Alternations: An Introduction (and Some Further

Alternations: An Introduction (and Some Further
Explorations) for Conlangers

Author: Doug Ball

MS Date: 07-06-2018

FL Date: 10-01-2018

FL Number: FL-000055-00

Citation: Ball, Doug. 2018. “Alternations: An Introduction

(and Some Further Explorations) for
Conlangers.” FL-000055-00, Fiat Lingua,
. Web. 01 October
2018.

a Creative Commons Attribution-NonCommercial-
NoDerivs 3.0 Unported License.

http://creativecommons.org/licenses/by-nc-nd/3.0/

Fiat Lingua is produced and maintained by the Language Creation Society (LCS). For more information
about the LCS, visit http://www.conlang.org/

Alternations: An Introduction (and Some Further
Explorations) for Conlangers

Doug Ball

Creator of Skerre; Truman State University

July 6, 2018

Abstract

This essay explores the nature of alternations: variations in form across different contexts. In addi-
tion to providing a basic introduction of the phenomena in both English and in other languages, it
considers several frameworks for understanding the behavior of alternations in natural languages.
This essay also offers some recommendations for the creation of alternations in constructed lan-
guages and gives some examples to illustrate these recommendations. It is a revised and expanded
version of a talk given at the 7th Language Creation Conference (July, 2017) in Calgary, Alberta,
Canada.

Introduction

One might think that an optimal system of communication would have a collection of invariant
symbolic entities corresponding to a collection of singular meanings; and, indeed, computer lan-
guages are often designed along the lines of this principle. However, for better or worse, human
languages do not have this property in general and in fact, the idealized correspondence breaks
down in several different ways. The focus of this paper is the variation at the symbolic level: for
spoken languages, this boils down to the fact that there may be different pronunciations (includ-
ing, in the most complex cases, rather different strings of sounds) which correspond to a particular
meaning. As it turns out, such variation – henceforth called alternations – is hardly rare in natural
languages and, in some languages, the amount of alternation can be surprisingly high or surpris-
ingly complex.

As an apparently common property of natural languages, the inclusion of alternations in con-
structed languages (henceforth conlangs) would seem to be desirable, especially if a language
creator (henceforth conlanger) intends to create a naturalistic language. However easy it might be
to include some variation in forms in a given conlang, creating realistic alternations is on the trick-
ier side of language creation (henceforth conlanging), as this appears to not be one of the more
transparent aspects of linguistic structure. Thus, in the tradition of articles in model railroading
magazines that look to help model railroaders reﬁne their technique, this article looks to provide
some aid to conlangers in improving their technique with respect to alternations. Alternations, of
the kind that we will focus on here, lie mostly within the realm of form, generally treated under
the linguistic disciplines of phonology and morphology. So, this paper will pursue a discussion
of alternations with a careful eye to both their phonological and morphological aspects. Thus,
this paper will delve into some details about how linguists view the grammar of sounds systems
(“phonological theory”) and the grammar of word-building systems (“morphological theory”). As
the discussion proceeds, I will consider both synchronic (language considered at a single time) and
diachronic (language considered across time) aspects of phonology and morphology, as both have
something to offer in better understanding how alternations work.

1.1 Caveats About Linguistic Theory and Conlanging

Unquestionably, there is a certain danger in considering particular theoretical frameworks in con-
langing. Conlanging does not, in any existential sense, require any linguistic theory: conlangs
are languages and, if learners can come to learn languages without any explicit linguistic theory, I
believe there is surely nothing that forces a conlanger to employ linguistic theory in the creation of
their languages. On the other hand, there does not seem to be any notable harm in learning about
linguistic theory either, and conlanging with respect to theoretical notions may result in particu-
larly interesting works of art. The fact that I will, in fact, be talking about the interrelation between
linguistic theory and conlanging probably betrays the fact that I personally have found this route
interesting and have used it in my own conlanging (no doubt a function of the fact that I have been
trained as a theoretical linguist and use a fair amount of theoretical linguistics in my “day job” as a
professor of linguistics). However, this paper is written without the thought that a conlanger must
master the material contained herein to create a “true” conlang (whatever that might be).

Theoretical frameworks have two further issues that are worth noting. First, by their very
nature, theoretical frameworks are incomplete models of a particular phenomenon, so they most

likely do not encompass everything there is to know about a particular topic. Second, theoretical
frameworks all involve making particular commitments about what is important and what is not;
this can result in a skewed view of the area of inquiry. To combat these issues, I have been fairly
intentional in steering clear of what I see as irrelevant theoretical issues and from any unnecessary
details of formalization, all while providing references to relevant works from the academic liter-
ature that an interested reader can investigate. I also include discussion of two different (though
perhaps sympathetic) theoretical approaches to alternations, which, if nothing else, should help
lessen the blind spots of any single theoretical framework. Nevertheless, these above issues might
not be completely overcome in the pages that follow.

1.2 A Road Map

This essay covers some signiﬁcant ground. The section immediately after this one introduces alter-
nations in some depth, using both English and non-English examples. The third section examines a
two-way typology of alternations, discussing both the diagnostics that are used to differentiate the
two types as well as some example alternations from English that simultaneously serve to illustrate
how the diagnostics can be applied. The fourth section discusses how one might, in a conlang,
create realistic alternations, of the easier-to-create type from the third section. The ﬁfth section
goes into greater depth on the more difﬁcult-to-create type from the third section, introducing both
a different framework for understanding this type and also considering some of the historical as-
pects to the phenomenon. The sixth section discusses how one might, in conlangs, create realistic
alternations, of the more difﬁcult type, and provides an example from my own chief constructed
language, Skerre. The seventh section concludes the essay.

If you, as the reader, are fairly new to doing phonological and morphological analysis on natural
languages (or constructed languages), this essay might be best consumed in small doses. Never-
theless, I have tried to write this essay in a way accessible to conlangers of all sorts of levels. I do
assume that the reader is able to read transcriptions in the International Phonetic Alphabet (IPA);
many of the examples make use of the IPA, because a full understanding of these phenomena is
difﬁcult without close attention to the precise sounds of particular words.1

Introducing Alternations

Having dealt with the preliminaries, let me turn to the concrete topic of what form-based alterna-
tions actually are.2 A fairly simple, but reasonably accurate, working deﬁnition of an alternation is
that it is variation in the form of a subpart of a word in different contexts. In the classic Structural
approach to linguistics widely practiced in the ﬁrst half of the twentieth century (see, for instance,
Bloomﬁeld 1933), this was understood in terms of different instantiations of either phonemes or
morphemes (or both); however, alternations still exist as a phenomenon regardless if one believes
that phonemes and morphemes are useful theoretical devices or not.

1The transcriptions are of American English pronunciations, as a result of me speaking that dialect.
2The descriptor ‘form-based’ is useful because there are also what are known as diathesis alternations, which
involve variation in how particular participants are linguistically encoded in sentences (as in the differences between
active and passive sentences). This is an interesting topic on its own, but reasonably far aﬁeld from the alternations
under discussion here.

A well-trod example of an alternation (found in numerous introductory linguistics textbooks
[including Fromkin et al. 2014 and Finegan 2015] as well as on the “Alternation” page of Wikipedia
[Wikipedia contributors 2017]) is in the behavior of the “regular” English plural sufﬁx -s (I will
call it the Z Alternation here). In spite of the conventional spelling, there is, in fact, variation
in exactly how this chief English signal of the plural3 is pronounced. There are three possible
pronunciations: [s], [z], or [1z]. (Each of these different forms is known as an allomorph, which,
more broadly, might be deﬁned as a contextual variant of what seems like a single morphological
unit.) Examples of words showing these different allomorphs are given in Table 1 (relevant forms
are in red). Closer examination of this variation reveals that it is not random; rather, patterns occur

Table 1: English Plural Words Illustrating the Z Alternation
[z]

[1z]

[s]

Canucks
cats
ﬁfths
tips

[kh@n2ks] badgers
[khæts]
conlangs
logs
[fIf(T)s]
[thIps]
taxis

>
[bæ
dZô
”
[kh˜Anl˜æNz]
[lAgz]
[thæksiz]

cases
judges
leashes
mirages

[kheIs1z]
>
>
dZ1z]
dZ2
[
[liS1z]
[m@ôAZ1z]

based on the pronunciation qualities of the ﬁnal sound in the base (the non-afﬁxal part of the word),
to which the sufﬁx is immediately adjacent. The three generalizations for the different allomorphs
are as follows:

• [s] occurs after a non-sibilant, voiceless sound

• [1z] occurs after a sibilant4 sound

• [z] occurs after a non-sibilant, voiced sound (≈ everywhere else)

Having the “choice” of allomorph crucially depend on the nature of the neighboring sounds – as
seen above – is quite common with alternations, though it does not occur in every last instance.

At a slightly higher level, it seems that at least two requirements of English prompt the Z
Alternation: (1) if there are the two adjacent consonants, both must have the same voicing property
(in the plural forms, the voicing value is controlled by the voicing of the end of the base) and (2)
two sibilants cannot appear in a row. The voicing requirement seems to be behind why both the
sounds [s] and [z] are involved – a voiceless sound is needed for voiceless-ﬁnal bases and a voiced
sound for voiced-ﬁnal bases. The requirement against two sibilants in a row seems to be behind
why the sound [1] is included in the allomorph [1z]: without the [1], two sibilants would appear in
a row.

A simple lesson that conlangers can take from the Z Alternation is that, if they are to create
alternations, they must pay attention to the environment; that is, the sound-related “neighborhood”
surrounding either key sounds or key morphological units (at key locations, like at the boundary
between a base and an afﬁx) involved in an alternation.

3There is an interesting question of how to treat plural sufﬁxes like -en in oxen, -i in alumni, -a of phenomena,
as well as the internal changes found in mouse–mice and foot–feet. However, since all these patterns noted in this
footnote are clearly of a different etymological source from -s, I leave them aside here.

4Sibilants are all sounds that have a high intensity noise and include hissing and hushing fricatives and affricates:
>
dZ]. Some phonologists characterize this class with the feature [+ strident].

in English, [s, z, S, Z,

>
tS,

The pattern of having multiple afﬁxes for a single morphosyntactic category where the distri-
bution of afﬁxes is based on phonological properties of the base appears to be common in natural
languages. Another language that shows a surprisingly similar kind of variation (to English) in its
plurals is Luiseño, a Uto-Aztecan language indigenous to Southern California. As shown in Table
2,5 the sufﬁx -m occurs with vowel-ﬁnal bases, while -um occurs with consonant-ﬁnal bases (in
this example, the base is identical to the singular form).

Base gloss

Table 2: Luiseño Plurals
plural
singular
base ending in a vowel

‘arrow’
‘house’
‘owl’

huulam
huula
kiicham
kiicha
muutam
muuta
base ending in a consonant
hunwut
’amaay
’awaal

hunwutum
’amaayum
’awaalum

‘bear’
‘child’
‘dog’

In my experience as an instructor in language construction classes, simple alternations where
there are two different afﬁxal forms and the forms are deployed based on whether the relevant
edge of the base has a consonant or vowel (or a particular type of consonant or vowel) – like in
the plurals in English and Luiseño – are fairly easy for novice conlangers to conceive of and, thus,
include in their languages. However, the behavior with alternations can be a bit more complex than
this, as will (hopefully) become apparent as this essay unfolds.

While the above examples from English and Luiseño involved sufﬁxes, alternations – as a
general phenomenon – can be found in both bases and afﬁxes. English, where alternations are
generally not very pervasive, nevertheless exhibits base alternations with some words. Table 3
gives some select English examples; sounds and sound-related phenomena (like stress) alternating
in these forms are noted in red:6

Table 3: Example English Stem Alternations

Base Word

[Elf]
[h@”ôaIzn
]
”
[khip]
[“phleIRoU] Platonic

elves
horizontal
kept

Derived Word
[Elvz]
[hOô@”z˜Antl
]
”
[khEpt]
[phl@”th˜AnIk]

elf
horizon
keep
Plato

Key, in the very least, to spying any alternations is morphology. Additional morphological ma-
terial changes the environment – at both the sound and morphological levels – and this can allow

5The data in Table 2 come from several sources: Kroeber & Grace 1960 (as cited in Grune 1997), Steele 1988,
Steele 1990, and Hyde & Elliot 1994. All forms cited are ‘non-possessed nominatives’; possessed forms and ac-
cusatives, among some others, present some other complications not worth going into here, though the basic general-
ization given in the text appears to hold throughout the nouns of Luiseño.

6Note that Plato∼Platonic and horizon∼horizontal have segmental changes that are probably consequences of

different locations of stress; I have not noted them in red.

further forms to emerge, if they exist in the language. Without even the possibility of additional
morphological material, a form cannot change and there will not be any alternations. Observe
that if English had only the form horizon and not horizontal, there would be no indication that
another form of horizon exists and horizon would always just be horizon. This effect can also be
seen in reverse with the English word mundane: the etymological predecessor in Latin of the En-
glish adjective, mund¯anus, had about 14 different possible forms (though only one stem), while the
current English word has just one form. Any of the “extra” forms of mundane are completely un-
known to modern-day English speakers, unless they have independently learned Latin, and English
mundane is just the single form, mundane. As a consequence of the morphological-sensitivity of
alternations, languages without much morphological complexity very likely will have few-to-no
alternations.7

Introductory texts (or instructors using them) may give the misleading impression that data
such as that from Italian presented in Table 4 (original data provided in Cipollone et al. 1998;
one scholarly discussion of these facts is in Bertinetto & Loporcaro 2005) involves an alternation.
In fact, these sorts of data indicate that the sounds [n] and [N] can be characterized by particular

[tinta]
[tEnda]
[dansa]
[nero]
[dZEnte]
[sapone]

Table 4: Italian nasals

‘dye’ (n.)
‘tent’
‘dance’ (n.)
‘black’
‘people’
‘soap’

[tiNgo]
[tENgo]
[fuNgo]
[bjaNka]
[aNke]
[faNgo]

‘I dye’
‘I keep’
‘mushroom’
‘white’
‘also’
‘mud’

phonotactic (sound sequencing) generalizations. The relevant generalizations could be represented
as in Figure 1.8 But because single bases or collections of related word forms do not allow for

[n]: vowel ∨ coronal consonant

[N]: velar stop

Figure 1: Phonotactics of Italian Nasals

both [n] and [N] (in different contexts), the words in Table 4 do not, strictly speaking, exhibit
an alternation. For Italian to have an alternation involving [n] and [N], there would have to be
a form like [dasan] which had a morphologically-related complex form [dasaNka] (where -[ka]
is some sort of sufﬁx). Generally, to qualify as alternation, there must be multiple forms (with
different, though related sounds) of the same morphological unit in a disjoint distribution across
several (related) words, not just different sounds in a disjoint distribution across (unrelated) words.
(To put it differently, alternations must involve allomorphy; instances of “pure” allophony are not
alternations.)

Nevertheless, given the focus by many theoretical phonologists since about 1990 (and some
before) on modes of analysis that crucially make reference to phonotactic generalizations (or, as

7However, even though languages like Mandarin and Vietnamese do not have much in the way of inﬂectional mor-
phology, that does not mean they have no morphology. The idea that so-called isolating languages have no morphology
might be more a myth than a fact.

8The underscores in Figure 1 serve as an indicator of where the sound before the colon appears in strings of sounds

in the language.

sometimes called, syllable structure), statements about phonotactics are important to make in any
in-depth linguistic description, of either a natural or a constructed language. So I encourage con-
langers to make such statements within their descriptions of their languages, independent of the
presence of any alternations. (Among the beneﬁts of doing so is that phonotactic generalizations
clarify what a possible word is in a given conlang, guiding a conlanger to make uniform words
as they generate new vocabulary.) As will become clear, phonotactic generalizations will play an
important role in several bits of discussion in the following sections.

3 A Classic Division of Alternations

If one takes a closer look at alternations within single languages and across languages, it becomes
apparent that they do not all behave in quite the same fashion. A classical division of alterna-
tions divides them into two types.9 One is known as phonologically-conditioned alternations (also
known as Automatic alternations [so-called because they appear to occur automatically] or phone-
mic alternations). As their name suggests, they are determined entirely by sound environment. The
other is known as morphologically-conditioned alternations (also known as Morphophonological
alternations or morphophonemic alternations). As their name suggests, they are determined, at
least in part, by morphological environment, though they may have phonological aspects, too.

As it turns out, the type of a particular alternation is not obvious from mere superﬁcial inspec-
tion, so diagnostics to probe the type of alternation have been developed. These diagnostics and
their related properties give some indication about the differing behavior of the two kinds of al-
ternations, so they are worth going over, as they could be interpreted (by the conlanger) as design
desiderata. The discussion of these properties over the next two subsections is largely based on the
discussion in Haspelmath & Sims 2010, ch. 10. Each subsection will use a particular alternation
from English to illustrate the properties presented.

3.1 Phonologically-Conditioned Alternations

Phonologically-conditioned alternations have three chief diagnostic properties. To illustrate them,
I will consider the alternation from English commonly known as Flapping. Flapping, which is
especially common in American English varieties (though found in other dialects, too), involves
variation between the sounds [t]10 and [R] (the latter sound known as the tap or ﬂap, hence the name
of the alternation). The [R] only occurs in a particular environment in English: between syllabic
sounds (that is, between vowels and/or consonants that are, exceptionally, behaving like vowels);
for the precise English distribution, there is the further restriction that the post-tap syllabic sound
must be unstressed. In contrast, the sound [t] never occurs in English between syllabic sounds
where the following syllabic sound is unstressed. However, it does occur in several other sound

9This division ﬁrst appeared in one of the ﬁrst linguistic works on alternations: Baudouin de Courtenay 1895.
The division appears to be continued in the works of major Structuralists like Edward Sapir and Leonard Bloomﬁeld.
Though classical generative phonology (as in Chomsky & Halle 1968) did not acknowledge the difference, the dis-
tinction was resurrected by practitioners of Natural Phonology (see Stampe 1979; Donegan & Stampe 2009) and is
commonly assumed by many phonologists (sometimes tacitly) from the 1970s to the present day.

10Actually [d] participates in these alternation, too. An interesting but tangential – for our purposes – property of
Flapping is that it is neutralizing, meaning that Flapping does not require one to make a distinction between /t/ and /d/,
at the location where Flapping occurs.

environments, where, in turn, [R] does not occur. The pair write [“ôaIt] and writer [“ôaIRô
”
Flapping alternation.

] exhibit the

Let me now consider the diagnostic properties and how Flapping exempliﬁes them.

Phonologically-Conditioned Diagnostic Property 1 These alternations are very general in ap-
plication; thus, they occur wherever the alternation’s conditioning environment does. For some
phonologically-conditioned alternations, that means they even occur across words.

This diagnostic property requires consideration of what one could consider the “domain” of the
alternation. When one considers this for Flapping in English, this alternation appears to be quite
general; in fact, so general that it can occur across words. To see that this is so, let us consider the
phrase write a novel. This English expression can be pronounced in a number of different ways, but
one way (probably more characteristic of fast, casual speech) is [“ôaIR@”nAvl
]. In this example, there
”
is a [R] where the student of English might otherwise expect a [t]: at the end of write. Furthermore,
this “morphing” of sound occurs between two syllabic sounds – [aI] and [@] – and the latter is, in
fact, unstressed (in connected speech, the indeﬁnite article a is rarely stressed); thus, the relevant
sounds meet the conditioning environment of this alternation, across words! Further examination
would yield numerous other parallel examples. So Flapping seems to clearly have this property.

Phonologically-Conditioned Diagnostic Property 2 The sound sequence patterns that the phono-
logically-conditioned alternation “creates” are also found in morphologically simple words.

This diagnostic property requires one to compare the phonotactics of relevant words with the alter-
nating sounds and with afﬁxes (morphologically complex words) to relevant words without afﬁxes
(morphologically simple words). There are, thus, several kinds of words to consider. In the case
of Flapping, one can observe the pattern “stressed vowel–tap–syllabic r” in morphologically com-
plex words like writer [“ôaIRô
]. (It is no accident that the pattern of interest includes the tap and
”
the sounds immediately before it and after it – this is precisely the sort of pattern that one needs to
consider for this diagnostic.) One can also observe the same pattern – “stressed vowel–tap–syllabic
r” – in words without any afﬁxes, like litter [“lIRô
]. Identical patterning in both morphologically
”
complex and morphologically simple words11 is exactly what one hopes to ﬁnd, if the alterna-
tion is a phonologically-conditioned one. Again, further examination of English data would yield
numerous other parallel examples. So Flapping seems to have this property, too.

Phonologically-Conditioned Diagnostic Property 3 If the sounds involved in the alternation are
non-contrastive in the language, the alternation is surely phonologically-conditioned.

This diagnostic property involves consideration of how the sounds involved in the alternation gen-
erally behave in the language. In particular, the key question is whether the language uses the
sounds in question to distinguish between different words (such sounds are often considered “con-
trastive”). Operationalizing this further, one could conceive of the relevant question being “Are
there any two words in the language that are distinguished solely by having either the alternating
sound #1 or alternating sound #2 in an identical location?” If not, the sounds are non-contrastive
and a phonologically-conditioned alternation is surely occurring. With Flapping in English, this

11The fact that morphology seems to be irrelevant for the application of this alternation also speaks strongly against
considering it a morphologically-conditioned alternation, as the diagnostic properties of that kind of alternation in
section 3.2 will make clear.

is what an analyst ﬁnds: there are no pairs of words in English that are distinguished solely by
whether they have a [t] or [R] in a given location. Thus, Flapping involves non-contrastive sounds,
and so it clearly tests, by this diagnostic, as phonologically-conditioned.

This third diagnostic property is worded carefully: while a non-contrastive pair of sounds
clearly points to a phonologically-conditioned alternation, ﬁnding the alternating sounds to be
contrastive does not actually point clearly in either direction. Contrastive pairs of sounds occur
with both phonologically-conditioned alternations and morphologically-conditioned alternations
(maybe at higher rates with the latter, but, nevertheless, found with both). So this is just a one-way
implication, not a bidirectional one.

In best case scenarios, all three of the above diagnostic properties will point in the same di-
rection. While there certainly could be cases where these diagnostic properties are not entirely
conclusive – or unable to be applied, due to independent properties of the language – the above
do seem to be quite reliable. While other diagnostic properties have been proposed (including that
phonologically-conditioned alternations involve more articulatorily- or acoustically-related alter-
nating sounds, that phonologically-conditioned alternations are more subject to speech style or
other sociolinguistically-conditioned variation, and that phonologically-conditioned alternations
are more reliably extended to loanwords), the diagnostic properties above seem, to me, to be the
ones that best capture the distinction, are the easiest to apply, or give the clearest results.

3.2 Morphologically-Conditioned Alternations

Morphologically-conditioned alternations also have multiple diagnostic properties. In some cases,
one considers the same kind of question as those considered above for the phonologically-condi-
tioned alternations; one just looks for a different result. Other diagnostics, however, consider rather
different aspects of the alternation. Ideally, then, an analyst would want to consider both types of
diagnostics – those of section 3.1 and of this section – before coming to a conclusion about a given
alternation.

To illustrate the diagnostic properties of morphologically-conditioned alternations, I consider
the English alternation known as Velar Softening. Velar Softening occurs with several sounds, but
I chieﬂy focus the discussion just on the part of the alternation that involves variation between
[k] and [s]. The [s] alternant has a tendency to appear before high front vowels, though there are
some complications with this statement, as the following discussion will reveal. Electric [IlEktôIk]
and electricity [@lEkthôIsIRi] comprise a well-known example pair of related words exhibiting this
alternation.

Let me now turn to the diagnostic properties for morphologically-conditioned alternations and

how Velar Softening exempliﬁes them.

Morphologically-Conditioned Diagnostic Property 1 These alternations are limited in their ap-
plication: they only occur with certain afﬁxes or certain bases; consequently, they only occur in
some morphologically complex words and never in morphologically simple words or across words.

Like the ﬁrst diagnostic property of the phonologically-conditioned alternations, this requires con-
sideration of the “domain” of the alternation. However, contrary to any phonologically-conditioned
alternation, any morphologically-conditioned alternation should occur in a quite limited domain
and, crucially, show some kind of sensitivity to the morphological environment. In the Velar Soft-
ening alternation, the fact that a restricted domain is involved is quite evident.

To ﬁrst get a handle on the domain of Velar Softening, one has to be aware of some of the
afﬁxes that participate in this alternation. As the example of electric∼electricity indicates, the
sufﬁx -ity [IRi] is one. Another sufﬁx participating in this alternation is -ism [Izm
] (cf. critic
”
[khôIRIk]∼criticism [khôIRIsIzm
]); a third is -ist [Ist] (cf. physics [fIzIks] – physicist [fIzIsIst]). (There
”
are some others, but these three are ﬁne for the present purposes.) It is clear that these afﬁxes share
something in common in their form: they all start with the sound [I]. The bases (in the original,
not in the derived form) also share something common in their form: they end with the sound [k].
These facts are key in the next portion of the diagnostic.

Having identiﬁed [I]-initiality as a common thread of afﬁxes participating in Velar Softening,
one might consider other afﬁxes in English with this sound-based property to see how they behave.
One afﬁx presents itself: -ish [IS]. Next, one wants to see how this afﬁx “reacts” with a [k]-ﬁnal
base (to see the behavior in the full Velar Softening environment): for this purpose, the base weak
[wik] will do nicely. If the -ish added to weak to form weakish, the complex word is pronounced
[wikIS] and certainly not *[wisIS], the form it would have if it were undergoing Velar Softening.
Thus, one has to conclude that, interestingly enough, -ish does not participate in Velar Softening.
Furthermore, the weakish behavior (pardon the pun) clearly indicates that there must be some
morphological sensitivity involved with Velar Softening: ostensibly, the same sounds – [kI] –
would be occurring across the base + afﬁx boundary, yet, in the presence of some afﬁxes, Velar
Softening occurs, while, with others, it does not.

As a further testament to Velar Softening’s morphological sensitivity, the phenomenon exhibits
a sensitivity to the kind of base, in addition to the sensitivity to afﬁxes noted in the previous para-
graph. As already noted above, the bound base physic- shows Velar Softening with -ist: physics
[fIzIks]∼ physicist [fIzIsIst]. However, the bound base anarch- does not show Velar Softening: an-
archy [ænAôki]∼anarchist [ænAôkIst]. While spelling might initially obscure the nature of anarch-
example, when one looks purely at the pronunciation, it becomes crystal clear that the [k] is main-
tained in both forms. Anarch- is exceptional – perhaps ironically, given its meaning – but this
exceptionality is precisely expected if Velar Softening is morphologically-conditioned.

Further telling, Velar Softening does not occur across words and, in fact, is quite strange in that
location. Given the phrase look it over, no English speaker would ever come close to producing
it as *[lUs It oUvô
], in spite of the /k/ ostensibly being in the same kind of “sound neighborhood”
”
as it would be in words like electricity and physicist. Instead, the above phrase would be pro-
]. This further shows that this process only occurs in morphologically complex
duced [lUk It oUvô
”
environments, as this ﬁrst diagnostic leads us to expect.

Overall, Velar Softening clearly has this ﬁrst diagnostic property, as the Softening occurs with

a limited set of afﬁxes, with a limited set of bases, and never across words.

Morphologically-Conditioned Diagnostic Property 2 The sound sequence patterns found in the
neighborhood of the alternation are “opaque”; namely, the phonotactic sequence without the al-
teration would have been allowed by the language and so the alternation could not be motivated
by conforming to the language-speciﬁc sequential requirements, and, from that standpoint, is egre-
gious.

This diagnostic goes hand-in-hand with the ﬁrst diagnostic: it supports and conﬁrms the results
from the ﬁrst diagnostic test through approaching the data from a more phonological perspective
(rather than a morphological perspective, as in the ﬁrst diagnostic). The chief task for the analyst
with this diagnostic is to compare three kinds of sound sequencing patterns:

1. the would-be patterns if the base and afﬁx were put together as is,

2. the actual patterns at the relevant base-afﬁx boundary, and

3. the patterns with similar sounds in morphologically simple words.

For an instance of Velar Softening, in, say, electricity, the pattern expected across the base-afﬁx
boundary if electric [IlEktôIk] was put together with -ity [IRi] as is would be […VkI…]. However,
as we know from words like electricity [@lEkthôIsIRi], the actual pattern across this boundary is
[…VsI…]. In morphologically simple words in English – while not plentiful12 – [VkI] is attested,
as shown by the words given in Table 5. Because English allows the sequence [VkI], it is clear

Table 5: Morphologically-Simple […VkI…] in English

achilles
akin
bucket
cricket

[@”kIliz]
[@”kIn]
[“b2kIt]
[“kôIkIt]

that “canonical” phonotactic restrictions within English are not behind the change from [k] to
[s] in this context.
If “canonical” phonotactic restrictions are not behind this change, it is, in
some sense, “egregious”. (Why change if nothing forces the change?) However, it seems that
only morphologically-conditioned alternations have such egregiousness. So the Velar Softening
Alternation also tests to be morphologically-conditioned by this diagnostic as well.

As with the phonologically-conditioned alternation diagnostics, the above diagnostics for mor-
phologically-conditioned alternations should point in the same direction, although it is possible
that not all the diagnostics will be conclusive or even available in a given language or for a given
alternation. Also akin to the phonologically-conditioned alternations, a few more diagnostics have
been proposed for morphologically-conditioned alternations than just the ones I discussed in-depth
above. Among them are the ideas that, in morphologically-conditioned alternations, the alternating
sounds can potentially be articulatorily- or acoustically-distant, that the alternations likely will not
be extended to loanwords, and that the alternations are “absolute” and not subject to speech style
or other sociolinguistically-conditioned variation. However, again, the diagnostics that I went over
in some detail seem, to me, to be the most reliable or applicable to conlanging.

4 Creating Realistic Phonologically-Conditioned Alternations

in Conlangs

Of the two kinds of alternations discussed in section 3, phonologically-conditioned alternations
are clearly easier to create. In this section, I go through an example of how one could create an
phonologically-conditioned alternation in some depth.

To create natural language-like phonologically-conditioned alternations, there are two straight-
forward steps. First, one must ﬁnd (or, more likely, decide upon) an attested sound change that

12Interestingly enough, it seems that the sequence [VkI] is much more common in English in morphologically

complex words – like making and freakish – than in morphologically simple ones.

is conditioned: conditioned meaning that the sound-related environment is crucial for the sound
change to apply. (Historical linguistics textbooks are great resources for seeing what sound changes
are well-attested; see, for example, the discussion in Campbell 2013, 23–39.) Second, one must
apply that sound change to all possible would-be forms in the language. Although these steps are
stated simply enough above, going through an example would be useful.13

Given in Table 6 is a collection of would-be forms for some language, transcribed in the IPA.
It might be useful to think of these as proto-forms (i.e. ancestral words to the current language’s,

Table 6: Would-be Forms
[lanas]
v.
[tosi]
vi.
[sani]
vii.
[wonsa]
viii.

[sina]
[ko:s]
[mu:so]
[kansi]

i.
ii.
iii.
iv.

on an evolve-your-language approach), though these could just as easily be seen as sketch words
whose precise shape is getting ﬂeshed out by the invention process. While ultimately one would
want to attach meanings to all these forms, the actual meanings will not really matter for this
exercise. It sufﬁces to assume that all the would-be forms mean something different from the other
forms present. Within Table 6, the reader will observe there is a preponderance of the sound [s].
This is by design, as the conditioned sound change chosen for this demonstration involves the
sound [s]. In particular, the chosen sound change is the one in (1):

(1)

s > S/

The sound change in (1) is a (particularly common) kind of change known as palatalization.14
(Observe that the [s] is articulated in a more palatal way by becoming [S].) The notation in (1) is
to be interpreted as saying that any “original” [s] that once was next to the sound represented by
[i] will change into ( ≈ “be henceforth pronounced as”) the sound represented by [S]. The sound
change in (1) will affect several of the forms in Table 6, due to the sounds present within them.
By (1), three of the words from Table 6 change, as given in (2) (the other words will stay as they
were):

(2)

i.
iv.
vi.

[sina] → [Sina]
[kansi] → [kanSi]
[tosi] → [toSi]

13This example is inspired by an alternation that existed once upon a time in my constructed language, Skerre.

However, the details are solely created for illustration here.

14This sound change – or a very similar change where the resulting sound is the voiceless alveolo-palatal fricative
[C] – has been proposed for lots of languages or their ancestors. Among the languages it has reported in include
Chinese (Late Middle Chinese to Standard Mandarin) (Baxter, 1992), Japanese (Shibatani, 2009), Korean (Sohn,
1999, 155), the Tibeto-Burman language of northwest India Hakha Lai (Peterson, 2016), the western African language
Hausa (morphologically-restricted only) (Newman, 2009), pre-Nahuatl in Mesoamerica (Campbell, 2013, 21), and the
southern African language Lamba (ﬁrst discussed by Doke 1938; further discussed in several phonology textbooks,
including Odden 2012). Assuredly, the above list is just a subset of the total number of languages that have undergone
this change. Interestingly, a structural prerequisite for this sound change appears to be that the earlier stage must have
only a single sibilant; though, perhaps further investigation would reveal that this is only a strong tendency.

At this point, a particular phonotactic pattern has been established – [S] before [i] – but no alter-
nations exist (yet). To have the latter, we need to a morphological unit (say, a sufﬁx) that contains
an [i] sound that could follow (at least some of) the historical [s] sounds. To choose a path of
very little resistance, let us assume that there is a sufﬁx -[i] that is semantically compatible with
(ii) [ko:s] and (v) [lanas].15 The diagrams in (3) depict what occurs when this sufﬁx is attached to
these forms, and the change in (1) is applied:

(3)

[ko:si] → [ko:Si]
[lanasi] → [lanaSi]

However, if [ko:s] and [lanas] by themselves are also allowed forms in the language (or these
forms also occur before some sufﬁx that does not start with [i]), there will be an alternation: in
some contexts, there will be the forms [ko:S] and [lanaS] and in other contexts, there will be [ko:s]
and [lanas] – precisely the set of circumstances that should hold if we have an alternation.

Further icing on the cake would be for the changes to occur across words. Thus, if the sequence
of words /lanas ike/ was actual pronounced [lanaS ike], there would be no doubt that this was a
phonologically-conditioned alternation, since it could occur across words. Something somewhat
like this happens in English, though the trigger is not the vowel [i], but the glide [j]. The phrase
miss you can be uttered as [mISu], illustrating that palatalization can occur across words in English.
(No doubt, this is a tricky instance to analyze as the trigger of the palatalization – [j] – does not
even appear in actual pronunciation [see Krämer & Urek 2016 for some further discussion about
this case and some other related ones]; nevertheless, on the view that this instance is an actual
palatalization of [s], it does illustrate that the process can occur across words).

5 More on Morphologically-Conditioned Alternations

Before diving into how one could create realistic morphologically-conditioned alternations, it is
worthwhile to explore some further ideas connected to these kinds of alternations, to have a better
background before tackling this tricky task. This section explores some of these ideas. It ﬁrst con-
siders an alternative approach to the theoretical understanding of alternations; namely, the stratal
approach to alternations, which focuses more on the domain of application of a particular alterna-
tion. The discussion then turns to some observations about the nature of alternation behavior over
time.

5.1 The Stratal Approach to Alternations

To some, the basic division of alternations into phonologically-conditioned ones and morpholog-
ically-conditioned seems like a good way of understanding the array of alternations possible in
natural language. To others, though, this division seems inadequate: in some respects, it underdif-
ferentiates the types of alternations and, in other respects, the classic division seems to divide al-
ternations in the wrong ways. People who have made these criticisms of the “classic” approach are
(often) practitioners of what I will call the stratal family of frameworks. These frameworks include

15As is evident from the discussion, this sort of match-up can be done without recourse to speciﬁc meanings, though

having meanings already established for the bases and afﬁxes likely makes the process more straightforward.

constraint-based frameworks like Stratal Optimality Theory (Kiparsky 2000, 2015; Bermúdez-
Otero 2010, 2018) and Cophonologies (Orgun 1996; Inkelas et al. 1997; Inkelas 2014) as well as
their rule-based forebearer, Lexical Phonology and Morphology (Kiparsky 1982; Mohanan 1986).
While there are notable differences between these different stratal approaches, these differences
are not signiﬁcant for our purposes and I will leave discussion of them aside here.

Central to the name “stratal” is that fact that this family of frameworks all recognize strata
(singular: stratum); that is, different “layers” of structure within the grammatical combinations in a
language. These strata amount to particular domains where certain phonological or morphological
generalizations might hold, so, in essence, the stratal approach takes the notion of “domain” that
was so important to some of the alternation-classifying diagnostics and builds the framework on
top of that.

At the bare minimum, it seems that stratal theorists recognize three strata. These are conven-

tionally named and deﬁned as such:

Stem: a minimal collection of form that must or can have further morphological material attached

to it (but may not be a free form)

Word: a minimal collection of form that is inﬂectionally complete (could serve as a free form)

Phrase: a minimal collection of form that could consist of more than one word

Though often presented as central to phonology, the strata are intimately tied with morphology,
as stem and word are long-standing useful notions in the study of word structure. However, it is
crucial for the stratal approach not only that the strata exist but also that they interact with each
other. They are arranged in a “Russian nested doll” fashion (to borrow a metaphor from Bermúdez-
Otero 2018). Because every phrase is made up of words and all (or maybe just almost all) words
are made up of the stems, conditions at the phrase-level apply to everything, whereas conditions at
the other levels apply to successively smaller domains. Thus, a generalized view of the structure is
as in (4):

(4)

[phrase … [word … [stem FORM ] … ] …]

With these three kinds of domains in play, stratal analysts can actually recognize three kinds of
In the stratal mode of analysis, there can be
alternations, instead of just the two from earlier.
stem-level alternations, word-level alternations, and phrase-level alternations. However, it does
seem that the types of alternations – regardless of framework – are highly correlated, so, in fact, all
morphologically-conditioned alternations from the initial framework are viewed stratally as either
stem-level alternations (most likely) or word-level alternations and all phonologically-conditioned
alternations are viewed stratally as either phrase-level alternations or word-level alternations.

The stratal approach might have some beneﬁt – for both theorist and conlanger – in offering a
slightly greater set of alternation possibilities and a sharper way to ﬁgure out the kind of alternation
present via its domain of application. Yet, this is not without complication, as, in many cases in
many languages, the word and the stem are coextensive or close to it – so determining if something
is a stem-level or word-level alternation must be done carefully.16 In the discussion that follows,

16Furthermore, this mode of analysis does lend itself to the thinking that maybe the best treatment of a particular
phenomena is at a level higher or lower, which, if not tied to some empirical observations, has the potential to be
abused.

I will make use of the different domains as they appear to be useful, though for many conlanging
applications, the choice of framework probably will not matter signiﬁcantly.

5.2 Historical Aspects of Alternations

Regardless of whether one adopts the classic phonologically-conditioned vs. morphologically-
conditioned approach, a stratal approach, or some other approach to alternations, it does seem
that there is a fairly decent understanding of at least how alternations change their behavior over
time. This is an important aspect of alternations to understand; furthermore, this understanding is
crucial, I believe, for the creation of realistic alternations in constructed languages.

5.2.1 Common Diachronic Pathways for Alternations

Like has been observed for particular syntactic/semantic changes (under the rubric of what is some-
times called “grammaticalization” see Heine & Kuteva 2002 for a discussion of “grammaticaliza-
tion”; interesting parallels between grammaticalization and “phonologization” are drawn in Hyman
2013), the behavior of alternations over time does follow recurrent progressions. On the “classic”
view, the progression is as in (5):

(5)

Phonologically-conditioned → Morphologically-conditioned

On the stratal view, the progression is the same in nature, but is, of course, domain-sensitive.
Graphically, the picture is as in (6):

(6)

phrase-level → word-level → stem-level

Examples of progression along the lines of (5)/(6) are easy to ﬁnd. As one example, the mu-
tations found throughout the Insular Celtic languages appear to involve a change where particular
(phonologically-conditioned) phrase-level alternations became restricted to at least the word-level
(with some non-trivial restructuring, so that the present-day patterns are all strongly sensitive to
syntactic context), becoming morphologically- (or morphosyntactically-) conditioned.

Another example would be the [f] ∼ [v] Alternation in English, as occurs in the forms elf-
elves, illustrated earlier in Table 3.
In Old English, [f] and [v] had a predictable distribution:
[f] occurred word-peripherally (including word-ﬁnally) and [v] occurred word-medially, between
voiced sounds. Thus, the changes between [f] and [v] between singular and plural (as occurred
in Old English [ælf] ‘elf’ – [ielv@] ‘elves’17) were quite expected, since the [f] ∼ [v] Alternation,
at the time, was a phonologically-conditioned alternation. Of course, the distribution of [f] and
[v] has changed from Old English to Modern English and, relatedly, this particular alternation has
become a morphologically-conditioned stem-level alternation (restricted to a rather speciﬁc set of
stems, at that) in the modern variety.

There are several ideas useful for the conlanger from the above pathway observations. First is
the idea that the same sorts of sound details (that is, the sounds alternating and the environments)
can be used for all the different kinds of alternations; thus, a conlanger does not necessarily need to

17As these examples reveal, the paradigm involving elf was not the same in Old English as it is now, suggesting that

the patterns surrounding this word have been restructured through the years.

come up with a strange sound relationship for a morphologically-conditioned word-level or stem-
level alternation (though, morphologically-conditioned stem-level alternations do seem to be where
the quirkiest of sound relationships occur). Second is the simple fact that many morphologically-
conditioned alternations are just narrowed “old” phonologically-conditioned alternations, suggest-
ing that if a conlanger would like a particular alternation to be morphologically-conditioned, they
just need to make sure that domain (of, likely, some old sound change) is restricted in some fashion.

5.2.2 Other Pathways to Morphologically-Conditioned Alternations?

What might be other historical sources of morphologically-conditioned alternations, aside from
the narrowing process noted in the previous subsection? I am not aware that a survey has been
undertaken, as of yet, that comprehensively documents all the ways that this might occur. But
from my own preliminary informal surveying, it seems that there are other possibilities.

One appears to be borrowing: by borrowing bunches of (related) words with particular form-
related patterns, a language can import a pattern (or more). Especially if the borrowed-in alterna-
tions are rather different in nature from any pre-existing alternations (if those even occur), these
borrowed words can prompt new (and usually morphologically-conditioned) alternations to come
into being. This was surely what happened with English’s Velar Softening (discussed above in
section 3.2): this alternation existed already in French (arising out of a sound change in the Late
Latin to Old French time frame); it was borrowed into English right along with the plethora of
borrowed words from French.

Another phenomenon related to the historical pathways of morphologically-conditioned alter-
nations is analogy. Analogy is a term used for a kind of historical change where one form (or set
of forms) changes to match, in some fashion, a larger pattern. Garrett & Blevins (2009) discuss a
particular instance where analogy appears to have created an unusual alternation in Ancient Greek.
The relevant alternating sounds are the coronals stops, [t] and [d], together with [s]; forms like
pseúdomai ‘I am lied to’ and épseusmai ‘I was lied to’ show the alternation. By well-established
sound change in Greek, [d] (or any other coronal stop) is not expected to appear as [s] before [m].
Yet, this occurs in several forms in the paradigm. Table 7 shows a relevant part of the overall
paradigm of the verb pseúd¯o ‘lie’ – the forms of the perfect middle. This table also provides, for
each form, either the etymology by regular sound change (each earlier form has an asterisk in front
of it) or the contrary-to-fact result by regular sound change (these forms appear in italics). As is

épseusmai

Table 7: Ancient Greek Prefect Middle of pseúd¯o ‘lie’
‘I was lied to’
(cid:30) should be épseudmai

(cid:30) should be epseúdmetha

epseúsmetha

‘we were lied to’

épseusai

‘you (sg.) were lied to’

épseusthe

‘you (pl.) were lied to’

(cid:31)*epseussai ← *epseudsai
‘he/she was lied to

épseustai

(cid:31) *epseudtai

(cid:31) *epseudthe

epseúdatai

‘they were lied to’

evident from Table 7, Greek speakers adopted the ahistorical -pseus- stem for the ﬁrst person forms
épseusmai and epseúsmetha. Garrett & Blevins suggest that the motivation for this was analogy:
instead of just having the [s]-ﬁnal stem before coronal sounds, speakers expanded the range of this
stem so it appeared before all consonants (and through almost all of the perfect middle paradigm).

Once Greek speakers remodeled these forms, it became a quirk of particular morphological forms
that a stem-ﬁnal [d] changed to [s] before [m].

While analogy offers another interesting possibility for how morphologically-conditioned al-
ternations change over time in a given language, how widespread this kind of change is and what
sorts of restrictions it might have (such as whether it ever creates truly new alternations or just
spreads already existing ones to further locations) is currently not well-understood and remains to
be explored in future linguistic research.

6 Creating Realistic Morphologically-conditioned Alternations

in Conlangs

6.1 Basic Recommendations

As I have noted several times already, creating realistic morphologically-conditioned alternations
is more difﬁcult than creating realistic phonologically-conditioned alternations. Nevertheless, sev-
eral points from earlier in this essay can contribute to making (more) realistic morphologically-
conditioned alternations. I detail them here.

First, some pointers emerge from the diagnostics of section 3.2. Principal among these is
that one should restrict the alternation to particular bases or afﬁxes. This is crucial to achieving
the requisite morphological-conditioning. To make the morphological-conditioning even more
apparent, one could have (certain) base + afﬁx combinations undergo the alternation while other
phonologically-similar base + afﬁx combinations stay unchanged. The latter is not absolutely
required for a morphologically-conditioned alternation, but is deﬁnitely not unexpected, given
natural language behavior.

Second, the stratal approach and historical pathways from section 5 together suggest a basic
methodology for creating a morphologically-conditioned alternation: one could select a particular
sound change (like the palatalization exempliﬁed in section 4), but restrict that change to a limited
domain, like just within the word, or just within the stem. This, of course, requires that the con-
langer creating the language carefully consider what counts, phonologically, as a word and as a
stem in their language (probably no two languages do this in exactly the same way). But whether
the conlanger is working speciﬁcally through the diachrony of their language or just is roughly
simulating it, this method gives the conlanger some starting point for how the sound alternation
could behave within a larger system. Of course, as happens with natural languages, additional
sound changes (or morphological changes) could further obscure the original relationship between
the forms – and thus, there is no requirement that a sound change behave in the “present” exactly
as it did when it historically entered the language. Furthermore, if a conlanger does not wish to
have an alternation emerge out of the native vocabulary of their language, an alternation could,
of course, be “imported” through the borrowing of a collection of alternation-ﬁlled words from
another language.

The above works nicely as a checklist of things to consider in making a morphologically-
conditioned alternation. But it would probably be useful to consider an example. Let me, then,
consider an example from Skerre, my own conlang.

6.2 An Example from Skerre

I began working on Skerre in 1994 and it has been my primary conlang since. The linguistic
sophistication of Skerre has strongly correlated with my knowledge of linguistically-interesting
phenomena and theoretical ideas; the language, hopefully unsurprisingly, has been subject to a
number of revisions through the years that incorporate those (in addition to a collection of revi-
sions that one could consider more whimsical). So, the facts detailed below are a relatively recent
inclusion to the language.

Skerre has what I will call proprietive denominal verbs; that is, there are verbs made from
nominal bases which have roughly the meaning ‘having NOUN’ or ‘be NOUN-y’. (The difference
between verbs and adjectives in Skerre is limited, if not non-existent, so I just consider the relevant
words here to be verbs.) In Table 8, I give some examples of proprietive denominal verbs, along
with their related base. The Skerre words are rendered in both romanization (in italics) and in
a fairly narrow IPA transcription (in square brackets). The key sound portion of the proprietive
denominal verbs is highlighted in red in the transcription. As is evident in Table 8, the relevant

Table 8: Proprietive Denominal Verbs in Skerre

Base

Derived form

eher
iniquih
kitsa
nahi
oonaqua
rasi
saan
toyos
woris

[EhER]
[inikwiç]
>
[ki
tsA]
[nAhi]
[o:nAkwA]
[RAsi]
[sA:n]
[tojos]
[woRis]

waateher

[wA:tEhER]
[wA:tinikwiç]
>
[wA:ki
tsA]
[wA:nAhi]

‘(the) cold’
‘belief(s), religion’ waatiniquih
‘point, sharp tip’
‘stomach’
‘strength’
‘tooth’
‘blur’
‘balance
‘shame’

waakitsa
waanahi
waatoonaqua [wA:to:nAkwA]
waarasi
waatsaan
waatoyos
waaworis

[wA:RAsi]
>
tsA:n]
[wA:
[wA:tojos]
[wa:woRis]

‘be cold’
‘be devout, religious’
‘be pointed’
‘be nervous’
‘be intense’
‘teethe’
‘be blurry
‘be physically skilled’
‘be ashamed’

preﬁxal forms for proprietive denominal verbs are waa- or waat-: the latter is found before vowels
and before [s]-initial bases.18 With these two preﬁxal forms, it is evident that an alternation is
occurring here, involving [t] and the absence of a sound. I will call this the T ∼ ∅ Alternation.
Carefully observing the red highlighting in Table 8 for patterns, one can see that no sequences of
>
consonants (I consider [
ts] a single, complex sound here) occur across the afﬁx + base boundary.
This distributional fact would seem to provide some rationale for why the alternation occurs: if
the [t]-ﬁnal allomorph were added to most of the consonant-initial bases, a sequence of consonants
would result and this sort of sequence does not appear to be allowed in this location.

The T ∼ ∅ Alternation displayed in sound change format (that is, in the same format as (1)) is

as in (7):19

18More strictly speaking, with [s]-initial bases, it seems that the afﬁx-ﬁnal [t] coalesces with the base-initial [s] to

form a single affricate.

19It is worth considering the possibility that the additional [t] with the vowel-initial proprietive denominal verb
stems should be treated as an insertion rather than a deletion ((7) encodes the deletion analysis). However, the insertion
analysis is less optimal because [P] independently seems to intervene when two vowels would otherwise come together,
as evident in Table 9; probably most analysts would prefer to just recognize one epenthetic consonant instead of two.
However, even if insertion were the chosen analytical route, it actually would not crucially change the points below,
as there is still a difference in sound sequencing behavior between this particular afﬁx and other phenomena in the

(7)

t > ∅/ Cnon−sibilant

This notation says that the sound [t] should be eliminated (that is, turned into ∅) before a non-
sibilant consonant. While not commonly given as a sound change by historical phonologists,
something like (7) is found as a phonological process (as noted by by Jun 2011) between words
in Basque, between all sorts of units in the West African language Diola Fogny, and (as noted by
Kang 2000) in morphological concatenation before stops in Attic Greek (see also Côté 2000, ch.
1 for discussion of all three of these and some others). So, this kind of sound-based change does
seem to have natlang precedent.

Turning to a different (though ultimately related) phenomenon in Skerre, the language also has
afﬁxes that signal the subject of the clause; they attach to verbs and a few other kinds of words.
Among the relevant subject afﬁx forms are -wo ‘1ST PERSON PLURAL’, -na ‘2ND PERSON SINGU-
LAR’, -ra ‘2ND PERSON PLURAL’, and -ti ‘3RD PERSON PLURAL’.20 Some example forms (again
in Skerre orthography and phonetic transcription; key portions of the complete words appear in
the transcription in blue) are given in Table 9. Contrasting with the behavior shown in Table 8,

Table 9: Subject-Afﬁxed Forms in Skerre

Base

koot
kenat

[ko:t]
[kEnAt]

‘how’
‘build’

sorit
aket

[soRit]
[AkEt]

‘agree’
‘understand’

Complex forms

[ko:tnA]

koot-na
hikenat-na [hikEnAtnA]
hikenat-ra
[hikEnAtRA]
esorit-ti
[PEsoRit:i]
eTaket-ti
[PEPAkEt:i]
eTaket-wo
[PEPAkEtwo]

‘how … you … ’
‘you are building’
‘y’all are building’
‘they will agree’
‘they will understand’
‘we will understand’

in Table 9, a base-ﬁnal [t] is maintained in a complex form, aside from the fact that two [t]s in
a row yields an elongated consonant, [t:] (arguably, maintaining the [t] as well). In particular,
observe that particular sequences of consonants – like [tn], [tR], and [tw] – are found, unaltered, in
the subject-afﬁxed forms, even though these same sequences are apparently avoided in the propri-
etive denominal verbs. (Recall that, in Table 8, ‘be nervous’ was [wA:nAhi] and not *[wA:tnAhi].)
The dissimilar phonotactic patterns – and their related dissimilar treatment of [t] – represent a
signiﬁcant difference between the two kinds of morphological combinations.

Let us consider the T ∼ ∅ Alternation – and the other [t] behavior noted above – through the
lens of my recommendations from section 6.1. One question to consider would be whether the T
∼ ∅ Alternation involves a plausible sound change. As (7) and the discussion around it indicate,
this desideratum is met: this alternation seems like a plausible sound change, at least at some
point in time. A second – and perhaps more important – question to consider is whether the T
∼ ∅ Alternation is morphologically-conditioned. Already from the above discussion, it is clear
that the T ∼ ∅ Alternation does not occur everywhere it possibly could – it does not occur with
subject afﬁxes, for instance – already pointing in the direction of a morphologically-conditioned
alternation. And, in fact, further evidence from elsewhere in Skerre’s morphophonological patterns
also supports this conclusion. While there are no other stop-ﬁnal preﬁxes in the language, there

language.

20In Skerre romanization, all these elements are obligatorily preceded by a dash, as shown in in-body text.

are compounds, some with a ﬁrst component ending in [t]. In compounds, even before sounds that
elsewhere trigger the T ∼ ∅ Alternation, these components behave just as the subject afﬁxes do,
form-wise: the ﬁnal [t] is maintained. This is exempliﬁed with the compound in (8):

(8)

seet

[sE:t]

Components
ree

‘15’

[RE:]

‘4’

Compound
[sE:tRE:]

seet-ree

19’

Thus, the T ∼ ∅ Alternation in Skerre is surely a morphologically-conditioned alternation, occur-
ring in a domain that includes the afﬁx waa(t)- (and possibly only with the afﬁx waa(t)-).

With the conclusion that the [t]-deletion behavior of the T ∼ ∅ Alternation is morphologically-
conditioned [t] behavior, it would seem likely that the [t]-maintenance behavior found in other
morphological constructions is surely the phonologically-conditioned behavior in Skerre. Thinking
about these patterns stratally, waa(t)- might be seen as a stem-to-stem afﬁx (it attaches to stems and
creates new stems21) while the subject afﬁxes are word-to-word afﬁxes (so they create words from
other words22). Therefore, on this view, the deletion of the [t] would be a stem-level alternation,
while the maintenance of the [t] is the [t]-related behavior at the word-level (or perhaps even the
phrase-level). Regardless of approach, it is deﬁnitely the case that, one of two – the [t]-deletion
behavior or the [t]-maintenance behavior – is “opaque” with respect to the other: one would not
think that a language would have this sort of differing behavior with the same sound in the same
sorts of sound-based environments without some non-phonological conditioning behind it.

There also could feasibly be a historical aspect to this differential behavior. It could have been
the case that earlier stages of Skerre did not allow any sequences of consonants (or, maybe, just
relevant sequences) when waa(t)- became an afﬁx, and so the [t]-deletion of (7) was warranted
when the afﬁx “entered” the word. Subsequently, however, the possibilities allowed elsewhere
in the language changed. Or perhaps, the loss of the [t] is a newer change, but this change just
affected elements inside of words (maybe to make them more like other insides of words) and
became restricted that way. I actually did not need to decide on a precise historical pathway to
include it in Skerre; it is sufﬁcient that plausible historical progressions exist that could lead to the
“present-day” state of affairs.

It is true that I could have chosen just one way to deal with [tC] sequences in Skerre, treating
them the same whenever those kinds of sequences arose: plenty of languages do this sort of thing
(one might call this a phonologically-conditioned solution). However, I felt it would add a further
bit of nuance to Skerre to have different behavior with different kinds of afﬁxes, so I opted to
have the behavior detailed above (and have what one might call a morphologically-conditioned
solution).

Overall, I hope that the above discussion has shown how all the various portions of the earlier
sections can be incorporated and considered in a single example. Additionally, I hope that this dis-
cussion gives some insight into the different kinds of structural properties that might be considered
and decided upon when creating a morphologically-conditioned alternation.

21Why might it be a stem-to-stem afﬁx? For one, category-changing morphology (like here where a noun is be-
coming a verb) is often taken to be stem-creating. The phonotactic patterns here also match those found within
morphologically-simple stems in Skerre.

22The fact that subject afﬁxes appear to attach to fully inﬂected words seems to speak for this treatment. The kinds

of phonotactic patterns here also match those that can occur across words.

7 Wrapping Up

So, after all of this, where do we stand? I hope that the preceding pages have illuminated that: (1)
there are changes in form across words called alternations and that they are plentiful in languages,
both natural and constructed; (2) tools are available for understanding how alternations work in
natural languages (both synchronically and diachronically); and (3) that there are ways in which
a conlanger could take ideas about how alternations work to create more realistic instances in
conlangs. Furthermore, I hope to have shown some good ways for creating what (I think) many
conlangers seek: somewhat exceptional behavior – of the kind that languages are notorious for
exhibiting – yet created in a principled way, so it is not randomly thrown in (or so that randomly
thrown in aspects could be retconned into more principled understandings).

There is no doubt still much more to be learned about the nature of alternations: how precisely
they behave across languages, what sorts of further structural interactions they have, and the ﬁne
details of how (and why) they start and progress over time, so, for the interested conlanger, further
research – some already conducted, some still yet to come – could reveal even more interesting
behavior and an even better understanding of how this sort of phenomenon works. So this is hardly
the last word on the matter; perhaps, it is just an invitation for interested conlangers to explore the
area further.

References

Baudouin de Courtenay, Jan. 1895. Versuch einer Theorie phonetischer Alternationen [Attempt at
a theory of phonetic alternations]. Strassburg: Trübner. English translation published in 1972
as part of A Baudouin de Courtenay Anthology: The Beginnings of Structural Linguistics. ed
and trans. by Edward Stankiewicz. Bloomington: Indiana University Press. 144–215.

Baxter, William H. 1992. A handbook of Old Chinese phonology. Berlin: Mouton de Gruyter.
Bermúdez-Otero, Ricardo. 2010. Stratal optimality theory: an overview. [Online; accessed 16-
July-2017]. http://bermudez-otero.com/Stratal_Optimality_Theory.htm.
Bermúdez-Otero, Ricardo. 2018. Stratal phonology. In S.J. Hannahs & Anna R. K. Bosch (eds.),
The Routledge handbook of phonological theory, Abingdon: Routledge. Also available at
http://ling.auf.net/lingbuzz/003118.
Bertinetto, Pier Marco & Michele Loporcaro. 2005.

The sound pattern of Standard Ital-
Journal
ian, as compared with the varieties spoken in Florence, Milan and Rome.
of
doi:https://doi.org/10.1017/
S0025100305002148. Also available at https://www.zora.uzh.ch/id/eprint/
33383/1/S0025100305002148a.pdf.

the International Phonetic Association 35(2). 131–151.

Bloomﬁeld, Leonard. 1933. Language. New York: Henry Holt.
Campbell, Lyle. 2013. Historical linguistics: An introduction. Cambridge, MA: MIT Press 3rd

edn.

Chomsky, Noam & Morris Halle. 1968. Sound pattern of English. New York: Harper and Row.
Cipollone, Nick, Steven Hartman Keiser & Shravan Vasishth (eds.). 1998. Language ﬁles: Mate-
rials for an introduction to language & linguistics. Columbus, OH: Ohio State University Press
7th edn.

Comrie, Bernard (ed.). 2009. The world’s major languages. London and New York: Routledge.

Côté, Marie-Hélène. 2000. Consonant cluster phonotactics: A perceptual approach: Mas-
sachusetts Institute of Technology dissertation. Available as ROA-548, Rutgers Optimality
Archive, http://roa.rutgers.edu/.

Doke, Clement M. 1938. A textbook of Lamba grammar. Johannesburg: Witwaterstrand Press.
Donegan, Patricia & David Stampe. 2009.

Poz-
nan Studies in Contemporary Linguistics 45(1). 1–31.
doi:10.2478/v10010-009-0002-x.
Also available at http://www.ling.hawaii.edu/faculty/donegan/Papers/
2009hypotheses.pdf.

Hypotheses of Natural Phonology.

Finegan, Edward. 2015. Language: Its structure and use. Boston: Cengage 7th edn.
Fromkin, Victoria, Robert Rodman & Nina Hyams. 2014. An introduction to language. Boston:

Cengage 10th edn.

Garrett, Andrew & Juliette Blevins. 2009. Analogical morphophonology. In Kristin Hanson &
Sharon Inkelas (eds.), The nature of the word: Studies in honor of Paul Kiparsky, 527–545.
Cambridge, MA: MIT Press.

Grune, Dick. 1997. A survey of the Uto-Aztecan language Luiseño. [Online; accessed 27-July-

2017]. https://dickgrune.com/NatLang/Summaries/Luiseno.pdf.

Haspelmath, Martin & Andrea D. Sims. 2010. Understanding morphology. London: Hodder 2nd

edn.

Heine, Bernd & Tania Kuteva. 2002. World lexicon of grammaticalization. Cambridge, UK:

Cambridge Unversity Press.

Hyde, Villiana & Eric Elliot. 1994. Yumáyk yumáyk. Berkeley: University of California Press.
Hyman, Larry M. 2013. Enlarging the scope of phonologization. In Alan C.L. Yu (ed.), Origins of

sound change: Approaches to phonologization, 3–28. Oxford: Oxford University Press.

Inkelas, Sharon. 2014. The interplay of morphology and phonology. Oxford: Oxford University

Press.

Inkelas, Sharon, C. Orhan Orgun & Cheryl Zoll. 1997. Implications of lexical exceptions for the
In Iggy Roca (ed.), Constraints and derivations in phonology, 393–418.

nature of grammar.
Oxford: Clarendon Press.

Jun, Jongho. 2011. Positional effects in consonant clusters. In Marc van Oostendorp, Colin Ewen,
Elizabeth Hume & Keren Rice (eds.), The Blackwell companion to phonology, volume 2, 1103–
1123. Chichester: Wiley-Blackwell.

Kang, Yoonjang. 2000. The phonetics and phonology of coronal markedness and unmarkedness:

Massachusetts Institute of Technology dissertation.

Kiparsky, Paul. 1982. Lexical phonology and morphology. In In-Seok Yang (ed.), Linguistics in

the morning calm, 3–91. Seoul: Hanshin.

Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17(2–4). 351–365. doi:10.
1515/tlir.2000.17.2-4.351. Also available at http://web.stanford.edu/~kiparsky/
Papers/vdhulst-ritter.pdf.

Kiparsky, Paul. 2015. Stratal OT: A synopsis and FAQs. In Yuchau E. Hsiao & Lian-Hee Wee
(eds.), Capturing phonological shades within and across languages, 2–44. Newcastle upon
Tyne: Cambridge Scholars Publishing. Also available at http://web.stanford.edu/
~kiparsky/Papers/taipei.2014.pdf.

Krämer, Martin & Olga Urek. 2016. Perspectives on palatalization. Glossa: A Journal of General

Linguistics 1(1). 31, 1–17. doi:10.5334/gjgl.226.

Kroeber, Alfred L. & George William Grace. 1960. The Sparkman grammar of Luiseño. Berkeley:

University of California Press.

Mohanan, K.P. 1986. The theory of Lexical Phonology. Dordrecht: D Reidel Publishing Company.
Newman, Paul. 2009. Hausa and the Chadic languages. In Comrie (2009) 618–634.
Odden, David. 2012. Introducing phonology. Cambridge, UK: Cambridge Unversity Press 2nd

edn.

Orgun, C. Orhan. 1996. Sign-based morphology and phonology: with special attention to Opti-
mality Theory: University of California-Berkeley dissertation. Available as ROA-171, Rutgers
Optimality Archive, http://roa.rutgers.edu/.

Peterson, David A. 2016. Hakha Lai. In Graham Thurgood & Randy J. LaPolla (eds.), The Sino-

Tibetan languages, 258–276. London and New York: Routledge 2nd edn.

Shibatani, Masayoshi. 2009. Japanese. In Comrie (2009) 741–763.
Sohn, Ho-Min. 1999. The Korean language. Cambridge, UK: Cambridge Unversity Press.
Stampe, David. 1979. A dissertation on Natural Phonology. New York: Garland Publishing.
Steele, Susan. 1988. Lexical categories and the Luiseño absolutive.

International Journal of

American Linguistics 54(1). 1–27. doi:10.1086/466073.

Steele, Susan. 1990. Agreement and anti-agreement: A syntax of Luiseño. Boston/London:

Kluwer.

Wikipedia contributors. 2017. Alternation (linguistics).

[Online; accessed 24-January-2018].

https://en.wikipedia.org/wiki/Alternation_(linguistics).