /k/ is deleted and not its skeletal position, for if the skeletal position of the Onset were deleted along with the /k/, such a spreading would not likely occur. However, it is not simply a matter of spreading the appropriate vowel whenever a preceding /k/ has been deleted. The I0 element of the vowels /i/ and /e/ spreads to form [j] when the deleted /k/ is preceded by a liquid, but not when the /k/ is preceded by a vowel. /y/ and /u/ (but not /0) form [v] when the deleted /k/ is surrounded by identical vowels 2 . Round (labial)  79    and non-round Glide Formation are in complementary distribution--labialization occurs intervocalically but not between a liquid and a vowel; jodformation occurs between a liquid and a vowel, but not in intervocalic position. The derivation of Glide Formation proceeds as follows: (7) a. derivation of  arjen  'weekday' gen. sg . gradation of /k/  0 R 0 R I I	 :\ N  N\  X XX X : 1 / \	 1  a r k I0 n : * 0 :  O R 0 R : I I\ N N \ I I	 I	 \ X X X X / \ \ a r I0 n   spread of I° (colons indicate that the elements below do not spread)  arjen  A+ I+  80  (7) b.  derivation of luvun 'number' gen. sg . OROR  gradation of /k/  \ N \  N I1111\ XXXXX 1111 1 1111 1  1 u k U° n 0 O R OR 1 :\ 1 N 1 N \\  spread of Uo  1111 xxXX  1 u  \: U0 n   >  luvun  The above derivations demonstrate the operation of the glideformation processes. I shall leave the question of distribution of the processes--why they occur where they do or fail in other circumstances for further research.  81  Notes to Chapter Six 1. The lack of native Cp clusters involving non-sonorants makes it impossible to examine the gradation pattern of forms such as the hypothetical *kaspa or *kahpa. When preceded by a non-assimilated sonorant, p alternates with v: kalpa kalvan 'sword' nam. sg ./gen. sg ., arpi arven 'scar'. -  -  2. According to Collinder (1957, 6) the sound represented by <v> where it has been derived by Glide Formation is [A] rather than the [v] found elsewhere in Finnish. This sound is derived phonetically by the spread of the labial element UP into the position left empty by the deletion of k. Certainly, it is not the result of sate freakish velar/labial alternation.  82  CONCLUSION  The purpose of this thesis has been to investigate the adequacy of the theory of Phonological Government with respect to Finnish. In the main, this performance has been good. The theory predicted with reasonable accuracy the kinds of heterosyllabic clusters one finds in Finnish. The most notable exception, the lack of clusters ending in p, represents a genuinely exceptional feature of Finnish and falls outside the context of the Principles and Parameters approach of the theory. The theory also provided an account of the interaction e-Deletion and t-Deletion, being able to treat the variations in root-final as as an artifact of structure rather than as a rule of grammar. The theory provided the beginnings of a full account of Finnish Consonant Gradation. The virtue of the theory of Government in this regard is the description of the context of Consonant Gradation in an active manner (an Onset Governed by a Branching Rime) rather than in a passive manner (the onset of a closed syllable). The theory's insistence that there are no syllables with the structure CVVC also offered an elegant account of how these superficially closed syllables fail to trigger Gradation. Although there is much work yet to be done with regard to Consonant Gradation (particularly in the verbal system, which I have not examined), it is clear that Government is a valuable notion in this regard. The theory of Phonological Government is still in its infancy. As it becomes more fully developed and articulated--particularly in terms of how constituent structure is erected--and as it is brought to bear on a greater number of languages and linguistic problems we shall see how well it fares against rival theories. This single case study cannot answer any  83  of the big questions--a large cross-linguistic survey would be needed for that. Even so, in the ongoing battle of "Generative Phonology vs. Finnish Phonology", Government theory emerges as an able contender.  84  REFERENCES Anderson, John, Colin Ewen and Jorgen Staun (1985). "Phonological Structure: Segmental, Suprasegmental and Extrasegmental." in: Colin J. Ewen and John M. Anderson (eds.) Phonology Yearbook 2. Cambridge: Cambridge University Press. 201-222. Campbell, Lyle (1981). "Generative Phonology vs. Finnish Phonology: Retrospect and Prospect". in: D. L. Goyvaerts, ed. Phonology in 1980's. Ghent: E. Story-Scientia. 147-182.  the  Chamsky, Noam and Morris Halle (1968). The Sound Pattern of English. New York: Harper and Row. Collinder, Bjorn (1957). Survey of the Uralic Languages. Uppsala: Almqvist and Wiksell. Foley, James (1977). Foundatio of Theoretical Phonology. Cambridge: Cambridge University Press. Ito, Junko (1986). Syllable Theory in Prosodic Phonology. UMAss. PhD Dissertation. Hayes, Bruce (1980). A Metrical Theory of Stress Rules. MIT PhD Dissertation. (1986). "Inalterability in CV Phonology." Lanauaae 62. 321 351. (1988). "Compensatory Lengthening in Moraic Phonology". MS.   Karlsson, Fred (1983). Finnish Grammar. trs. Andrew Chesterman. Juva: Werner SOderstrOm Osakeyhtiti. Kaye, Jonathan (1987). "Government in Phonology: The Case of Moroccan Arabic". MS. 24 pp. Kaye, Jonathan, Jean Lowenstamm and Jean-Roger Vergnaud (1985). "The Internal Structure of Phonological Elements: a Theory of Charm and Government." in: Colin J. Ewen and John M. Anderson (eds.) Phonology Yearbook 2. Cambridge: Cambridge University Press. 305-328.   . (1987). "Constituent Structure and Government in Phonology". MS. 49 pp. . (1989). "Konstituentenstruktur and Rektion in der Phonologie". Linauistiche Berichte, Sonderheft 2/1989: Phonologie. 31-74. (Published version of the preceding).    Keyser, S. J. and Paul Kiparsky (1984). "Syllable Structure in Finnish Phonology." in: Mark Aronoff and Richard T. Oehrle (eds.) Lanauaae Sound Structure. Cambridge, MA: MIT Press. 7-31.  85  Lehiste, use (1965). "The Function of Quantity in Finnish and Estonian." Language 41: 447-483. Lehtinen, Meri (1962). Basic Course in Finnish. American Council of Societies. Research and Studies in Uralic and Altaic Learned Project No. 57. New York: Humanities Press Inc. Languages, Lieber, Rochelle (1987). A Integrated Theory of Autosegmental Processes. Albany: SUNY Press. McCarthy, John J. and Alan S. Prince (1986). Prosodic Morphology. MS. 108 pp. Prince, Alan S. (1984). "Phonology with Tiers." in: Mark Aronoff and Richard T. Oehrle (eds.) Language Sound Structure. Cambridge MA: MIT Press. 234-244. Schane, Sanford (1987). "The Resolution of Hiatus." in: Anna Bosch, Barbara Need and Eric Schiller (eds.) Papers from the 23rd Annual Regional Meeting ot the Chicago Linguistic Society--Part Two: Parasession on Autosegmental and Metrical Phonology. Chicago: CLS. 279-290. Selkirk, Elisabeth (1984). "On the Major Class Features and Syllable Theory." in: Mark Aronoff and Richard T. Oehrle (eds.) Language Sound Structure. Cambridge MA: MIT Press. 107-136. Shuken, Cynthia (1984). "[?], [h], and Parametric Phonetics." in: Jo-Ann W. Higgs and Robin Thelwall, eds., Topics in Linguistic Phonetics in . T. Uldall, Ulster: The New University of Ulster 111Honour ofE ms 139. Suomi, Kari (1985). "On Detecting Words and Word Boundaries in Finnish: A Survey of Potential Word Boundary Signals". Nordic Journal of Linguistics 8. 211-231. Vennemann, Theo (1988). Preference Laws for Syllable Structure and the Explanation of Sound Ltwitsm. New York: Mouton de Gruyter. ,  Whitney, Arthur H. (1956). Finnish. Teach Yourself Books, Bungay: Hodder and Stoughton.  86  


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items