Lexical tone contrast in Izon as ubiquitous floating tone

This paper establishes the lexical tone contrasts in the Nigerian language Izon, focusing on evidence for floating tone. Many tonal languages show effects of floating tone, though typically in a restricted way, such as occurring with only a minority of morphemes, or restricted to certain grammatical environments. For Izon, the claim here is that all lexical items sponsor floating tone, making it ubiquitous across the lexicon and as common as pre-associated tone. The motivation for floating tone comes from the tonal patterns of morphemes in isolation and within tone groups. Based on these patterns, all lexical morphemes are placed into one of four tone classes defined according to which floating tones they end in. Class A morphemes end in a floating (cid:2) (/ tà (cid:2) / ‘wife’), class B in (cid:2) (/ fụ́ (cid:2) / ‘salt’), class C in (/ wún / ‘sand’), and class D in (cid:2) (/ wò (cid:2) / ‘him’). This paper provides extensive empirical support for this analysis and discusses several issues which emerge under ubiquitous floating tone. Issues include the principled allowance of OCP(T) violations, and the propensity for word-initial vowels and low tone to coincide.


Introduction
One of the most influential developments in 20 th century phonology was formally separating the segmental tier from the suprasegmental tier, codified in the work of Autosegmental Phonology (Goldsmith 1976).A major part of its continuing success is its ability to represent multiple kinds of tonal configurations involving a tone (T) and a tone-bearing unit (TBU, conventionalized as τ), the structural unit to which the tone associates.Potential configurations are shown in Table 1, which include 'deficient' structures lacking either a tone (a.) or a TBU (b.), non-deficient structures where the two are associated (c.), structures where they both occur but remain unassociated (d.), or some combination (e.):

T | τ
Several of these configurations involve tones unassociated to TBUs, a representation referred to as 'floating tone' and notated with a circled Ⓣ.In most languages, floating tones are restricted in some way, such as appearing only on a minority of morphemes within the lexicon, or being restricted to some grammatical context (i.e.grammatical tone).The analytic claim of this paper is that in Izon -an Ijoid language of Nigeria's Niger Delta region -all lexical morphemes contain floating tones as part of their underlying representation.I refer to this as UBIQUITOUS FLOATING TONE, a typologically unusual system wherein floating tones constitute a core component of phonological contrast, with remarkable stability across the Izon dialect continuum (Williamson 1988).The motivation for floating tone comes from the tonal effects lexical morphemes exhibit in isolation and within multi-morpheme tone groups.The tone of all non-initial morphemes within the tone group is deleted and replaced by floating tones idiosyncratically associated by the initial morpheme.
Building on pioneering work on Ijoid in Williamson (1988), I present evidence that lexical morphemes fall into four tone classes, looking at two closely related dialects of Izon -Gbarain Izon and Kolokuma Izon.Classes are defined based their systematic effect on the following words, e.g.replacement by a LH pattern (class A), replacement by all H (class B), by all L (class C), or by a HL pattern (class D).Each of these replacement patterns is analyzed as a sequence of floating tones which appear after any pre-associated tones.This is illustrated below with four examples of nouns in object position, whose floating tones replace those of the verb /ẹ rı̣ / 'see': (1) Four tone classes in Izon -Defined based on their replacive tone pattern a. Class A: ends in ⓁⒽ / tàⓁ Ⓗ + ẹ rı̣ ́ / → [ tà ẹ rı̣ ] 'see wife!' [Gb-20190714:60] b.Class B: ends in Ⓗ / fụ Ⓗ + ẹ rı̣ ́ / → [ fụ ẹ rı̣ ] 'see salt!' [Gb-20190714:60] c.Class C: ends in Ⓛ / wún Ⓛ + ẹ rı̣ ́ / → [ wún ẹ rı̣ ̀ ] 'see sand!' [Gb-20190714:60] d.Class D: ends in ⒽⓁ / wòⒽ Ⓛ + ẹ rı̣ ́ / → [ wẹ ẹ rı̣ ̀ ] 'see him!' [Gb-20190706:30] This paper highlights a number of issues which emerge under ubiquitous floating tone representations.First, Izon allows OCP(T) violations within a morpheme if the first tone is pre-associated and the second is floating (i.e.H Ⓗ), as long as the identical tones are linked to different phonological words.Second, vowelinitial targets have special effects on floating tone association, wherein a preference emerges for low tone and onsetless syllables to coincide.I analyze the initial vowel here as extrametrical, resulting from a misalignment of morphological and prosodic constituency (following Downing 1998).Finally, towards the end of this paper, I directly compare the floating tone analysis to an alternative involving 'obligatory tone spreading', whereby the final tone of a word is required to spread across its word boundary.Common to both approaches is that the tonal effects can be interpreted as the phonologization of pitch carry-over within its tonal domain (McPherson 2016).
This paper is structured as follows.§2 provides a brief overview of the Izon language, its tone system, and defines certain terms used throughout this paper.§3 establishes the four tone classes (A-D, above) and explicates their analysis via floating tone.§4 presents discussion on three final issues: the frequency of each tone class, a comparison of the attested Izon patterns to the logically possible patterns to determine systematic gaps, and contrasting the analysis of this paper with the alternative involving obligatory tone spreading.A summary is in §5, followed by four appendices.Appendix D in particular provides information on the conventions used in data citation, as well as background on data collection, subsequent databases, archiving information of recordings, and select .wavfiles for data points used in this paper (found in the supplemental materials).

Relevant background on Izon
The Izon language Izon [orthography: Ịzọn -IPA: ɪzɔ̃ -ISO code: ijc] is an Ijoid language spoken in the extreme south of Nigeria in the Niger Delta region, often known by alternative names 'Ijo', 'Ịjọ', or 'Ijaw'.Jenewari (1989) characterizes Izon as constituting approximately 30 dialects, the most prominent in the linguistics literature being Kolokuma due to the work by Kay Williamson.Despite a high ethnic population of at least one million (Jenewari 1989: 107), Izon is an endangered language due to the growing dominance of Nigerian Pidgin English [pcm] (as well as Standard Nigerian English).In what follows, I use 'Izon' to refer to the language and 'Ijoid' to refer to the language family, and avoid the names 'Ijo'/'Ijaw'. 1he focus of this paper are two dialects of Izon, Gbarain Izon ([ɡbàrã] or [ɡbàrãɪ]) and the aforementioned Kolokuma Izon ([kólókùmá]).They are geographically proximate and each other's closest Izon relative (99% cognates in a modified Swadesh list -Lee & Williamson 1990).The data from Gbarain Izon were collected by the author in collaboration with Jumbo Gift during fieldwork in Nigeria in 2017 and 2019.The Kolokuma Izon data come from Williamson (1965Williamson ( , 1978Williamson ( , 1983Williamson ( , 1988) ) and Williamson & Timitimi's (1983) Izon dictionary (updated in Williamson & Blench 2011).For all data points in this paper, I cite the source in subscripted square brackets, which minimally includes the dialect (abbreviated as [Gb] and [Ko]), the date collected, and the page number within the archived field notes (again, see Appendix D).

Linguistic profile
All Izon varieties exhibit advanced tongue root (ATR) harmony, with [+ATR] /i e o u ı̃ ẽ õ ũ/, [-ATR] /ɪ ɛ ɔ ʊ ɪ̃ ɛ̃ ɔ̃ ʊ̃/, and neutral /a ã/.As common in Nigerian orthographies, [-ATR] is written as a dot under the vowel and contrastive nasal vowels are written with an <n> following the vowel, e.g. as in the name Ịzọn.Within a word, all non-low vowels must agree in ATR value, but the neutral vowels /a ã/ may co-occur with both sets.Although ATR harmony is categorical within morphemes, few alternations are seen across morphemes.For example, the [+ATR] root [seì] 'be bad/spoiled' appears in a derived causative form [seìmɔ̌] 'spoil' and shows mixed ATR values.In general, Izon has a very low degree of bound segmental morphology, and bound morphemes are generally treated as simple cliticization.There is no clear vowel length distinction.For this paper, the few morphemes which appear to have long vowels are interpreted as adjacent short vowels, e.g.[ɡbẹ ẹ kı̣ ] 'short'.

Tone preliminaries
Tone in Izon is a contrastive property of morphemes, therefore meeting the definitional criteria of a tone language (Welmers 1959, 1973, Hyman 2018).Izon has two tonemes, high tone /H/ and low tone /L/.In this paper, all tone is marked overtly with standard conventions, e.g.acute [á] for high tone and grave [à] for low tone.Instances of falling [â] and rising [ǎ] are interpreted as a sequence of high and low tones on a single mora, which I take to be the tone-bearing unit (TBU).
The position of underlying tones is not predictable, e.This paper does not examine the regular tone rules in Izon which are not morphologically-/lexicallyconditioned.There are at least two such operations in Gbarain Izon, H-absorption and low-to-mid raising.These are described in Appendix A.

Tone classes
It is impossible to discuss underlying tonal contrast in Izon without discussing 'tone classes' and 'tone groups', terminology stemming from Williamson (1978).In Izon, morphemes belong to idiosyncratic TONE CLASSES, defined informally as the following: (2) Tone class: a lexical item's unique combination of (i) pre-associated tones, and (ii) the systematic tonal effect it has on the following words to their right The number of tone classes in Izon depends on dialect and level/type of analysis.For the Gbarain and Kolokuma dialects, I establish four tone classes labeled A, B, C, and D, and a number of subclasses.As stated, class A lexical items assign a LH pattern to the following sequence of words to its right, class B assigns an all H pattern, class C an all L pattern, and class D a HL pattern; see (1) above.In this paper, I attribute these tonal effects as stemming from sequences of underlying floating tones.These floating tones cannot be predicted based on the pre-associated tone of either the triggering morpheme or the target.This is seen in the following minimal pair from Kolokuma Izon which have identical segments and preassociated tonal structure: (3) Kolokuma Izon -Minimal pair based on tone class In the underlying representation, floating tones appear only at the right edge of a lexical item and replace all tones to its right.In (3) above, the underlying tone of the second word -/òvùrùⓁ Ⓗ / 'space between' and /àpụ ́ràⓁ/ 'skin' -has no effect on the surface.
2 Due to Ijoid's distinct structural profile compared to the neighboring Niger-Congo families, the lack of convincing systematic sound correspondence to other Niger-Congo families, or clear form-meaning vestiges of the Proto-Niger-Congo noun class system or verbal extension system, sources such as Glottolog (Hammarström et al. 2021) treat Ijoid as an isolate family altogether.

Tone groups
The domain within which these floating tones show an effect is called the TONE GROUP, defined in (4): (4) Tone group: The multi-morphemic unit which includes the morpheme sponsoring the floating tone and those morphemes to its right to which the floating tones are systematically assigned Crudely, within a tone group the tone of all morphemes but the first are deleted, and are replaced by the floating tones of this first morpheme.Within the Africanist literature, this type of tonal phenomenon is often called 'replacive tone' (Welmers 1973: 132-133), and can be classified as a type of dominance effect (Kiparsky & Halle 1977, Inkelas 1998, Rolle 2018).The result is a complete neutralization of the underlying tones of the target morphemes, shown in a [MODIFIER NOUN] micro-paradigm in Table 2. Here, the lexical tone contrast of the nouns is neutralized and obligatorily replaced with the floating tones of the modifiers, whether ⓁⒽ from class A, Ⓗ from class B, or Ⓛ from class C. The data below is from Gbarain Izon.A simple spreading rule is insufficient to capture the tonal changes: some class B morphemes (subclass B2) end in L but sponsor a floating Ⓗ tone, while some class C morphemes end H but sponsor Ⓛ.Nor is a simple dissimilation rule sufficient, since class A both ends in and sponsors low, while subclass B3 ends in and sponsors a high.The tonal changes here are orthogonal to any phonological markedness conditions, e.g. the grammar preferring H over L, non-contours to contours, phrase-final low tones, etc.I will return to the role of tone spreading in understanding these patterns, and entertain an alternative analysis which attributes most floating tone effects to obligatory spreading ( §4.3).What defines a tone group?In Izon, tone groups largely correspond to major phrases: the noun phrase (NP) consisting of pre-modifiers and ending at the noun, and the verb phrase (VP) consisting of the complement object and ending at the verb.These may combine too, for example a [MODIFIER] [NOUN] [VERB] construction forms a single tone group.This paper will not discuss the details of tone group formation, only the tonal effects which take place after its formation, namely tonal replacement of all but the first.See Rolle (2018: §6.3ff.) for one proposal on tone group formation, discussing several competing analyses.An important issue which I return to is the behavior of the proposed floating tones when the following word's tones are not replaced (see §3.2.3).

The four tone classes
In each subsection below, I detail the patterns of the four tone classes (and their subclasses).This includes tonal forms in isolation and their forms in various tone groups.By virtue of all lexical items being classified into one of these four classes, all lexical items carry a floating tone, which was referred to as ubiquitous floating tone above.The analysis of this paper superficially resembles Williamson's (1988) analysis of Kolokuma Izon tone, which also captured tonal behavior by positing a combination of pre-associated and floating tones.A rudimentary comparison to Williamson's analysis is provided in Appendix C.

Tone class A -Analysis as floating ⓁⒽ
We begin our discussion with class A which ends in a floating tone sequence ⓁⒽ.I first lay out the basic patterns of the Gbarain Izon dialect, then compare it to Kolokuma Izon.

Core patterns of class A in Gbarain Izon
Class A morphemes bear all low tones and sponsor a floating ⓁⒽ sequence.Example (5) shows that monosyllabic class A morphemes in Gbarain Izon are pronounced in isolation with a final rising tone, both for morphemes with a single vowel (a.) and those with two vowels forming a (surface) diphthong (b.).This is straightforwardly interpreted as the floating Ⓗ co-occurring with the lexical low. (5) Gbarain Izon -Sample of class A (ⓁⒽ) monosyllabic forms in isolation a.
With multi-syllabic class A items, multiple surface patterns are attested.The majority of tokens collected show an all low-pattern with a rise on the final TBU, mirroring the patterns in (5).An example set is in Table 3, including forms which end in a single vowel (a.-b.) and those with two (c.).[ sùbeì́ ] 'gun' In these patterns, the floating Ⓗ docks to the final TBU of the word which sponsors it, which I will refer to as 'self-association'.In general, self-association in Izon primarily occurs when there is no host to the right of the trigger, though certain exceptions exist which I bring up at the appropriate point.In these patterns there is no evidence that the floating Ⓛ has any effect, and I therefore assume it deletes in isolation (or perhaps coalesces with the pre-associated low).During data collection, a small amount of variation with respect to tone association was documented.In the text body here, I strictly describe the majority patterns. 3 In Gbarain, when class A lexical items are leftmost within the tone group, the floating ⓁⒽ sequence associates to the phonological words to the right of the sponsoring morpheme.This is exemplified below, with a single phonological word as the target.Example a. shows the class A noun /bùrùⓁ Ⓗ / 'yam' assigning its floating tones to the following lexically toneless words /kpọ / 'also' and /kụ mọ / 'only'.With single-TBU targets like /kpọ /, it was common to find both [kpọ ́] and [kpọ ] surface variants.Further, example b. shows a pre-nominal modifier /ebı̀Ⓛ Ⓗ / 'good' whose floating tones overwrite the lexical tones of the following noun.In this context, the majority of tokens show an all low pattern but high on the final TBU.
3 For example, a minority of class A tokens were pronounced in isolation with a level high on the final TBU, e.g./tụ ̀kpàⓁ Ⓗ / 'lamp' as [tụ ̀kpá], and /ı̣ kpọ ̀sọ ̀ⓁⒽ / 'dirt' as [ı̣ kpọ ̀sọ ́].Other items showed a different minority pattern, that of all low with no high at all, e.g.tokens of /tụ ̀bọ ̀ⓁⒽ / 'child' as [tụ ̀bọ ̀], and /bılèⓁ Ⓗ / 'dive in' as [bıle].The lexical items exhibiting these minority patterns patterned identically to other class A items in all other regards, and in general class A is subject to more speaker and dialect variation than other tone classes.I therefore take these surface forms to be incidental surface variants of a single tone class.It is here that we see the evidence motivating the floating Ⓛ: the TBUs immediately following the first word are low-toned.
In tone groups with more than two morphemes, the ⓁⒽ sequence docks between the second and third phonological words in the tone group.In (7), the floating tones of the first word replace those tones of the second and third words.The second word appears with all low tones while the third appears with all high tones.This happens regardless of the number of TBUs in the second word, i.e. with 2 TBUs in (a.), 3 in (b.), and 4 in (c.).The tone group in c. is underlined.The words before and after it are not part of the tone group.
( The tonal association statements thus far hold for consonant-initial target words.Different patterns emerge if the target is vowel-initial, i.e. it begins with an onsetless TBU (effects of this type were previously identified for Izon in Williamson & Timitimi 1983:xxx).I refer to such vowel-initial TBUs as 'V-TBUs' 4 As discussed in footnote 3, a small amount of variation exists in these larger contexts.A minority of tokens were collected where a 2-TBU target surfaces with a final rising tone: (i) Gbarain Izon variation: Minority pattern (cf.Majority pattern) / bùrùⓁ Ⓗ kụ mọ / → [ bùrù kùmọ ] 'only yam' ([ … kụ ̀mọ ́ ]) This variation is found with larger targets as well, e.g.targets with three or more TBUs, below.Across this variation, the floating Ⓛ consistently associates to the initial TBU of the target and the floating Ⓗ to the final.
(ii) Gbarain Izon variation in larger tone groups (and its consonant-initial counterpart as 'C-TBUs').In (9), the floating Ⓛ docks to the first V-TBU and the floating Ⓗ must dock immediately after it, onto the second TBU of the word (and thereafter).These association patterns hold regardless of the number of TBUs in the target (a.), or the number of words (b.).
( These data contrast with the patterns with C-TBUs, ( 6)-( 8).There, in the majority pattern the Ⓗ associates to the rightmost TBU in the next word, e.g.[ebı̀ dànɡılòkó], or the leftmost TBU of the third word, e.g.
[òsı̀ nàmà wárı̣ ].Importantly, the variation exhibited by C-TBUs is not replicated with V-TBUs (see footnotes 3 and 4); variants like *[òpù òpòrıò̀pó] 'big pig' are not attested.These data suggest that low tones are attracted to V-TBUs, and that the floating tones in the ⓁⒽ sequence dock to adjacent TBUs.Attraction of the Ⓛ sequence to a V-TBU is illustrated with a 4-word tone group in (10) below.In a., the ⓁⒽ associates between the second and third words.In contrast, in b. the ⓁⒽ sequence docks entirely within the second word, pulled to the left from its 'normal' position of association.The pattern in c. shows that this floating sequence can also be pulled to the right if the third word begins with a V-TBU.Finally, d. shows that when both the second and third begin with V-TBUs, the sequence is oriented to the left.
(10) Gbarain Izon -Class A association with V-TBU targets in larger tone groups How can we account for the disparate behaviors of C-vs.V-TBU targets?I adopt an analysis where the initial vowel of the target domain is 'extrametrical' and not parsed as part of the following phonological word (p-words, ɷ).Compare the C-and V-TBU targets in Table 4 below, using data introduced above.In rows a. and b., the Ⓛ associates to the initial TBU in both.In row b., however, the initial vowel to which the Ⓛ associates is marked as extrametrical, marked in < > brackets.This therefore entails that the following C-TBU is what actually begins the phonological word.We may assume that the floating Ⓗ preferably associates to a word edge: if the left edge is occupied by the Ⓛ tone then it associates to the right edge (a.), but otherwise the Ⓗ associates to the leftmost free left-edge (b.).In the autosegmental representations, the underlying tonal structure of the target is in grey, constituting what is systematically replaced.The TBU to which the floating tones associate is represented as a solid line, and any TBUs to which this tone subsequently spreads are represented by dashed lines.
This analysis holds for multi-word targets as well, e.g. in rows c.-d. the Ⓛ docks to the following TBU, which is a C-TBU in c. but an extrametrical V-TBU in d.Consequently, the Ⓗ docks to the leftmost TBU of the first free phonological word, which is the third word (dı̣ bá) in c. but the second word (bı) in d. (from <e>(bı) ebi 'good').Finally, in e. the floating Ⓛ is attached to the extrametrical <ò>, and the Ⓗ docks immediately after this.

Trigger
Target Tone group a.
L ⓁⒽ /\ ebı̀ Cross-linguistically, vowel-initial syllables/words often show anomalous prosodic effects compared to consonant-initial patterns.Odden (1995) and Downing (1998) present evidence of such effects looking at tone and prosodic domains in several Bantu languages.In Odden's (1995) study of Kikerewe [ked], vowels without onsets cannot bear tone.In a conditional construction, subject markers are assigned a high tone if they are consonant-initial (a.), which subsequently doubles to the following syllable.However, if the subject marker is vowel-initial, this high tone shifts to the following TBU (which again shows tone doubling).
(11) Kikerewe -Vowels without onsets cannot bear tone (Odden 1995: 97) a. bá-ká-luunduma 'if they growl' b. a-ká-lúúnduma 'if he growls' The core of these analyses is that vowels without onsets are non-optimal for starting prosodic constituents, and also non-optimal as tone-bearing units.I follow Downing (citing Inkelas 1989, 1993, a.o.) in interpreting extrametricality in Izon as misalignment between morphological and prosodic constituents.We will see with the other Izon tone classes how V-TBUs behave exceptionally, as well.

Core patterns of class A in Kolokuma Izon
Largely identical patterns are found for class A lexical items in the other dialect of focus, Kolokuma Izon.With forms in isolation, the high of class A's floating ⓁⒽ sequence generally docks to final TBU of the trigger.However, the Kolokuma dictionary (Williamson & Timitimi 1983, hereafter [W&T83]) describes a complex set of conditions dictating whether the final TBU surfaces as rising or level high.This is reflected in their dictionary's transcription convention, with rising tones represented via a doubled vowel (note that the vowel is not inherently long, and its phonetic duration is unclear).Representative examples are in Table 5.This dictionary was updated as Williamson & Blench (2011), hereafter [W&B11].'be noisy' If a morpheme is monosyllabic (row a.), of the shape VCV (b.), or ends in a diphthong (c.), the form in isolation is realized with a final rising tone.In other contexts such as CVCV or VCVCV words (d.-e.), the final TBU is a level high tone.In a minority of the dictionary entries there is "doubl[ing of] the final vowel when standing alone or emphasized" (W&T83:xliii), e.g.variation in [bùrú]~[bùrùú] 'yam' (row e.).The authors note that "this varies somewhat with different speakers" (p.xliv), and that in general vowel doubling can be found for emphasis and stylistic effect (p. xlvi).As stated, whether there is an underlying short vs. long vowel distinction is currently not resolved in Izon.
For Kolokuma Izon only, there exists a small exceptional subclass which I classify as subclass A2 (all other A morphemes are consequently classified as A1 in Kolokuma).With these A2 morphemes, K&W83 state that their pronunciation is all low even in isolation, with no final high.A complete list of A2 lexical items is in Table 6.These forms may end in a single vowel (a.), a diphthong (b.), or a double vowel (c.).I represent A2 as /L ⓁⒽ* /, where the asterisk on the Ⓗ indicates that this floating tone does not show selfassociation.It remains to be tested whether these two classes in Kolokuma are consistently differentiated or are merely incidental variants.5(12).With a 1-TBU target, the floating sequence docks to this final TBU (a.), while with two TBUs each tone docks to a TBU (b.).In larger tone groups the ⓁⒽ sequence straddles the second and third word with an initial C-TBU (c.), but associates entirely within the second word with an initial V-TBU (d.).These association patterns are identical to the findings of Gbarain.
( However, one can observe an important difference in Kolokuma.In Gbarain, the floating Ⓗ spreads rightward to all following TBUs within the group, e.g.[ıǹè tàrà dı̣ bá búrú] in (10) above.In Kolokuma, in contrast, TBUs after the floating tones are realized with default low tone, i.e. [ıǹè òpú wàrı̣ ] 'my big house' in d. above.It can be concluded that Gbarain prefers to value toneless TBUs via spreading of the final tone, while Kolokuma prefers to value them by a default tone.

Tone class B -Analysis as floating Ⓗ
Next we examine class B which sponsors a floating Ⓗ tone.I treat Gbarain and Kolokuma Izon together as they show identical patterns.For each data set, I still denote it as Gbarain or Kolokuma.

Core patterns of class B in both dialects
Tone class B is split into three subtypes.The first (B1) is as an all H-toned morpheme which sponsors a floating Ⓗ tone, the second (B2) is represented as an all L-toned morpheme which sponsors a floating Ⓗ, and the third (B3) is a LH morpheme which sponsors a floating Ⓗ tone.These three subclasses are illustrated below with Gbarain Izon data, in isolation and in a simple tone group.In these tone groups, the tone of the target word (/bùrùⓁ Ⓗ / 'yam') is replaced with high tone.
( In isolation, there is no evidence of this floating Ⓗ when the sponsor ends in high tone already (i.e.[tárá] 'three' in b. above).In the B2 pattern, however, the floating Ⓗ self-associates to its sponsor, resulting in a rising tone [ẹ ndı̣ ].The presence of this rising tone distinguishes its use as the subject of a clause from its modificational use, shown in (15).In a., the modifier appears in subject position in a separate tone group and consequently surfaces as [ẹ ndı̣ ] with a final rise.The || indicates a tone group boundary, here between the subject and predicate.When acting as a modifier, the form surfaces as [ẹ ndı̣ ] (b.).
(15) Gbarain Izon -Class B2 minimal pair  12).An anonymous reviewer asks why high tone should spread in class B but not class A for Kolokuma.One way to understand this fact is that tone groups in Kolokuma prefer to maintain one single tone level, if possible.If the sole tone is H (as in class B contexts), then H spreads to maintain one tone level.If, however, the tones of the tone group contain both L and H (as in class A contexts), then spreading the final H would do nothing to maintain a one tone level.Instead, the (less marked) default low is inserted.An Optimality Theoretic constraint such as *CONTOUR(TONEGROUP) could model these patterns, defined as 'a tone group does not have a contour (i.e. a transition from L to H, or H to L)'.Crucially, in B3 contexts the initial L is considered outside of the tone group due to initial extrametricality, discussed immediately below.

Near-complementarity of subclasses B1 and B3
While I have presented B1 and B3 as two subclasses, in both dialects they are in near-complementary distribution.In the majority of cases, if a class B morpheme begins with a consonant then it bears all high tones (the B1 pattern), but if it is vowel-initial then this vowel bears a low tone followed by all highs (the B3 pattern).This is shown in the Gbarain data in (18).( 18 No simple two-TBU morpheme exists in either dialect of the form /LH Ⓗ / with an initial C-TBU, i.e. nonattested *[bàráⒽ].This near-complementarity is replicated in Kolokuma Izon.A greater number of exceptional vowelinitial B1 morphemes are found due to the available resources on Kolokuma being much greater at this point compared to Gbarain.The complete list is provided in Table 7.The near-complementarity of these subclasses is seen in pairs such as B1 /eŕı́Ⓗ/ 'dry' vs. B3 /ereíń Ⓗ / 'day' in Kolokuma.This near-complementarity in both dialects can be interpreted as another instance of an initial vowel showing extrametrical properties.We may represent the unexceptional B3 patterns as <e>reíń Ⓗ 'day' with initial extrametricality (again, in angle brackets), and the exceptional B1 form as eŕı́Ⓗ 'dry' without extrametricality.This allows us to state that class B1 and B3 morphemes both have entirely high tones across their tone group, with a potential low-toned vowel preceding it outside of the tone group domain.

An argument for /H Ⓗ / structure
Classes B1/B3 are analyzed as ending in the sequence /H Ⓗ /, where the final associated tone is followed by a floating tone of the same value.One might ask why posit this representation rather than simply allowing for the pre-associated high to spread.Evidence for /H Ⓗ / comes from the behavior of this class in complex tone groups, which we will briefly go through here using data from Gbarain.
To understand the argument, we must first establish the role of syntactic structure in dictating tone patterns.Within the noun phrase, we have seen several cases where pre-nominal modifiers replace the tones of the noun within the tone group, e.g./ıǹéⒽ/ 'your' and /ẹ ǹdı̣ ̀Ⓗ/ 'that' in (20).
(20) Gbarain Izon -Pre-nominal modifiers replacing tone of noun Other modifiers exist which follow the noun, and consist primarily of determiners (e.g.markers of definiteness/indefiniteness) and quantifiers.We follow Carstens' (2002) analysis of Izon syntax, whereby pre-nominal modifiers are phrases before the head noun, while post-nominal modifiers are syntactic heads after it (also Carstens 1991, Giusti 1995, Harry 2004:18, a.o.).This is in a. in (21) below.The main point here is that post-nominal determiners and quantifiers are structurally higher than pre-nominal modifiers, schematically represented in b.Toneless modifiers are valued by the floating Ⓗ tone sponsored by the noun; they have no tone to delete or to retain.However, unlike with the pre-nominal modifiers (20), b. above shows that the floating tone sponsored by the first word does not replace the tones of the second word.Instead, the floating Ⓗ co-occurs with the inherent L tone, resulting in falling tone [bı̣ ].These patterns are representative of the role of syntax in dictating tonal patterns: if the second word is syntactically higher than the first word -e.g.determiners and quantifiers (D⁰ and Q⁰ heads) -then the floating tone co-occurs with the inherent tone rather than replaces it.What happens when both pre-and post-nominal modifiers are present?In (23), a. shows that the floating tone of the pre-nominal modifier /ẹ ǹdı̣ ̀Ⓗ/ 'that' replaces the tone of the noun but does not affect the underlying low tone of the post-nominal modifier.In contrast, example b. shows that this if the post-nominal modifier is toneless then the floating tone spreads to it as well.These facts are consistent with the interpretation that post-nominal modifiers are structurally higher than the pre-nominal ones, and consequently why the tones of the latter are retained.
( Having established these facts, let us return to our argument for class B as /H Ⓗ /, rather than /H/.The important contrast to observe above is between forms like [námá bı̣ ] 'the animal' from ( 22) versus [ẹ ǹdı̣ ̀ námá bı̣ ] 'that animal' from (23).In the former, the floating Ⓗ docks to the post-nominal modifier, while in the latter it associates to the noun only and does not spread onto the post-nominal modifier.The same pattern holds if the pre-nominal modifier bears high tone itself, e.g./wẹ nı̣ ́Ⓗ/ 'walk' in (24).
Under the floating tone analysis, the floating Ⓗ in row a. associates to the post-nominal determiner /bı̣ / only when there are no free TBUs for it to dock to.This results in the falling tone.In contrast, in rows b.c. the Ⓗ is able to associate to toneless TBUs of the noun.Because it has a host already, it does not need to associate to the low-toned determiner, which it only does as a last resort.
Consider the alternative in the rightmost column, which I reject.Under this alternative, the representation of class B is /H/ without the floating Ⓗ.This H spreads obligatorily to the post-nominal determiner in a., while in b. and c. it spreads only to the head noun; it does not spread to the determiner in these latter contexts.To capture this, we would be forced to say that spreading must happen at least once.This would be a trigger-driven requirement, rather than a target-driven one.This is undesirable given the normal understanding of tone spreading, being licensed to avoid some marked structure (e.g.*TONELESS).If spreading of a phonological feature [F] is due to some markedness constraint prohibiting specific configurations, it is unclear what kind of constraint would require spreading onto /bị / in a. but prevent it in b. and c.This cannot be attributed to any boundedness of the spreading operation, whose unboundedness we have encountered in many examples thus far, e.g.[kúlúkúlú óbórı́ pı̣ ná tı̣ bı̣ ] 'a black goat's white head' in ( 14).
We return to discussing this alternative with obligatory spreading later in §4.3.I note that under Richness of the Base, both /H/ and /H Ⓗ / must be possible inputs and it may be the case that different speakers have different grammars but converge on the same surface patterns.Importantly, I emphasize that there is no contrast between /H/ vs. /H Ⓗ / in either Izon dialect, regardless of analysis.
Tone class C -Analysis as floating Ⓛ

Core patterns of class C in Gbarain Izon
The next class to discuss is Class C, which sponsors a floating Ⓛ. Beginning with Gbarain Izon, we can split class C into subclasses C1, C2, and C3, illustrated in (25).Subclass C1 ends in pre-associated L with a H somewhere before it, and sponsors a floating Ⓛ tone.Subclass C2 is all high-toned with a final floating Ⓛ, and C3 ends in pre-associated H (with a L before it) and sponsors a Ⓛ tone.
( For subclasses C2 and C3 (ending in H Ⓛ ) in isolation, in Gbarain the most common pattern is for the floating Ⓛ to delete, rather than self-associate to its sponsor.Note, however, that some class C morphemes in isolation waver between a final low tone (or falling tone if monomoraic) and a final high, in ( 27).
Most tokens of C2/C3 morphemes in isolation were pronounced with a final high tone (reflected in the Gbarain lexicon, located in the supplemental materials -see Appendix A for details).For those morphemes which showed variation, I attribute this to variation in the underlying form, e.g.a. above as C2 /wún Ⓛ / ~ C1 /wûn Ⓛ / 'sand'.Further research may reveal whether this is incidental variation involving optional selfassociation.6

Gbarain variation in underlying representation
In Gbarain, there appears to be variation in the underlying representation of class C1 morphemes /…HL Ⓛ / words which end in low tone, between /L/ and /L Ⓛ /.Evidence comes from effects with the post-verbal particle /kụ mọ / PROHIBITIVE (PROH) 'don't, shouldn't', a particle with particular tonal properties.To explain, consider the data in (28).When a verb and this particle form a tone group, the floating tones of the verb associate to /kụ mọ /.
(  [Gb-20170716:48] From these data no pre-associated tones with /kụ mọ / are detected, and I therefore take it to be underlyingly toneless.However, in larger contexts we see that it displays tone polarity properties when the floating tone within a tone group does not directly associate to /kụ mọ /.In a., we see that the floating tones on the object /bùrùⓁ Ⓗ / 'yam' dock to the toneless post-nominal modifier ọ mọ and the verb (whose tones are replaced).Importantly, the Ⓗ does not spread onto /kụ mọ /.Instead it appears with the opposite tone of whatever surface tone is on the word before it.Example b. is similar with the floating Ⓗ spreading to the verb and the particle bearing a polar low tone.In contrast, in c. we see the floating Ⓛ spread to the verb, but the particle bears the polar value high.
We can exploit this polarity property to diagnose underlying tonal structure of class C1 as /L/ or /L Ⓛ /.Under an analysis where class C verbs end in /L/, the verb would assign no floating tone to /kụ mọ /, and we would expect the form to surface as [kụ ́mọ ́] with a polar high when it is adjacent to the verb.In contrast, under the /L Ⓛ / analysis, the verb would assign the floating low and we would expect the form to surface as [kụ mọ ̀] without polar tone.
( In the examples collected, the consultant provided surface forms with the verb ending low followed by [kụ mọ ] with polar high, supporting an interpretation as /L/.However, the consultant also fully accepted forms where [kụ ̀mọ ̀] was low which supports /L Ⓛ /.This variation cannot be attributed to class C items in general.For example, (29) above, [àká ọ mọ ɡbòrò kụ mọ ́] 'don't plant corn', a hypothetical variant *[…ɡbòrò kụ ̀mọ ] was explicitly rejected when presented to the consultant.I interpret these findings as showing variation in the underlying form in the lexicon.This is plausibly due to analytic indeterminacy, noting that contexts which would differentiate /L/ vs. /L Ⓛ / are rare.Like we saw with the debate between /H/ vs. /H Ⓗ / in class B, I emphasize that regardless of analysis there is no contrast between /L/ versus /L Ⓛ /.Even under the analysis of C1 as /L/, floating tone would still be required for C2 and C3, and would still be pervasive across the lexicon generally.

Core patterns of class C in Kolokuma Izon
For Kolokuma Izon, there is clear evidence for three subclasses of class C: C1 (…HL Ⓛ ), C3 (…LH Ⓛ ), and C4 (L Ⓛ ).As stated earlier, class C4 is found in Kolokuma but not in Gbarain.Even in Kolokuma, it is rare and restricted to names for places or people.
( How are C3 (/…LH Ⓛ /) items pronounced in isolation in Kolokuma?Within the Kolokuma dictionary, most entries of class C3 morphemes are transcribed without a final low or falling tone (representing their isolation pronunciations), indicating that the floating Ⓛ generally does not self-associate to its sponsor (as in Gbarain).Examples are provided in (33).Williamson (1965: 101) in fact notes this overtly: "a unit of class [C] in initial position keeps its isolation tone pattern", whether ending in a single TBU (a.) or two (b.). (

Class C tone alignment with V-TBU targets (both dialects)
We have seen for both classes A and B that there are specific tonal effects of vowel-initial words.Such effects are seen with class C as well.For example, there is a clear asymmetry in the distribution initial high and low tone across class C. For morphemes which begin with a vowel, this initial vowel is overwhelmingly low-toned with a ratio of 15:1 in the Kolokuma lexicon, shown in Table 9 with representative examples.This can be compared to consonant-initial class C morphemes (rightmost column) which are more evenly distributed.An unexpected pattern emerges in Gbarain, however, with class C morphemes in tone groups.This involves classes C2 /H Ⓛ / and C3 /…LH Ⓛ /, which show variation in the association of the floating Ⓛ with respect to a vowel-initial target.In one variant, the floating Ⓛ associates to the initial TBU of the target, just as with the consonant-initial targets in (32) above.This variant is the one which is expected as it shows the alignment of low tone with a V-TBU.This is exemplified in ( 35 We may tentatively state that within a tone group in Gbarain, there is a constraint which forbids a falling tone across adjacent vowels in separate words (i.e.*…V" #V# …), but that it applies only optionally.This constraint is only found in Gbarain Izon.In Kolokuma Izon, the data uniformly shows the expected pattern where the floating Ⓛ associates to the initial TBU of the next word, whether it begins with a consonant (32), or with a vowel (Table 11).

Core patterns of class D in Gbarain Izon
In Gbarain Izon, no lexical items belong to class D. The only class D members are a small series of 'prevocalic pronouns'.Across Izon varieties, pronouns have two allomorphs, one used before consonants and one before vowels.This allomorphy is exemplified in (37) with an imperative construction.
(37) Gbarain Izon -Pronoun allomorphy based on following segment (consonant or vowel) a. [ Although the segmental shape and underlying tone of the pre-vocalic series is related to their preconsonantal counterpart, these alternations cannot be reduced to any regular phonological process.All pre-vocalic object pronouns pattern as class D by virtue of sponsoring ⒽⓁ floating tones which replace the tones of the following word.In (38), I divide them into three subclasses: D1 /H ⒽⓁ / (a.), D2 /L ⒽⓁ / (b.), and D3 /LH ⒽⓁ / (c.).These are each illustrated with the vowel-initial verb /ẹ rı̣ ́Ⓗ/ 'to see'.Note that in these examples, there is an independent process which assimilates the vowel of the pre-vocalic pronoun.
(39) Gbarain Izon -Pre-consonantal pronoun series (classes A, B, and C only) Let us examine the pronunciations of Kolokuma class D morphemes in isolation.For D1 (/H ⒽⓁ /) and D3 (/LH ⒽⓁ /), the floating Ⓛ does not self-associate to its sponsor in isolation, reflected in the transcription in the Kolokuma dictionary.Examples of class D are in (42).These patterns suggest that the Ⓛ deletes, exactly as it did in the majority pattern for class C, e.g.C3 /LH Ⓛ / [emeíń] 'manatee' from (33).( 42 With Kolokuma D2 /L ⒽⓁ / items in isolation, the floating Ⓗ of the ⒽⓁ sequence always docks to the final TBU of the sponsor, resulting in a surface high or rising tone.D2 surface forms are never realized with a final falling tone, indicating that the floating Ⓛ goes unrealized (as with D1 and D3).This renders D2 morphemes surface identical in isolation with class A morphemes (/L ⓁⒽ /), only differentiated when in tone groups.Representative examples are in Table 12, compared to analogous class A form.With VCV words (a.) and words ending in with a bimoraic syllable (b.), the surface form ends in a rising tone.However, CVCV (or bigger) forms end in high (c.).These patterns mirror the patterns of class A in isolation in Kolokuma (detailed in §3.1).That the high tone self-associates only in isolation suggests that it is truly floating and not a pre-associated H on the final TBU.8

Class D tone alignment with V-TBU targets (Kolokuma only)
In Williamson's (1988) discussion of class D in Kolokuma, she states that the floating Ⓗ "is thrown back on the final syllable of the first morpheme when the second morpheme begins with a vowel" (p.262), showing another tonal effect of V-TBU targets.This is demonstrated with the class D word /ı̣ mbẹ lẹ ̀ⒽⓁ / 'sweet, tasty, interesting' in (43).In a., the ⒽⓁ associates as expected, while in b. the floating Ⓗ selfassociates to its sponsor while the Ⓛ associates to this initial V-TBU.
(43) Kolokuma Izon -Effect of V-TBU on ⒽⓁ association in class D This is a robust pattern in Kolokuma, seen across the collection of phrasal forms in (44) derived from /bekèⒽ Ⓛ / 'English, White, European' ([beke] in isolation).If the target word begins with a vowel, the Ⓗ consistently self-associates to the sponsor itself (b.).Data come from the updated Kolokuma dictionary (W&B11).
( 'calendar month' [ beké ẹ kẹ n-òvùrù ] 'week' (of seven days) In contrast to the examples in ( 43)-( 44) with class D2 /L ⒽⓁ /, there is variation in the surface patterns with classes D1 (/H ⒽⓁ /) and D3 (/LH ⒽⓁ /) which end in a pre-associated high.Consider the D1 word /wárı̣ ́ⒽⓁ / 'house', whose floating tones straightforwardly associate to the target if it is consonant-initial, e.g.[wárı̣ tẹ mẹ ] 'wall gecko' (lit.'house spirit').However, (45) shows that if the following word is vowelinitial, there is variation between the Ⓛ associating to the vowel (a., presumably with deletion of the Ⓗ), versus the Ⓗ associating to the vowel and the Ⓛ following it (b.).There does not appear to be a consistent pattern and I take this to be incidental variation.For example, the same phrase for 'wall' which has Ⓛ align to the vowel in a. shows Ⓗ aligning to this vowel in a sentence in c.This variation also affects D2 /L ⒽⓁ / words, e.g. the derivatives of /kẹ ǹı̣ ̀ⒽⓁ / 'one, a certain' in (46).Example a. shows the expected pattern where the Ⓗ self-associates to its sponsor due to the presence of a V-TBU in the next word, while b.shows examples of it associating to this initial vowel of the target.

Frequency of each tone class
What is the frequency of each tone class in the Izon lexicon?In this section, I catalogue class frequencies in both dialects, based only on lexical morphemes, i.e. nouns, verbs, and adjectives.This sample does not include proper names, transparently derived words, ideophones/adverbials, or any grammatical morphemes (numerals, quantifiers, determiners, auxiliaries/light verbs, inflectional enclitics, among others).
Gbarain Izon frequencies are shown in Table 14.From a sample of 513 lexical morphemes, classes A, B, and C are evenly distributed, constituting about a third of the vocabulary each.If a lexical item showed variation between two classes, each class was given 0.5.A database of these morphemes is found in the supplemental materials (see Appendix D).
These 513 items were collected sporadically and opportunistically in order to have a large enough sample of vocabulary for deducing phrase and sentence level tonal changes.Nouns and verbs are represented in each subclass, however nouns of subclass C2 (/H Ⓛ /) are quite rare compared to verbs.Within class B (Ⓗ), no lexical morphemes are of class B2.This subclass strictly consists of a small number of grammatical morphemes, e.g. the demonstrative /ẹ ǹdı̣ ̀Ⓗ/ 'that'.Within class C, the distribution is fairly even internally, but recall that class C4 (/L Ⓛ /) is missing in Gbarain.Finally, as stated class D (ⒽⓁ) is limited in Gbarain to the pre-vocalic pronouns; no class D lexical morphemes exist.
A much wider sample of lexical morphemes can be catalogued for Kolokuma Izon, based on Williamson & Timitimi's (1983) dictionary.Excluding non-lexical morphemes (as well as proper names and ideophones), this results in a database of 1868 lexical morphemes.As above, the .5 designation indicates morphemes which varied between two subclasses.As above, see the supplemental materials for the corresponding database.
Classes A, B, and C are fairly evenly distributed, though fewer class C items occur than we saw in Gbarain.Class A were almost entirely /L ⓁⒽ /, with a marginal number of A2 in which the floating Ⓗ does not selfdock (indicated by the asterisk -see §3.1.2).Subclass B2 is entirely missing in Kolokuma.Class C is particularly skewed.Classes C1 /…HL Ⓛ / and C3 /…LH Ⓛ / are well represented, but C2 /H Ⓛ / and C4 /L Ⓛ / are very marginal classes for lexical morphemes.Finally, class D constituted the smallest class.It is therefore not surprising that this class does not exist in Gbarain Izon (a loss reported even among certain Kolokuma Izon speakers as well -W&T83:xxxii).
Nouns and verbs in general are well-represented within each class in Kolokuma, but have an interesting distribution within class D. While D1 has only a moderately skewed distribution of nouns (n=10.5/38.5)versus verbs and adjectives (n=28/38.5),class D2 is made up almost entirely of nouns (n=20/22) while class D3 almost entirely of verbs/adjectives (n=25/26).Although D2 is rare for monomorphemic verbs, a productive process derives D2 verbs with the final intransitivizing suffix -i/ị.Part of the morphological expression is changing the tone pattern to D2 (L ⒽⓁ ), overwriting lexical tone.This is demonstrated in (47).We can begin by examining tone strings of 0-2 tones, and comparing it to the databases of lexical morphemes for both dialects ( §4.1).Table 16 shows that lexical morphemes consisting of 0 tones (i.e.toneless) or 1 tone are absent in both dialects (a.-c.).I interpret this as a systematic gap for both dialects.All systematic gaps are denoted with asterisks, and are shaded gray.While absent for lexical morphemes, we have seen several functional morphemes which are toneless (e.g./kpọ / 'also') or have inherent but not floating tone (e.g./bı̣ / DEFINITE).
Tone strings consisting of 2 tones (d.-g.) correspond to several attested Izon patterns, though most logically possible combinations are unattested.These include the four gaps in (48) below, which I interpret as systematic.After these systematic gaps are taken into account, there are only four remaining patterns -L Ⓛ , L Ⓗ , H Ⓛ , and H Ⓗ -all of which correspond to B and C tone classes in Izon (indicated in Table 16 in parentheses).While L Ⓛ is absent in Gbarain and rare in Kolokuma, it exists in enough names for places or people that it should not be considered a systematic gap.The same holds for L Ⓗ , which does not occur as a lexical morpheme but found for some pre-nominal modifiers in Gbarain (/ẹ ndı̣ Ⓗ / 'that', §3.2.1).Because I interpret these as non-systematic gaps, I denote them with 0 rather than *.48), e.g.gaps due to tone identity (rows a., h.) and the floating tone requirement (rows c., f.).Many other logically possible patterns are not shown, as they would be automatically ruled out as well, e.g.sequences of all floating tones.Such a constraint on tonal troughs is common cross-linguistically and tonologically natural (Yip 2002: 137).This automatically rules out all of the patterns in f.Thus far, I have proposed that there is a contrast between pre-associated tones and tones which are not pre-associated and thereby 'floating'.It is worth asking at this juncture whether we can conflate these two tone types into a simpler representation.One alternative is having only pre-associated tones which delink under specific conditions, i.e. no floating tones in the input.A second alternative is having only floating tones which link to the sponsoring morpheme and target morphemes under specific conditions, i.e. no preassociated tones in the input.Such alternatives would have the advantage of being representationally uniform, and in most cases the corresponding surface pattern of the tone string is predictable from the tones in the input.For example, the tone string H L always corresponds to a morpheme with an underlying form /H Ⓛ /, and does not contrast with /HL/ or / ⒽⓁ / representations.The crucial evidence for positing representations with mixed pre-associated and floating structure comes from L H L tone strings (row c. in Table 17).Here, there is such a contrast, namely between /LH Ⓛ / (C3) vs. /L ⒽⓁ / (D2) in Kolokuma Izon.Examples are in Table 18.These data show that it is not predictable whether the H in these strings will be pre-associated to the final TBU of the sponsor or will be floating and self-associates only in the absence of a suitable target host.
Finally, let us examine tone strings of 4 to 6 tones, which represent the upper limit of tones for single lexical morphemes, in Table 19.Most of the logically possible patterns in this table are ruled out by the systematic gaps above.For others, new statements are required.To rule out a tone string L L H L mapping to /L ⓁⒽⓁ / (row c.), I propose a systematic gap on a sequence of three floating tones, in (51).
1 1 The systematic gap on 'no troughs' (49) automatically rules out several logically possible patterns (e.g., within rows f., j., k., l., and n.).Underlying sequences of a pre-associated HLH trough are actually attested, however, in both Gbarain Izon and Kolokuma Izon (k., q., r., s.)In both dialects, many of these words are loans, peripheral/special vocabulary (such as plants and animals), or reduplicants.Some words also appear to have multi-morphemic origins.For example, /òkókópọ ̀lı̣ Ⓛ / 'parrot' exhibits mixed vowel harmony categorically banned in single morphemes.Williamson & Blench (2011: 166) attribute this anomalous word to a borrowing from Igbo ([òkóòkó] 'parrot') in combination with the English word Polly, a common name for a parrot.Still, there are some basic vocabulary which cannot be explained away in this way, e.g.Kolokuma /áràụ Ⓛ / 'she (full pronoun), female' and /fúròúⓁ/ 'stomach'.These findings require a revision to the systematic gap on troughs to only applying to derived troughs, i.e. those where the HLH pattern is not a pre-associated sequence in the input.Competing analyses: Ubiquitous floating tone vs. obligatory spreading The analysis of this paper accounts for the various tonal patterns via ubiquitous floating tone: every lexical item in Izon has both pre-associated tone as well as floating tone at its right edge (with the possible exception of some class C subclasses - §3.3.2).This is typologically unusual, as floating tones are typically associated with only a minority of items in a lexicon or restricted to specific grammatical contexts.
A key aspect of this analysis is that a floating tone can be sponsored which is identical to the final preassociated tone, i.e. adjacent low tones in class A /L ⓁⒽ / and classes C1 and C4 /…L Ⓛ /, and adjacent highs for classes B1 and B3 /…H Ⓗ / and D1 and D3 /…H ⒽⓁ /.One obvious objection to this analysis is that it is a contradiction of the OCP (Obligatory Contour Principle), which bans adjacent identical elements (Leben 1973, Goldsmith 1976, Kager 1999: 398, a.o.).Even early on, however, there emerged several analyses of tone systems which tolerated OCP violations, e.g.Odden (1982) for Shambala [ksb] and Clements' (1984: 288) discussion of 'geminate tone melodies' in Kikuyu [kik].Further, given the widespread adoption of Optimality Theory (Prince & Smolensky 2004) where constraints are violable with there being no restrictions on underlying structure (i.e., Richness of the Base), underlying sequences such as /H Ⓗ / must be seriously entertained.Such an analysis in fact has been recently argued for the Papuan language Awa [awb] in McPherson (2016) based on Loving (1973).From the tonal patterns in Awa, McPherson (2016:e43) concludes that "all L-final nouns are followed…by a floating L tone", a representation which "may be viewed as the phonologization of phonetic L carryover".Izon may have had a similar trajectory.
An alternative approach brought up by an anonymous reviewer is OBLIGATORY TONE SPREADING of the final tone of the first word of a tone group, an alternative brought up already in §3.2.3.The result would be a much more limited use of floating tone in underlying representations, i.e. it would no longer be 'ubiquitous' across the lexicon.The two analyses are compared in Table 21 with a tone group [wó wárı̣ ] 'our house'.Under ubiquitous floating tone, the modifier would be represented /wóⒽ/, while under obligatory tone spreading it would be simply /wó/.Under this alternative, lexical items of all tone classes would undergo obligatory tone spreading.A rendition of how this would look for each subclass is provided in Table 22, which consists of a proposed underlying representation for each tone class, and how tone would spread within a tone group.Spread tones are denoted with a checked underline.Floating tones are circled, as throughout.One can see several commonalities between the analyses.First, while the tone spreading alternative eliminates floating tone in classes B and C, it still requires it for classes A and D. Second, both analyses require spreading at some level: in the alternative, it is the primary means for valuing toneless TBUs within the tone group, whereas under ubiquitous floating tone it is secondary to floating tone association.Third, I mentioned above the idea that identical floating tones are the phonologization of phonetic carryover (McPherson 2016).Such a functional motivation equally applies if one interprets obligatory spreading as what has been phonologized.
Turning to key differences between the analyses, if one were to adopt the spreading analysis, two issues need to be settled.First, under the alternative all classes would undergo tone spreading.For classes such as C3 and D2 in Table 22, effects of such spreading would not be transparent in the tone groups (their surface forms are [(L.H).(L.L)] and [(L.L).(H.L)], respectively).One would be forced to posit underlying representations with final contours, i.e. /L.H ͡ L/ and /L.L ͡ H Ⓛ /.Sample derivations involving C3 and D2 morphemes are in Table 23 (data from Kolokuma Izon).Furthermore, under the ubiquitous floating tone analysis non-initial words of the tone group are valued with floating tones due to a straightforward constraint against unassociated tones in the output (e.g.*FLOAT in an OT approach).The alternative analysis would be required to employ some equivalent constraint to enforce tone spreading which must take place even in those contexts when all TBUs are already valued.Thus, it cannot be reduced to toneless TBUs needing valuation (see §3.2.3).
Two types of constraints in the (OT) literature could accomplish this task.One type are constraints which favor spreading, e.g.constraints EXPRESS(F) (Cole & Kisseberth 1994), SHARE(F) (McCarthy 2011), or SPREAD(F) (Kimper 2011), where F stands for a phonological feature (in the Izon data, it would be replaced by 'T' for tone).The other type of constraints are indirect, prohibiting sequences of mixed values of some feature in or across some domain, e.g.*INTERWORD[-ATR][+ATR] in Kügler (2015) Kula & Bickmore (2015).This could be generalized as a constraint * INTERDOMAIN[αF][βF].

and *INTERWORD[H][L] in
Regardless of the exact type of constraint employed, it must specifically not apply iteratively to capture the fact that tone does not spread in several contexts, e.g./ẹ ǹdı̣ ̀Ⓗ/ does not spread to /bı̣ / in [ẹ ǹdı̣ ̀ námá bı̣ ] 'that animal' (cf.[námá bı̣ ] 'the animal'); again, see the data and argumentation in §3.2.3.How to accomplish this requires amending the grammar, e.g.incorporating Harmonic Serialism (Kimper 2011).I leave further comparison of these analyses to future work.
Finally, several tone patterns illustrated a propensity for low tone to align with a vowel-initial word (a V-TBU), in contrast with the tone patterns with consonant-initial C-TBU counterparts.Such exceptional tone effects are summarized in (58).As stated, in at least some of these contexts I have interpreted the initial vowel as being extrametrical, which following Downing (1998) constitutes misalignment between morphological and prosodic constituents.Finally, I compared the ubiquitous floating tone analysis advocated for here with an alternative involving obligatory tone spreading.While a full evaluation of this alternative remains, I identified several key differences and similarities.Importantly, positing ubiquitous floating tones at the right edge of a word or positing obligatory spreading of the final pre-associated tone both can be interpreted as the phonologization of pitch carry-over into the target domain.

Appendix A: General tone rules in Gbarain
There are two general tone rules in Gbarain Izon.The first rule is 'H-absorption', whereby a rising tone decontours and becomes low before a high tone, i. (59) Gbarain -H-absorption with /kụ ̀rẹ ı̣ / 'be able to, can' The second rule is 'low-to-mid raising', and applies when a high-low sequence appears at the end of an utterance, i.e.HL# > HM# utterance finally.In such sequences, the final TBU was produced higher in some tokens ([˦ ˧]) and lower in others ([˦ ˨], with a non-falling low tone), but was not produced with low tone which falls to the lowest part of the pitch range (i.e.*[˦ Å]).Low-to-mid raising is demonstrated in (61).
(61) Gbarain -Low-to-mid raising: HL# → HM# utterance finally [Gb-20190706:30] a. / wárı̣ ̀Ⓛ kpọ / → wárı̣ ̀ kpọ ̀ → [ wár kpọ ̄ ] 'also a house'  Classes A and B in Gbarain correspond straightforwardly to the same classes in Kolokuma (a.-c.).The same holds for the C1 subclass in Gbarain (d.).Thereafter, there are mixed class correspondences (e.-i.).The majority of C2 morphemes in Gbarain correspond to B1 /H Ⓗ / in Kolokuma.Class C3 in Gbarain is the most complicated and irregular set.It corresponds fairly evenly to all four tone classes in Kolokuma (and many subclasses as well), with no clear majority and each correspondence set being relatively small (n<15).Furthermore, as established class D is absent in Gbarain except for the few pre-vocalic pronouns.Class D lexical morphemes in Kolokuma generally correspond to the class C in Gbarain.One can see for D1 and D2 that the floating Ⓗ is 'pulled' one TBU to the left in Gbarain, e.g.docking to the final TBU in h.In contrast, no major patterns emerge with D3 given their limited number; they fairly evenly correspond to C3 (…LH Ⓛ ) and B3 (<L>H Ⓗ ) in Gbarain.

Appendix C: Comparison to the tone class analysis of Kay Williamson
The proposed analysis differs in crucial ways from the analyses in Williamson (1965Williamson ( , 1978Williamson ( , 1988) ) and Williamson & Timitimi (1983) of Izon.The clearest differences are with Williamson (1988), which examines tone in Kolokuma Izon against three other Ijoid varieties.Here, she posits three types of tone: pre-linked tone, domain tone, and floating tone.Pre-linked tone stays on the TBU syllable "with which it is associated in the lexicon" (Williamson 1988: 257).It is both immobile and does not spread to other TBUs.In contrast, domain tone (which I conventionalize as a circled lowercase ⓣ) is directly associated to the tone group itself, rather than to a traditional tone-bearing unit such as the mora.Williamson notes that the "defining feature of a domain tone is that it links to the leftmost free TBU and then automatically spreads through its domain until interrupted" (p.256).Finally, floating tones -circled uppercase Ⓣ -are mobile and "surfaces in different positions in accordance with rules which vary from dialect to dialect" (p.256).The behavior of these three tone types are summarized in Table 27.Williamson proposes that these three types combine to form five tone classes across Ijoid, which she variably labels with Roman numerals (Williamson 1965), common numerals (Williamson & Timitimi 1983), or letters (Williamson 1988).Table 28 provides a comparison of the conventions of this paper to these previous works by Williamson.Williamson proposes a complex grammar based on the three types of tone, coupled with multiple phonological cycles interacting with default tone.Full comparison between her analysis and the one in this paper would be too lengthy to include here; I leave the interested reader to compare the analyses directly.

Appendix D: Data collection and supplemental materials
The Gbarain data in this paper were collected in the summers of 2017 and 2019, in Port Harcourt and Ibadan, Nigeria.All Gbarain data comes from one proficient speaker named Jumbo Gift, from the Gbarain Izon community Okolobiri [òkólóbırı] in Bayelsa State (5°02'N 6°19'E).He is highly proficient in Izon and uses it frequently with family and fellow Izon people.Recordings were made with a Tascam DR- g. Gbarain H.L.L [ánɡɡı̣ sı̣ ] 'handkerchief', L.H.L [ọ ̀tọ ́kọ ̀] 'mud', and L.L.H [àkàlụ ́] 'moon'.Surface contours are typically found at the right word edge, e.g.L.L.H ͡ L [ı̣ sụ ̀sụ ̂] 'garden egg' or L.L.L ͡ H [eɡberı] 'story'.Surface patterns of strings of high tones are very common, e.g.L.H.H [ıkókú] 'waist' and H.H.H [teḱeĺe] 'lift up', and nearly all lexical items occur with a H tone somewhere (either pre-associated or floating).Entirely L-toned surface patterns do not occur in the Gbarain Izon dialect and are rare in the Kolokuma dialect where they are restricted to names for places or people, e.g.L.L [ıɡbòn] 'northerner, Hausa'.

( 47 )
Kolokuma Izon -Productive derivation of D2 verbs with intransitive -i/ị aLogically possible combinations of pre-associated tone and floating tone I have proposed that tone class contrasts be analyzed as underlying sequences of pre-associated and floating tones.How do these representations stack up against the range of logically possible combinations of preassociated tone and floating tone?This exercise will allow us to see common gaps, and help interpret them as accidental or systematic.In the tables below, possible combinations range from 0 to 6 individual tones in the input, forming various tone strings of H and L. For example, from a tone string of two tones H L, possible representations are (i) both pre-associated, HL, (ii) one pre-associated and the other floating, H Ⓛ and Ⓗ L, or (iii) both floating, ⒽⓁ .As throughout, underlying floating tones are circled.

( 48 )
Systematic gaps in Izon tone class contrasts a. Tone identity: No adjacent identical tones of the same type (pre-associated or floating) e.g.LL, HH, ⓁⓁ , ⒽⒽ (note: this does not exclude L Ⓛ or H Ⓗ ) b. Floating tone position: No floating tones before pre-associated tone e.g.Ⓛ H, Ⓗ L, Ⓛ L, L Ⓗ L, H Ⓛ H, etc. c.Floating tone requirement: No sequences without floating tones e.g.L, H, LH, HL, etc. d.No all floating: No sequences consist only of floating tones e.g.Ⓛ , Ⓗ , ⒽⓁ , etc.

( 53 )
Systematic gap (revised) -No derived troughs: No sequences of tonal 'troughs' H L H in the output which were not pre-associated in the input, e.g.HL Ⓗ , H ⓁⒽ , HLH Ⓗ , HL ⓁⒽ Finally, one row from Table19still remains without explanation: the string of tones L H H L in row g. corresponding to a pattern /LH ⒽⓁ /, which is unattested.A related string L H H (row d. in Table17) corresponding to a pattern /LH Ⓗ / is also virtually unattested in either dialect.Examples with a L H H (L) string generally involve an initial extrametrical vowel, i.e.D3 <à>kụ ⒽⓁ 'bitter', where the initial L is not parsed as part of the phonological word.To account for the non-occurrence L H H (L) patterns, I posit a final systematic gap, *LH#H.(54) Systematic gap -*(LH#H): No sequences of LH followed by H across a word boundary within a tone group This gap would not rule out forms like <à>kụ ⒽⓁ because the initial vowel would be outside of the domain where this ban is evaluated, i.e. the tone group. 10

( 58 )
Exceptional tone effects of vowel-initial words • Class A: Floating Ⓛ docks to the V-TBU, and floating Ⓗ docks immediately after it, analyzed as extrametricality of the initial vowel • Class B (Gbarain only): L tone from initial word spreads to V-TBU, and Ⓗ docks immediately after it • Class C: Vowel-initial words overwhelmingly have an initial low tone rather than high tone (e.g.ratio of 15:1 in Kolokuma) • Class D (Kolokuma only): if target is vowel-initial, floating Ⓗ self-associates rather than associating to the V-TBU • Class B and D: Subclasses B3 and D3 begin with an initial low-toned vowel, in (near) complementary distribution with consonant-initial high-toned B1 and D1 e. /LH.H/ à [L.H].Examples (59)-(60) illustrate the absorption of the high component of an underlying rising tone.The relevant portion is underlined.

Table 1 :
Types of tonal configurations using autosegmental representation

Table 2 :
Gbarain Izon -Tone neutralization of nouns in tones groups

Table 6 :
Kolokuma Izon -Exceptional class A2 which surface as all L in isolation Within tone groups, Kolokuma class A patterns by and large replicate the Gbarain patterns, shown in 12) Kolokuma Izon -Class A ⓁⒽ association in a tone group a. / ıfırı̀Ⓛ Ⓗ Unlike classes B1 and B3 which have many members, I have found only two words of class B2 in Gbarain, the other being /ı̣ ndàⒽ/ 'how many'.In Kolokuma, there are no members.Two examples with this latter modifier are in (16).Note that the question particle /à/ falls outside of the tone group, and therefore surfaces with its lexical low tone.(16)Gbarain Izon -Class B2 nominal modifier /ı̣ ndàⒽ/ 'how many' Notice in b. that the Ⓗ does not directly dock to the initial V-TBU of the target, unlike consonant-initial targets.This represents another instance of V-TBUs showing exceptional tonal behavior, preferably associating to low tone of the first word rather than the floating Ⓗ. Parallel data are found with /ẹ ǹdı̣ ̀Ⓗ/ 'that', e.g. in [ẹ ǹdı̣ ̀ òpórıó́pó] 'that pig'.
Comparable patterns are found for Kolokuma Izon subclasses B1 (/H Ⓗ /) and B3 (/LH Ⓗ /), in(17).These data are identical in Gbarain.[Note that the floating Ⓗ spreads to all subsequent TBUs in the tone group in Kolokuma Izon, as it does in Gbarain.This is unlike the floating Ⓗ of class A in Kolokuma, which associated only to one TBU and those thereafter were valued with default low.Compare the form of non-initial words in class B context [dıŕı́ gụ ́ọ ́ kı̣ ḿı̣ ] 'sorcerer' (a.immediately above) to class A context [bùrù gbòrò kı̣ mı̣ ] 'yam-planter' ( There are numerous members of the B3 class which begin with a low-toned vowel, a sample of which are in a. above.In contrast, there are very few class B1 members which begin with a high-toned vowel (hence, the near-complementarity); the complete list found for Gbarain is in b. above.All of these exceptional B1 forms begin with a non-high vowel, and either have no medial consonant (/óúⒽ/ 'cry') or a high-sonorant medial consonant /r y/.The consonant-initial B3 class is even smaller, consisting of only two items, /bọ ̀ụ bọ ụ Ⓗ / 'very soft' and /bụ ̀rụ bụ rụ Ⓗ / 'rotten' (both showing internal reduplication).An example is in (19), which is transparently related to class A /bụ rụ ̀ⓁⒽ / 'be rotten'. [Gb-20190702:2]

Table 9 :
Kolokuma Izon -Co-occurrence of V-TBU and low tone across class C
); the relevant portion is underlined.
PFTV'Ebi put out the corn'In another variant, however, the final H of the sponsor spreads to this first V-TBU, and the floating Ⓛ associates immediately after it.This results in a [H#HⓁ] pattern, exemplified in (36).This is not expected given the propensity for onsetless syllables and low tone to coincide.Both patterns were common in the Gbarain data collected.
PFTV'Ebi put me out' (as if on fire) b.Cf.*[EQ bı̀ ıǹè òvıńmọ m̀ ]3.4.2 Core patterns of class D in Kolokuma IzonIn Gbarain, tone class D is limited to the small class of pronouns.In contrast, in Kolokuma they have a much wider distribution, found with many lexical items as well as several modifiers (e.g.demonstratives).
) Kolokuma Izon -Class D lack of self-association of floating Ⓛ in isolation Notice as well that D1 and D3 are in complementary distribution: D1 always begins with a consonant and is all high-toned, while D3 begins with a low-toned vowel, followed by all high tones.This is parallel to the patterns in class B, where B3 patterns were analyzed with initial extrametricality of the low-toned vowel, i.e. /<e>reíń Ⓗ / 'day'.I therefore extend the extrametricality analysis to the b.forms in (42), e.g./<ı̣ >fı̣ fı̣ ýọ ́ụ ́ⒽⓁ / 'whistle'.While in class B there was only near-complementarity due to the presence of exceptional B1 forms such as /eŕı́Ⓗ/ 'dry', in class D there is complete complementarity, i.e. no items of the shape */átáⒽ Ⓛ / or */bàtáⒽ Ⓛ /.

Table 21 :
Ubiquitous floating tone vs. obligatory tone spreading

Table 22 :
Analysis comparison across tone classes

Table 23 :
Kolokuma -Ubiquitous floating tone vs. obligatory tone spreading(classes C3 and D2)Under the alternative, the underlying forms of these morphemes would have final contours, i.e. /òsûn/ and /bekěⓁ/.Intermediate representations would be required where the final tone of the contour spreads, followed by decontouring rules at the surface (i.e.falling H ͡ L#L becomes [H#L]).Although some decontouring is attested in Izon (see the general tone rules of Gbarain, Appendix A), we have seen several places already where falling tone appears before surface low, in (55).Here, there is no decontouring.

Table 26 :
Tone class correspondences between dialects

Table 27 :
Williamson's three types of tone across Ijoid

Table 28 :
Tone class equivalences compared to work of Kay Williamson