This article analyzes the process that I call Liquid Lowering which turns high vowels
$i$
and 
into
$e$
before liquids, schematically 
. The process began to operate in Polish in the 16th century. I look at the modern reflexes of Liquid Lowering in Standard Polish and in Kurpian, a dialect of Polish that dates back to the 17th century, and argue that the rule is dead in Standard Polish but not in Kurpian, where it is productive in derived environments.
The modeling of Liquid Lowering as a phonological process has implications for phonological theory. In particular, it calls for the recognition of derivational levels, as envisaged by Derivational Optimality Theory. It is argued that Standard Optimality Theory, with its principle of strict parallelism, cannot account for the data because it runs into insoluble ranking paradoxes. The analysis bears on the issue of abstractness by positing vowels that never occur phonetically. The abstract vowels are exchanged for the actually occurring vowels before reaching the surface representation. I term this type of exchange a virement shift.
The research reported here is based one hundred percent on my fieldworkFootnote [2] since, first, in general, the literature on Kurpian is extremely small and, second, Liquid Lowering is not even mentioned in any of the sources. Apart from brief remarks in books on dialectology such as Zduńska (Reference Zduńska1965) and Dejna (Reference Dejna1973), Kurpian, but not specifically Liquid Lowering, is discussed in Friedrich (Reference Friedrich1955), which is a survey of Kurpian as spoken in a number of villages in Kurpia in the 1930s.Footnote [3] The recent work on Kurpian is that of Rubach (Reference Rubach2009, Reference Rubach2011a, Reference Rubach2014a, Reference Rubachb) but none of these sources deals with Liquid Lowering.
This article is organized as follows. Section 1 introduces background information about Kurpian. Section 2 identifies two main classes of verbs that are relevant for Liquid Lowering. Section 3 discusses Liquid Lowering and states basic generalizations. Section 4 proposes an OT analysis of the data. Section 5 summarizes the conclusions. The development of the line of argumentation can be outlined as follows.
There are two main arguments in this article that are noteworthy from the point of view of general phonological theory: abstract vowels and a derivational step. Liquid Lowering turns high lax unrounded vowels into tense [e], so the rule appears to be: 
Footnote
[4]
$\rightarrow$
[e] before liquids. Modeling this rule in Optimality Theory (OT; McCarthy & Prince Reference McCarthy, Prince, Beckman, Dickey and Urbanczyk1995, Prince & Smolensky Reference Prince and Smolensky2004) runs into difficulty. Since the rule prohibits 
before a liquid, the OT’s principle of minimal repair predicts that 
should change into lax [ɛ] since then IDENT[
$\pm$
tense] is not violated and [ɛ] satisfies Liquid Lowering. However, this prediction is empirically incorrect because the attested output is tense [e] rather than lax [ɛ]. Since the Kurpian data do not warrant a further repair, ɛ
$\rightarrow e$
, it becomes evident that tense [e] can only be obtained if the input vowels themselves are [+tense]. That is the underlying vowels must be //i ɨ// rather than 
as then the rule 
before a liquid works as desired. The interest of this prediction made by OT is that //i ɨ// are abstract vowels that never occur phonetically in Kurpian. The abstract /i ɨ/ are changed into 
before reaching the surface representation, a context-free exchange that I call virement. For this analysis to work, OT must admit a derivational step: the virement 
must take place after Liquid Lowering 
has taken place, so OT must admit at least two derivational levels: Level 1 at which Liquid Lowering operates and Level 2 at which virement is active.
The derivational step argument is strengthened by the analysis of the alternation between surface lax 
and tense [o], as in 
‘to live’ – [zo+ɫ] ‘he lived’. OT cannot deliver a direct change from //i ɨ//
$\rightarrow$
[o] because this change runs against the principle of minimal repair. The principle predicts that the output should be [u], not [o], because [uL] satisfies Liquid Lowering by removing the prohibited [iL ɨL], where L stands for a liquid. Saving the analysis by prohibiting [uL] will not work because Kurpian admits [uL] in an unrestricted way. The solution is to postulate that Liquid Lowering takes //i ɨ// to /e/ at Level 1 and then, at Level 2, /e/ goes to [o]. In sum, the analysis requires a derivational step.
1 Background
Kurpian is a dialect of Polish spoken in Kurpia in northern Poland. Even though Kurpia is inhabited by some 70,000 people, the speakers of the dialect are few in number and are invariably people of the older generation (70 years or older). Therefore, the dialect is on the verge of extinction. However, it should be noted that much effort is currently being made by Kurpians themselves in order to revive their culture and language.
Kurpian has a rich system of vowels, considerably richer than that of Standard Polish. In (1) I show the system of vowels that are contrastive in minimal or near minimal pairs (Rubach Reference Rubach2011a).
Examples of words illustrating Kurpian vowels create an opportunity to explain the spelling system. The system was devised by Rubach (Reference Rubach2009) and is closely phonetic.
Let us add that prepalatals 
are always written with an accent in Kurpian, so 
‘gray’, śano [ɕanɔ] ‘hay’, 
[ʑɪma] ‘winter’, źarno [ʑarnɔ] ‘seed’, 
‘silent’, ćotka 
‘aunt’, dźådek [dʑɑdɛk] ‘grandpa’, 
[dʑɪvnɨ] ‘strange’, and 
[ɲɪska] ‘bowl’, ńebo [ɲɛbɔ] ‘sky’.Footnote
[8]
In terms of distinctive features, Kurpian vowels are classified as follows:
The feature [
$\pm$
tense] plays a prominent role in the classification of Kurpian vowels. The understanding of [
$\pm$
tense] follows Wood (Reference Wood1975), who defines this feature as a degree of constriction in the regions of hard palate, soft palate, pharynx and lower pharynx. The consequence is that upper high, upper mid and ‘backer’ (back retracted) vowels are [+tense]. Thus, Kurpian
$a$
[a] and å [ɑ] contrast by [
$\pm$
tense], the former is [–tense] while the latter is [+tense]. The operation of phonological rules supports this classification (see Rubach Reference Rubach2014a,Reference Rubachb).
While there is no doubt that //o// is a separate phoneme (underlying segment), there is a context in which [o] is systematically derived from //ɔ//. The context in question is that of a following nasal consonant. Rubach (Reference Rubach2011a) has noted a systematic phonotactic generalization: only [o] and never [ɔ] is found before a nasal. This generalization leads to the following rule. Informally:
The rule is entirely exceptionless, so we find [o] in, for example: dóm ‘house, tón ‘tone’, and kóń ‘horse’.
A reviewer asks whether it is necessary to derive phonetic [o] from underlying //ɔ//, as envisaged by Nasal Tensing. The reason for the query is that [o] before nasals never alternates with [ɔ], so, it appears, the Alternation Condition (Kiparsky Reference Kiparsky1968/1973a) would bar any discrepancy between the surface representation and the underlying representation, leading to postulating underlying //o// rather than //ɔ//. I argue below that the Alternation Condition is not applicable to Nasal Tensing and hence //ɔ// is a licit representation.
The Alternation Condition discussed in Kiparsky (Reference Kiparsky1968/1973a) but actually formulated in Kiparsky (Reference Kiparsky1973b) refers to neutralization rules and those are defined as follows.
Let us elucidate this definition of neutralization by looking at an example from Standard Polish.
The data in (6a) exhibit an alternation between a voiced and a voiceless obstruent, for example, [d]–[t] in kody [kɔdɨ] ‘code’ (nom.pl) – kod [kɔt] (nom.sg). The examples in (6b) exclude the possibility of analyzing the alternation in terms of intervocalic voicing because voiceless obstruents occur between vowels. Consequently, the alternation must be due to the rule of Final Devoicing. Informally:
Final Devoicing takes underlying //kɔd// to surface [kɔt], kod ‘code’. In the case of kot //kɔt// ‘cat’, the structure derived by Final Devoicing is present before Final Devoicing has a chance to apply, as the underlying //t// of kot occurs before a word boundary. This means that kot ‘cat’ constitutes a (C)BD, where
$C$
is an irrelevant context from the point of view of Final Devoicing. I conclude that Final Devoicing is s neutralizing rule.
This conclusion has consequences for the analysis of words that do not exhibit alternations. Typically these are words that do not occur in inflectional paradigms.Footnote [9]
The underlying representations of these words appear to be ambiguous because the final obstruent could be derived either from a voiced obstruent via Final Devoicing or from a voiceless obstruent: //tag//
$\rightarrow$
[tak] or //tak//
$=$
[tak] ‘yes’ and //lub//
$\rightarrow$
[lub] or //lup//
$=$
[lup] ‘or’. Since Final Devoicing is a neutralizing rule, the Alternation Condition sets in and resolves the ambiguity in favor of the underlying representations that are the same as the surface representations, so //tak// ‘yes’, //lup// ‘or’ and so forth. The resolution follows from the tenet that non-alternating forms must be entered in the underlying representation in their surface representation. The underlying representations enforced by the Alternation Condition are verified by novel formations: po+tak+iwa+ć
‘say yes’ and obocz+n+ość
Footnote
[10]
‘alternation’. The new words are derived from the representations with a voiceless obstruent.
The intention of Kiparsky (Reference Kiparsky1968/1973a) was, on the one hand, to limit the abstractness of underlying representations and, on the other hand, to keep underlying representations free of phonetic details that can be supplied by general rules. Kiparsky (Reference Kiparsky1973b) draws the dividing line between excessively abstract and excessively phonetically detailed representations by appealing to neutralization: only neutralization rules are subject to the Alternation Condition. This requirement is fulfilled by Final Devoicing, as just discussed, but not by Nasal Tensing (4). The point is that Nasal Tensing does not meet the criterion for a neutralizing rule because, unlike in the case of Final Devoicing, there are no CBDs to speak of: all instances of [o] before a nasal in the surface representation are an effect of Nasal Tensing and there is not a single instance on record that would have a tense [o] before a nasal from an independent source. Consequently, nothing in the system mandates that the vowel in dóm ‘house’, tón ‘tone’ and kóń ‘horse’ should be tense //o// in the underlying representation. On the contrary, the intention of the Alternation Condition is that phonetic details should not be entered in the underlying representation. There is a phonotactic generalization: lax [ɔ] cannot occur before a nasal. Classic generative phonology expresses this generalization by underspecifying underlying representations and postulating a rule (here: Nasal Tensing) that fills in predictable information, ɔ
$\rightarrow$
o before a nasal. Derivational Optimality Theory, which is the framework adopted in this article (see Section 4) shares with classic generative phonology the assumption that phonotactic generalizations are analyzed by evoking underspecification and rules/markedness constraints that enforce the derivation of the attested surface forms. Therefore the underlying representations of dóm ‘house’, tón ‘tone’ and kóń ‘horse’ are //dɔm//, //tɔn// and //kɔɲ//, respectively, with Nasal Tensing making sure that the phonetic representations are [dom], [ton] and [koɲ], as required.
It might be added that Nasal Tensing is a powerful process. It is entirely exceptionless and extends to all borrowings, including foreign names, for instance Camerón.Footnote [11] Furthermore, Nasal Tensing is the source of phonological interference (incorrect pronunciation) in the speech of some Kurpians speaking Standard Polish.
2 Ï -verbs and E -verbs
Liquid Lowering applies to a huge number of verbs that have 
as the verbalizing morpheme 
but never to otherwise similar verbs that have
$e$
as the verbalizing morpheme (
$e$
-verbs). This distinction in behavior provides crucial arguments for the discussion of Liquid Lowering and the theoretical issues that it entails. Therefore, it is necessary to identify the class of 
and the class of
$e$
-verbs. The task is not easy because the phonological structure of these classes is masked by the effects of two processes: Infinitive Lowering and Infinitive Tensing.
Looking at the issues from a general perspective, it should be noted that verbs in Kurpian, like in Standard Polish, are of two types: inherent verbs and derived verbs. Inherent verbs are those whose root is a verb, for example, paś+ć ‘pasture’ and zy+ć ‘live’. Derived verbs are formed from other parts of speechFootnote [12] by adding a verbalizing suffix. The word los+owa+ć ‘draw lots’ illustrates the point: los ‘lot’ is a noun, -owa is a verbalizing morpheme and -ć is the infinitive. Kurpian, like Standard Polish, has many verbalizing morphemes and verbs are divided into classes, depending on the suffix that they use, so los+owa+ć is an -owa class (see Rubach Reference Rubach1984).
With this background, we look at the following verbs:
Looking at dzwóńéć ‘ring’ (10a) and suńéć ‘hum’ (10b), for example, it would appear that the analysis should be as follows:
Since we see é [e] occurring after the noun root in both verbs, it seems that, first, both verbs belong to the same class and, second, the verbalizing suffix is é //e//. However, closer investigation shows that none of these statements is true. The similarity in (10) is apparent and is a result of some phonological processes that mask the underlying structure.
In order to determine whether we are dealing with one or two classes of verbs in (10), we look at selected forms of the conjugation paradigm.
The similarity of the verbs in (11) is true only in the infinitive and in the present tense. The past tense and the perfect participle are markedly different. For example, the 3rd person masculine form in the past has ó [o] in dzwóńół ‘he rang’ but å [ɑ] in suńåł ‘he hummed’. Similarly, the front vowels in dzwóńély ‘they rang’ (masc) and suńely ‘they hummed’ (masc) are different: tense é [e] versus lax
$e$
[ɛ], respectively. These differences are found in what are ostensibly the same contexts, so they cannot be ascribed to the operation of phonological rules.
Within their paradigms the verbs exhibit alternations. Looking at the past paradigm, the past tense morpheme appears either as -ł or as -
$l$
.
The verbalizing é [e] appears in two other shapes: as ó in dzwóń+ó+ł ‘he rang’ and as 
in 
‘having rung’. Therefore, the alternation chain is é [e] – ó [o] – 
Given the default assumption that one of the alternating segments is the underlying segment, the question is which segment to pick for the underlying representation. This question can be answered by looking at a larger fragment of Kurpian phonology.
We argue in the next section that Kurpian has a rule of Liquid Lowering, 
before a liquid, a descendant of the well-known process that affected Polish and its dialects in the 16th and 17th centuries. Given Liquid Lowering, é [e] is derivable, so it should not be posited in the underlying representation. This leaves us with the choice between the ó [o] of dzwóń+ó+ł ‘he rang’ and the 
of 
‘having rung’ as the representation of the verbalizing suffix. There are two reasons that argue against postulating ó //o// as the underlying representation. First, ó [o] is contextually restricted to closed syllables: it occurs in dzwóń+ó+ł ‘he rang’, but not in dzwóń+é+ł+a ‘she rang’. Second, the verb dzwóń+é+ć takes the endings containing 
in the present tense conjugation while most verbs take the endings containing
$e$
[ɛ].
The generalization true for both Standard Polish and Kurpian is that verbs containing front vowels as verbalizing morphemes take the endings 
while other verbs take the endings -es, -
$e$
, as shown in (13).
The contextual restriction of ó [o] to closed syllables and the distribution of the endings in (13) both argue for postulating 
as the verbalizing suffix in dzwóńéć ‘ring’:
Establishing the underlying representation of the verbalizing morpheme in suń+é+ć ‘hum’ is built on a similar line of reasoning. In surface terms, we have the following chain of alternating segments:
Of these alternants, the back vowels
$a$
[a] and å [ɑ] are not viable underlying representations for the same reason as in the case of dzwóń+é+ć. In the present tense, the endings are 
rather than -es and -
$e$
: 
‘you hum’ and 
‘he/she/it hums’. This leaves us with the choice between the tense é [e] occurring in the infinitive suń+é+ć and the lax
$e$
[ɛ] occurring in the past plural form suń+e+l+y ‘they hummed’ (masc). Both vowels are front, so the generalization concerning the endings in the present tense, shown in (13), is fulfilled in either case. That is, the question is whether suń+é+ć has the underlying representation 
with tense //e// or 
with lax //ɛ//.
The answer to this question is prompted by the observation that verbs of different classes invariably have tense é [e], and never lax
$e$
[ɛ] in the infinitive.
We thus arrive at the following distributional generalization: the only type of
$e$
-vowel that can occur in the infinitive is tense é [e]. Schematically:
Returning to suń+é+ć ‘hum’, since the occurrence of tense é [e] in the infinitive is predictable from Infinitive Tensing, it is lax e [ɛ] that is present in the underlying representation: 
, so //ɛ// is the verbalizing suffix.
A reviewer asks why é [e] could not be postulated as the underlying representation of the verbalizing suffix in the suńéć class of verbs in (9b). The vowels alternate (11), which motivates postulating a single underlying representation.Footnote
[16]
There are two reasons that point to underlying //ɛ//. First, as just explained, the alternation [ɛ] – [e] can be accounted for by Infinitive Tensing (17), //ɛ//
$\rightarrow$
[e] in suń+é+ć. The other alternant, [ɛ] occurs in suńely ‘they hummed’ (masc), which is predicted if //ɛ// is the underlying segment. Second, if //e// were the underlying vowel in suń+é+ć, we would need a rule that would take //e// to [ɛ] in suńely. A rule with that effect could not be general as Kurpian has not only [ɛ] but also [e] before liquids, for example, télny ‘back’ and grdély ‘strong men’. The putative e
$\rightarrow$
ɛ rule would therefore have to be postulated just for the past tense of the masculine forms. This complication is avoided by the analysis that postulates //ɛ// as the underlying vowel in suń+é+ć. The alternant é [e] is derived from //ɛ// by Infinitive Tensing (17), a rule that exists in the Kurpian grammar anyway, regardless of how we analyze the suńéć class of verbs.
We are now able to answer the question of whether dzwóńéć ‘ring’ and suńéć ‘hum’ belong to the same class of verbs. The answer is negative: suńéć is an
$e$
-verb while dzwóńéć is an 
The occurrence of tense é [e] in the infinitive is an effect of Infinitive Tensing. This is straightforward in the case of e-verbs: 
by Infinitive Tensing
$\rightarrow$
by other rules. In the case of 
, we need a lowering rule.
A derivational scenario for dzwóńéć ‘ring’ would thus be the following:
To conclude, the morphological system of Kurpian includes inter alia two classes of verbs: 
whose verbalizing morpheme is 
and
$e$
-verbs, whose verbalizing morpheme is
$e$
//ɛ//.
3 Liquid Lowering: facts and generalizations
Historical grammars, for example, Stieber (Reference Stieber1952) and Rospond (Reference Rospond1973), note that in the 16th century Polish developed a rule that changed ir [ir] and yr [ɨr] into er [er] and il [il], ił [iɫ], yl [ɨl], ył [ɨɫ] into [el], [eɫ]. I will call this rule Liquid Lowering and state it informally as follows:
The grammars are not clear on whether the output of the rule was lax [ɛ] or tense [e]. Both are viable because Polish made a phonemic distinction between tense and lax mid vowels, exactly as does Kurpian today. The distinction was lost in the 19th century.
Standard Polish today has obviously lax [ɛ] in the context of liquids because tense [e] does not exist as a sound any longer, so Standard Polish is not helpful in clarifying the query of whether the output of Liquid Lowering was lax [ɛ] or tense [e]. However, Kurpian is helpful at this point because systematically only tense [e] is found in the Liquid Lowering context in modern Kurpian (see the data below). Therefore, the odds are that the historical Liquid Lowering produced tense [e] rather than lax [ɛ], so I assume the output [e] in (22).
The data are based on Stieber (Reference Stieber1952) and Rospond (Reference Rospond1973) for Old Polish and on my fieldwork for Kurpian. Some representative examples, transcribed in IPA, are given in (22).
The effects of Liquid Lowering are preserved today as tense [e] in Kurpian and as lax [ɛ] in Standard Polish. The vowel is lax because, as noted earlier, Standard Polish lost tense mid vowels in the 19th century.
Liquid Lowering, so perfectly productive in the 16th and 17th centuries, seems to have been lost in Standard Polish, that is, the rule does not exist any longer. This is documented by the assimilation of borrowings. The early borrowings (16th and 17th centuries) exhibit the effects of Liquid Lowering today, so Latin subtilis ‘subtle’ that existed as subtylny [suptɨlnɨ], with [ɨ] before [l] in Old Polish, is subtelny in Modern Polish, with [ɛ] before [l]. In contrast, later borrowings (19th and 20th centuries) defy Liquid Lowering and have
$i$
[i] or
$y$
[ɨ] before liquids, as the following examples illustrate.
I conclude that Liquid Lowering does not exist in Standard Polish any longer. The situation in Kurpian is different. Liquid Lowering has exerted a huge influence on Kurpian, as is evident from the following comparison.
Liquid Lowering has affected also recent (20th century) borrowings from Standard Polish into Kurpian.
In spite of its huge productivity, Liquid Lowering has recently accumulated exceptions. The list in (26) is not exhaustive.
Liquid Lowering does not seem to be an active process any longer. An interesting piece of evidence strengthening this conclusion comes from the Old Polish word siła [ɕɨɫa] that underwent Liquid Lowering and was pronounced [ɕeɫa]. In today’s Kurpian the word has two meanings: one is ‘how many’ and the other is ‘strength’. In the meaning ‘how many’, the word is pronounced [ɕeɫa] with [e].Footnote [23] In the meaning ‘strength’, the pronunciation is [ɕɪɫa]. The conclusion that suggests itself is that Liquid Lowering is dead as a rule of Kurpian. However, inspection of verbs shows that this conclusion is overstated.
Liquid Lowering is perfectly productive with verbs, and the evidence is of the strongest type: alternations. In (27) I look at four forms: the infinitive and the past forms: the 3rd person singular and the plural feminine forms marked as ‘she’ and ‘they’ (fem) and the 3rd person plural masculine form marked as ‘they’ (masc). So the glosses for the first line are: ‘live’, ‘she lived’, ‘they lived’ (fem) and ‘they lived’ (masc). The 3rd person singular masculine will be discussed later.
As the data show, 
and 
alternate with é [e] when the suffix begins with a lateral (see footnote 21), which is exactly what Liquid Lowering predicts.
The class of inputs to Liquid Lowering is considerably larger than shown in (27) because it includes all 
such as dzwóńéć ‘ring’ discussed earlier. The complication is that, unlike in (27), the input 
is masked by the action of Infinitive Lowering (18) and Infinitive Tensing (17) changing //ɪ// into é [e], but the //ɪ// still occurs on the surface in the perfect participle 
‘having rung’.
The question is how to reconcile the data in (26) showing that Liquid Lowering is dead with the data in (27) and (28) showing that the rule is active and productive. Inspection of the structure of the ‘exceptions’ in (26) and the undergoers in (27)–(28) shows a systematic difference: in (26) both the input 
and the trigger (
$r$
, ł or
$l$
) are contained in one morpheme. In contrast, the data in (27)–(28) show a morpheme boundary between the input 
and the trigger ł or
$l$
, as in the past feminine forms of zy+ć ‘live’ and dzwóń+é+ć ‘ring’: zé+ł+a ‘she lived’ and dzwóń+é+ł+a ‘she rang’.
The situation just described – the requirement that a morpheme boundary be present in the context – is known as the Derived Environment Condition (DE). The idea played an important role in Lexical Phonology since all cyclic rules were assumed to be sensitive to the DE Condition (Kiparsky Reference Kiparsky, van der Hulst and Smith1982, Booij & Rubach Reference Booij and Rubach1987). One of the interesting claims of Lexical Phonology is that in their historical evolution rules become restricted to DE before they disappear altogether (Rubach Reference Rubach1984). The fate of Liquid Lowering supports this point. In Standard Polish the process has gone its complete course: from birth (rule addition) to death (rule loss). In Kurpian, Liquid Lowering is at the DE stage, so the process is active but carries a DE restriction. Interestingly, once DE is recognized as a condition, Liquid Lowering becomes a totally exceptionless rule.
4 Analysis
This section offers an OT analysis of Liquid Lowering, arguing that Derivational Optimality Theory but not Standard Optimality Theory is able to account for the full range of the data. Liquid Lowering is restated as a constraint.
Liquid Lowering prohibits combinations of 
with [l ɫ] (see footnote 21), as well as with [r]. The context before
$r$
is in fact empty once (30) carries a DE restriction because Kurpian does not have suffixes beginning with
$r$
.
4.1 Underlying tense //
${\rm \small{i}}\,-{\rm \small{i}}$
//
This section constructs an OT grammar of Kurpian with a view to satisfy Liquid Lowering. Since the Kurpian high unrounded vowels are the lax 
, the effort is to exclude the combinations *ɪl, *ɪɫ and 
.The repair is to lower the high vowels and thereby eliminate the prohibited 
before a lateral. At this point the analysis runs into difficulty.
The principle of OT is to execute a minimal repair that is necessary to obey a high-ranked markedness constraint (here: Liquid Lowering). A minimal repair is one that is the least costly in terms of how many faithfulness constraints it violates. In the case at hand, a minimal repair is to change 
to lax [ɛ] rather than to tense [e], because the latter change, 
, would violate additionally IDENT[
$\pm$
tense] since the input vowels are lax while the output vowel is [+tense]. The difficulty is that it is [e] rather than [ɛ] that is the attested surface form. The derivation cannot be repaired by postulating an additional constraint that outlaws [ɛl ɛɫ] and hence enforces the change from [ɛ] to [e]. This repair is not available because Kurpian has many [ɛl ɛɫ] combinations in DE contexts.
The conclusion is that it is the input rather than the derivation that needs to be fixed. The logic of minimal repair dictates that the tense vowel [e] can be obtained if the input itself is a tense vowel, so //i ɨ//
$\rightarrow$
[e] as then IDENT[
$\pm$
tense] is not violated. The bottom line is that it must be //i ɨ//, and not 
, that are the underlying vowels in Kurpian. The phonetically attested 
are derived by Laxing, 
.The interest of this analysis is twofold: first //i ɨ// are abstract vowels that never occur phonetically in Kurpian and, second, it is the theoretical apparatus of OT that discovers these vowels. The scenario just outlined is fleshed out in the remainder of this section.
In preparation for a formal analysis, let us look at constraints that can potentially play a role in the evaluation of candidates.Footnote
[25]
Liquid Lowering violates two faithfulness constraints: IDENT[
$\pm$
high] because a high vowel lowers to a mid vowel and IDENT[
$\pm$
tense] because the inputs are the lax vowels 
and the output is the tense vowel [e]. Further, we need to make sure that 
, for example, do not satisfy Liquid Lowering by changing to [ur]. This change is outlawed by IDENT[
$\pm$
round].
The underlying representation of 
‘they lived’, the past masculine plural form of zy+ć 
‘live’, is 
.Footnote
[26]
As mentioned in footnote 21, the //ɫ// is motivated by the alternation between [l] in 
‘they lived’ (masc) and [w] in 
‘they lived’ (fem). The gender marker for the masculine form is the front vowel //ɪ//, which appears overtly in the surface representation of verbs that have //n// rather than //ɫ// in the past tense, for example, 
[zatsəɲɪ] ‘they began’ (masc).Footnote
[27]
The derivation of the surface representation 
from underlying //ze+ɫ+ɪ// is complicated. As in Standard Polish, Palatalization changes dark
$l$
//ɫ// into soft (palatalized) /lj/. Unlike in Standard Polish, however, a lateral is never soft in the surface representation. This generalization is expressed as Hardening, 
. The output [l] is a hard consonant and hence carries the feature [+back]. This feature spreads to the following vowel causing the retraction of the vowel from 
. Below I summarize these developments and compare the derivation of Kurpian 
and Standard Polish 
. Informally:
The derivation in (32) is not a formal analysis but just a scenario showing where our analysis is going.
Forerunning the analysis, let us look at some differences between Standard OT and Derivational Optimality Theory (Kiparsky Reference Kiparsky1997, Reference Kiparsky2000; Rubach Reference Rubach and Roca1997; Rubach Reference Rubach2000a, Reference Rubachb, Reference Rubach2003a; Bermúdez-Otero Reference Bermúdez-Otero1999; DOTFootnote [28] hereafter) that will become relevant shortly. DOT rejects the Standard OT’s principle of strict parallelism that prohibits derivation. Evaluation proceeds in steps called levels. Kiparsky (Reference Kiparsky2000) proposed three levels: the stem level, the word level and the postlexical level (sentence level). Rubach (Reference Rubach2011b and Reference Rubach2016) has extended this model by adding the clitic level placed between the word level and the postlexical level. The levels constitute miniphonologies: the constraints are the same at all levels, but their ranking may be different. Reranking of constraints between levels must be minimal and requires motivation (Rubach Reference Rubach2000a). The input to Level 1 is the underlying representation, the input to Level 2 is the optimal output from Level 1, the input to Level 3 is the optimal output from Level 2, and the input to Level 4 is the optimal output from Level 3. The domains increase in size as we move down from one level to another: the stem at Level 1, the word at Level 2, the clitic phrase at Level 3, and the sentence at Level 4.
Returning to zé+l+y ‘they lived’ (masc), let us look at the evaluation at Level 1 at which Liquid Lowering is active. We focus on Liquid Lowering and not on Palatalization (PAL). To have an effect, Liquid Lowering must outrank IDENT[
$\pm$
high]. The undesired winner is indicated by the left-pointing hand 
. The icon 
marks the desired winner that has lost in the evaluation.
Candidate (33a) fails on PAL because [ɫ] is hard, i.e. [+back], and [ɪ] is [–back], so the consonant and the vowel do not share the value of [
$\pm$
back].Footnote
[29]
Both (33a) and (33b) offend Liquid Lowering since they contain the prohibited 
. Candidate (33c) has avoided this prohibition by changing 
into [u], but this change violates IDENT[
$\pm$
round] since the input 
is [–round] while the output [u] is [+round]. The same objection holds for candidate (33d). The result of the evaluation is incorrect since (33f), [zeljɪ], is the desired winner at Level 1. The reason for the failure is clear. An OT grammar is constructed to execute a minimal repair (the least costly change) when it is necessary to obey a higher ranked markedness constraint. In the instance at hand, the minimal repair is to flip 
to [ɛ]. This change satisfies Liquid Lowering as there is no high vowel before [l]. Candidate (33f), [zeljɪ] satisfies Liquid Lowering in the same way but additionally violates IDENT[
$\pm$
tense] because the input vowel 
is [–tense] while the output vowel [e] is [+tense], so unlike in 
, the repair is not minimal.
It appears that in order to outlaw the undesired candidate [zɛljɪ] in (33e) and tip the balance in favor of the desired winner [zeljɪ] in (33f), we could add a new constraint prohibiting lax [ɛ] before a liquid: *ɛL. Ranked above IDENT[
$\pm$
tense], *ɛL would eliminate candidate (33e), [zɛljɪ]. This rather heavy-handed solution is not available, however. The reason is that it is precisely [ɛl] that is found in the desired winner of verbs representing different classes. Some examples are the following.
Since, as (34) shows, postulating a constraint that bans el [ɛl] is closed as an option, let us reconsider what it is exactly that Liquid Lowering does. As the name says, it lower vowels, specifically, it lowers 
to mid tense [e], crucially, not to lax [ɛ], even though lowering to [ɛ] would be more natural than lowering to [e]: the inputs 
are lax, so they should lower to lax [ɛ]. Given this logic, lowering to tense [e] would be natural and expected if the input had tense vowels. This is the solution then. The inputs to Liquid Lowering are the tense vowels //i ɨ// and not the lax 
. For this analysis to work, we need to assume that lax 
are not members of the underlying inventory. Rather, Kurpian has underlying //i ɨ//. The lax 
are an effect of a late process that I call Laxing. Schematically:
In terms of OT, Laxing is implemented as a reranking of the segment inventory constraints:
We crucially need //i ɨ// at the level at which Liquid Lowering is active, which is probably Level 1. At a later level, level 2, 3 or 4, Laxing sets in and /i/, /ɨ/ are replaced by [ɪ], 
, respectively. As far as I know, there are no processes in Kurpian that would require the high vowels to be lax, that is, Laxing does not interact with other rules, and hence plays no phonological role. All Laxing does is add ‘phonetic detail’ to the system and hence can apply at the end of phonology, which, given DOT, is Level 4.
Notice that this analysis is available in DOT but not in Standard OT, because it crucially assumes that evaluation is carried out at different levels and not simultaneously in a strictly parallel manner at one level, as Standard OT would have it. To conclude, 
are phonetic and not underlying segments in Kurpian. The corresponding underlying segments are //i ɨ//.
One advantage of this analysis is that the inventory of underlying vowels in Kurpian becomes symmetrical because all high vowels are tense and we do not have unexplained gaps in the system, with lax 
having no tense correspondents and tense //u// having no lax correspondent. The inventory in (1b) repeated in (38a) is now replaced by the inventory in (38b).
Returning to the analysis of zy+ć ‘live – zé+l+y ‘they lived’ (masc), the underlying representations are now different and have tense //i ɨ// rather than lax 
hence we have 
and //zɨ+ɫ+i//, respectively. Similarly, 
such as dzwóń+é+ć ‘ring’ all have underlying //i// and not underlying //ɪ//, so they are better called
$i$
-verbs rather than 
Therefore, the infinitive and the past tense forms are the following.
Now let us return to the sample derivation zé+l+y ‘they lived’ that we evaluated in (33) with no success. Since the inputs are the tense vowels //i ɨ//, the relevant IDENT constraint mandating their preservation in the output is IDENT[
$\pm$
tense], familiar from (33).
The failed evaluation of zé+l+y ‘they lived’ in (33) is now improved in (40). The icon 
marks the desired winner.
The result is correct. As shown by the scenario in (32), there are some further processes that apply in the derivation of zé+l+y. At Level 2, soft /lj/ loses palatalization and is hardened to /l/, /zelji/
$\rightarrow$
/zeli/ (Hardening). Since hard /l/ is [+back], the front vowel /i/ changes into the [+back] vowel [ɨ], /zeli/
$\rightarrow$
[zelɨ] (Retraction). Finally, at Level 4 Laxing exchanges /i ɨ/
for 
, yielding the final outcome 
, which is the attested surface form.
Looking at a larger picture, the analysis that proposes an exchange of segments uses what I will term virement as a theoretical tool. The Kurpian virement takes //i ɨ// to 
. It is a shift rather than a neutralization processFootnote
[30]
as no 
exist prior to virement.
Let us look at Laxing, 
, from a diachronic perspective. Old Polish had the tense vowels [i ɨ] in both surface representations and underlying representations. Kurpian that developed from Old Polish changed the vowel inventory by adding Laxing as a rule. The analysis of Liquid Lowering shows that the newly developed 
do not play a role. Kurpian phonology still works on the now abstract assumption that the vowels are //i ɨ// rather than 
, so the underlying representations of zy+ć ‘live’ and dzwóńéć ‘ring’ are 
and 
, respectively. The Laxing virement, 
, must be pushed off to a later level than the level at which Liquid Lowering takes place. The later level may be the last level (i.e. the postlexical level) because Laxing does not interact with any phonological processes. The placement of Laxing at the postlexical level is consistent with the classic tenet of diachronic generative phonology that rules are added the end of the grammar.
4.2 Ranking paradox
The argument for derivational levels in OT motivated by the virement analysis is strengthened by the analysis of the 3rd person past forms such as zó+ł ‘he lived’ and kośó+ł ‘he mowed’. The claim is that the derivation //zɨ+ɫ//
$\rightarrow$
[zoɫ] and //kɔs+i+ɫ//
$\rightarrow$
[kɔɕoɫ] is insoluble without the intermediate stage /zeɫ/ and /kɔɕeɫ/, which calls for a derivational level. This section motivates the need for an intermediate stage (derivational level) by pointing to a ranking paradox. The following section completes the argument by demonstrating that //zɨ+ɫ//
$\rightarrow$
[zeɫ] and //kɔs+i+ɫ//
$\rightarrow$
[kɔɕeɫ] occur at level 1 and [zeɫ]
$\rightarrow$
[zoɫ] and /kɔɕeɫ/
$\rightarrow$
[kɔɕoɫ] take place at Level 2.Footnote
[31]
The ranking paradox argument developed in this section rests upon the same principle of OT as the argument presented in the preceding section: the tenet of minimal repair. The objective is to take //zɨ+ɫ// to [zoɫ]. The change 
is different from the one discussed in the preceding section that took //ɨ// to /e/, zély (40) ‘they lived’ (masc): 
operates in closed syllables, as in //zɨ+ɫ//
$\rightarrow$
[zoɫ] while 
is found in the remaining contexts. The problem is that the minimal repair eliminating [ɨɫ] that is outlawed by Liquid Lowering is to change //ɨɫ// to [uɫ]. The change is minimal because it violates one constraint: IDENT[
$\pm$
round]. The desired change 
additionally violates IDENT[
$\pm$
high] and hence is judged suboptimal by an OT grammar. An attempt to force 
in one step by postulating an additional constraint that prohibits [uɫ] fails since [uɫ] is a well formed combination in verbs such as zuł //zu+ɫ// = [zuɫ] ‘he chewed’.
The derivation //zɨ+ɫ//
$\rightarrow$
[zoɫ] requires that IDENT[
$\pm$
high] be ranked low as the constraint is violated in the desired winner [zoɫ]. In contrast, //zu+ɫ//
$\rightarrow$
[zuɫ] (no change) calls for IDENT[
$\pm$
high] to be undominated because //u// cannot be permitted to change to [o] in zuł in response to the putative *u
ł constraint. We witness a ranking paradox. The remainder of this section fleshes out this analysis.
Our analysis of Liquid Lowering in Kurpian is incomplete because we have not looked at the forms of the past 3rd.sg.
We see that 
and [ɪ] alternate with tense ó [o]. Actually, we already know that 
come from underlying //i ɨ//, so the data in (42) show the operation of a new rule that takes //i ɨ// to ó [o]. The rule is motivated further by the alternations in
$i$
-verbs, for example:
The presence of //i// in (43) is masked by Infinitive Lowering that takes underlying //i// to 
and further 
, by Palatalization, 
, and Nasal Tensing, ɔ
$\rightarrow$
o. To conclude, the underlying representation of
$i$
-verbs has the verbalizing suffix //i//, so the occurrence of ó [o] in the 3rd.sg forms, must be an effect of a rule taking //i// to [o].
The system that we have in place now generates tense [e] by Liquid Lowering, //i ɨ//
$\rightarrow$
[e], as in //zɨ+ɫ+a//
$\rightarrow$
[zeɫa] ‘she lived’. Consequently, if nothing is done, zó+ł ‘he lived’ will also end up with [e], //zɨ+ɫ//
$\rightarrow$
*[zeɫ], which is incorrect. What we need is a constraint that will ban *[zeɫ] ’he lived’, but not [zeɫa] ‘she lived’. The generalization is that [e] from Liquid Lowering cannot occur in closed syllables; compare zół ‘he lived’ and zéła ‘she lived’: in the masculine form the syllable is closed, [zoɫ], while in the feminine form as well as in the other past tense forms that contain é [e], the syllable is open zéła [ze.ɫa] ‘she lived’, 
‘they lived’ (fem), 
‘they lived’ (masc).Footnote
[32]
This generalization is expressed as a constraint banning é [e] in closed syllables. The constraint is related to Liquid Lowering in the sense that the syllable must be closed by a liquidFootnote
[33]
and the DE restriction is relevant.
We are now ready to look at the evaluation of zó+ł ‘he lived’. For the sake of an argument, I assume Standard OT in (45). Also, I ignore the fact that phonetically the high vowels are lax and dark
$l$
is represented as [w] (see footnote 21).
Notice that IDENT[
$\pm$
round] must be low ranked because it is violated by the optimal candidate [zoɫ]. The result is incorrect and there is no way of repairing the evaluation by ranking the constraints in a different way because the desired winner [zoɫ] has a superset of the violations of the wrong winner *[zuɫ].
The reason for the failure of //ɨɫ//
$\rightarrow$
[oɫ] is the same as the reason for the failure of 
discussed in the preceding section: an OT grammar is wired to make the least costly repair in response to a markedness constraint. The least costly way of eliminating the [ɨɫ] that is prohibited by Liquid Lowering is to change //ɨ// into [u] rather than into [o]. The change //ɨ//
$\rightarrow$
[u] violates one constraint, IDENT[
$\pm$
round], while the change //ɨ// into [o] violates two constraints: IDENT[
$\pm$
round] and IDENT[
$\pm$
high].
An option of introducing a new constraint that bans the wrong winner [zuɫ], that is, a constraint prohibiting *uł, is not available because uł is the attested correct surface representation of many verbs in Kurpian.
Let us assume for the moment the putative constraint *uL)
$_{\unicode[STIX]{x1D70E}}$
prohibiting [u] followed by a liquid in a closed syllable. The evaluation of //zu+ɫ// ‘he chewed’ is as follows.
The evaluation works correctly on the condition that IDENT[
$\pm$
high] is ranked higher than *uL)
$_{\unicode[STIX]{x1D70E}}$
because otherwise candidate (47b), *[zoɫ], would be the winner. The problem is that the ranking IDENT [
$\pm$
high]
$\gg$
*uL)
$_{\unicode[STIX]{x1D70E}}$
delivers the wrong result in the evaluation of zó+ł ‘he lived’ considered earlier in (45).
The result is incorrect because [zoɫ] ‘he lived’ is the attested output. We have identified a ranking paradox: in order to obtain the correct output [zuɫ] ‘he chewed’ in (47) we need IDENT [+high]
$\gg$
*uL)
$_{\unicode[STIX]{x1D70E}}$
, but to get the correct output [zoɫ] in (48), the ranking must be reversed: *uL)
$_{\unicode[STIX]{x1D70E}}$
$\gg$
IDENT [
$\pm$
high]. The paradox is insoluble in Standard OT because [zuɫ] ‘he chewed’ and [zoɫ] ‘he lived’ are a minimal pair exhibiting contrast. The problem is easily solved by DOT, as the following section demonstrates.
4.3 Levels
As noted in Section 4.1, Derivational Optimality Theory (DOT) recognizes four derivational levels; the stem level, the word level, the clitic level, and the postlexical (sentence) level. Even though the assignment of particular affixes to either Level 1 or Level 2 is a language-specific matter, the default assumption in a Slavic language is that level 1, the stem level, contains roots plus suffixes, whereas Level 2, the word level, expands the morphological domain by including prefixes,Footnote [34] so the structure subject to evaluation comprises stems (optimal outputs from Level 1) and prefixes.Footnote [35]
Returning to the example meaning ‘lived’, the Level 1 inputs are the inflected stems:
These verbs admit prefixation, so, at Level 2, the evaluation is broadened by considering prefixed structures containing, for example, the prefix do-:Footnote [36] do+zé+ł+a ‘she lived till’, do+zé+ł+y ‘they lived till’ (fem), do+zé+l+y ‘they lived till’ (masc), and do+zó+ł ‘he lived till’.
The ranking paradox identified in the preceding section is resolved as follows. At Level 1, not only zé+ł+a //zɨ+ɫ+a// ‘she lived’, zé+ł+y //zɨ+ɫ+ɨ// ‘they lived’ (fem) and zé+l+y //zɨ+ɫ+i// ‘they lived’ (masc) but also zół //zɨ+ɫ// ‘he lived’ undergo the change from //ɨ// to /e/, thereby satisfying Liquid Lowering. The [e] derived at Level 1 is the attested surface vowel in zéła ‘she lived’, zéły ‘they lived’ (fem) and zély ‘they lived’ (masc). The masculine singular form /ze+ɫ/ is an intermediate representation. It enters Level 2 as /dɔ+ze+ɫ/ and changes to [dɔzoɫ] in the optimal candidate, an effect of *eL)
$_{\unicode[STIX]{x1D70E}}$
prohibiting [e] in closed syllables (recall (44)). The constraint *eL)
$_{\unicode[STIX]{x1D70E}}$
outranks IDENT[
$\pm$
round] and IDENT[
$\pm$
back], so that /e/ may change into [o].
The analysis appears to be contradicted by forms such as do+zół+ëm ‘I lived till’. The problem is that [o] occurs in an open syllable, [dɔ.zo.ɫəm], so *eL)
$_{\unicode[STIX]{x1D70E}}$
has no force to induce
$e\rightarrow o$
. The problem is apparent. The affixes -ëm (1st person singular) and -eś (2nd person singular) are clitics, which means that they become first available for evaluation at Level 3, the clitic level. The input to Level 3 is the optimal output from Level 2, here /dɔzoɫ/. Nothing needs to be done at Level 3. The resyllabification ensuing after the addition of the clitics -ëm and -eś has no effect on the vowel /o/. The grammar generates the correct surface forms. The details of the analysis are fleshed out below.
At Level 1, the system of the constraints already in place generates all the past tense forms with /e/, exactly as shown in (40). This is the final output for zély ‘they lived’ (masc),Footnote [37] zéł+a ‘she lived’ and zé+ł+y ‘they lived’ (fem). In (50), we look at zéł+a ‘she lived’.
The result is correct. Notice that *eL)
$_{\unicode[STIX]{x1D70E}}$
is inapplicable because none of the candidates has a closed syllable. The masculine form zó+ł ‘he lived’ has essentially the same evaluation at Level 1 as zé+ł+a ‘she lived’ in (50).
The winner from Level 1, /zeɫ/, is now the input to Level 2 at which *eL)
$_{\unicode[STIX]{x1D70E}}$
is the driver for the eł
$\rightarrow$
oł change. To obtain this result, we need to rerank *eL)
$_{\unicode[STIX]{x1D70E}}$
above IDENT[
$\pm$
round].
Since the Level 2 input, /zeɫ/, has a mid rather than a high vowel, the relevant faithfulness constraint is IDENT [
$\pm$
high]. The evaluation of zó+ł ‘he lived’ from (51) continues at Level 2 in (53).
The result is correct. I conclude that DOT, unlike Standard OT, can successfully account for both the 
pattern and the 
pattern, as seen in zé+ł+a ‘she lived’, zé+ł+y ‘they lived’ (fem), zé+l+y ‘they lived’ (masc) and zó+ł ‘he lived’. The success of DOT is strengthened by the final batch of the data adduced below. In (54), I look at the masculine singular forms of the past tense but I extend the data to include the 1st and the 2nd person forms, so, for example, the glosses for the first verb are ‘live, ‘he lived’, ‘I lived’, and ‘you lived’, respectively.
The same pattern is true for
$i$
-verbs, for example:
The troubling cases are the forms with the endings -ëm (1st person) and -eś (2nd person). Their syllabification follows the universally default pattern that syllabifies CVCV as CV.CV, where the dot marks a syllable boundary: zó.łëm ‘I lived’ and zó.łeś ‘you lived’. Given these syllabifications the question is how it is possible to have ó [o] in zó.łëm and zó.łeś. Since the syllables are open, *eL)
$_{\unicode[STIX]{x1D70E}}$
, which is the driver for deriving ó [o], is mute, so the system of the constraints predicts, incorrectly, that the 1st person and the 2nd person forms should have é [e] rather than ó [o]: *zéłëm ‘I lived’, *zéłeś ‘you lived’, like we have zéła ‘she lived’, zéły ‘they lived’ (fem) and zély ‘they lived’ (masc) in which the syllables are open.
The data in (54)–(55) are incomplete in one important way. It needs to be noted that the endings -ëm and -eś are clitics. As would be expected of clitics, they can move around in the sentence, attaching to parts of speech other than verbs. Consequently, we always have more than one variant of a sentence with a clitic morpheme, as (56) illustrates.
The information in (56) is important for a DOT analysis. As explained earlier, DOT recognizes four levels of evaluation: the stem level, the word level, the clitic level, and the postlexical (sentence) level. The facts of Kurpian fit the DOT model. All that we need to assume is what we have been assuming all the time along, namely, that the do+zy+ć ‘live till’ – do+zó+ł ‘he lived till’ pattern is derived at Level 2. The derivation proceeds as in (51) and (53). The morphemes -ëm and -eś are not processed at Level 2 precisely because they are clitics. The 3rd person form has no ending, so the Level 2 derivation is the final stage. In the 1st person and the 2nd person, the clitic endings -ëm and -eś are first available at Level 3. Therefore, the inputs to Level 3 are /dɔ zoɫ+əm/ and /dɔ zoɫ+ɛɕ/. The faithfulness constraints guarantee that the input fares through Level 3 unscathed and is the optimal output. For instance, the input /dɔzoɫ+əm/ surfaces as [dɔzoɫəm], as (57) shows.
This scenario is exactly what DOT has been devised to handle. Standard OT with its strict parallelism prohibiting derivational levels cannot derive the zółëm, zółeś pattern.
5 Conclusion
The analysis of Liquid Lowering presented in this article provides supportive evidence for DOT and shows that Standard OT is unable to handle the relevant data. Even though DOT allows for four levels, this analysis requires only three levels. Levels 1 and 2 solve the ranking paradox that arises in the derivation of the z+yć ‘live’ – zó+ł ‘he lived’ pattern. Level 3 solves the contradictions involving clitic morphemes.
A diachronic look at Liquid Lowering shows that the process, once entirely productive, has become extinct in Standard Polish. In Kurpian, on the other hand, Liquid Lowering is still fully productive but has changed its status from a totally unrestricted rule to a rule that is constrained by DE (derived environments).
Optimality Theory makes two interesting predictions in the analysis of Liquid Lowering: first, a derivational step and, second, abstract underlying vowels. The derivational step (level) is required in the mapping of //ɨ// to [o], as in //zɨɫ//
$\rightarrow$
[zoɫ] ‘he lived’ discussed in Section 4.2. The //ɨ// cannot go to [o] directly because of the Optimality Theory’s principle of minimal repair. The minimal repair here is for //ɨ// to change to [u] rather to [o]. The problem is resolved in Derivational Optimality Theory because DOT admits derivational levels.
The principle of minimal repair leads to the discovery of underlying tense //i ɨ// that are postulated in lieu of lax 
. This is noteworthy because [i ɨ] do not occur phonetically in Kurpian and hence are abstract vowels. A process that I call Laxing turns the abstract //i ɨ// into the phonetically attested 
. There are two points worth noting in connection with Laxing. First, Laxing is a virement shift that exchanges segments context-freely, creating new contrasts in surface representations. Second, Laxing is a Kurpian development because the original vowels in Old Polish were tense [i ɨ]. Kurpian phonology ignores the outcome of Laxing and works on the assumption that the high vowels are tense, that is, //i ɨ//. Laxing does not interact with any rules and hence need not be activated before Level 4. This fact is in keeping with the well-known tenet of classic generative phonology that when new rules come into being, they are added at the end of the grammar.
.
in
‘drink’ sounds like the English i [
in
‘write’, where -a is a verbalizing morpheme and -ć is the infinitive suffix.
and it occurs in the surface representation of both zéł+y
‘they lived’ (
‘they began’ (
only. The derivation //kɔs+i+ɫ// 
