[This paper was originally published in 1981, in Texas Linguistic Forum 18:223-255, Papers in honor of Fernando Horcasitas. This version for the Internet, prepared in 2003, contains, at least in intent, the original text unchanged except for typographical adjustments, error corrections, and a few bibliographical updates. A more detailed discussion of one of the central cases of this paper may be found in Tuggy (1997). This paper should be viewed using a well-populated Unicode font for the cited data. Otherwise words like čīčƛi will look even more weird than they are supposed to.]
There are a number of phonological alternations in Classical Nahuatl (CN) and Tetelcingo Nahuatl (TN) which can be accounted for by a phonological rule of Epenthesis, which would insert an i to break up impermissible consonant clusters., In this paper I wish to show that such a rule can be motivated, but also to suggest that using it alone to account for the alternations is probably wrong. Most of the alternations can also be reasonably accounted for by rules deleting i’s, and historically in certain cases i’s have persisted when the environment for Epenthesis has been destroyed, or have started to surface where it never existed. I suggest that the proper way to treat these problems is not to abandon Epenthesis but to claim that multiple analyses are available to speakers, including competing rule analyses and treatment of alternations as suppletive even when they can also be accounted for by rule.
1. Motivation for Epenthesis
1.1. Subject Markers
In both CN and TN, the subject markers ‘I’ n-/ni-, ’we’ t-/ti-, ‘you sg.‘ t-/ti- and ‘you imperative’ š-/ši-, each have two forms: one which consists of a consonant and another which consists of a consonant plus i. The first occurs before vowels and the second before consonants. These alternants are exemplified in the forms in (1).
Subject markers are, as a rule, the first element in the verb (or noun or adjective)
in which they occur.
The following analyses for these alternations are possible under traditional generative phonology:
(i) They are suppletive. Each prefix has both allomorphs listed in the lexicon
with a specification as to its occurrence before a consonant or a vowel.
Both forms of each prefix exist at the level of underlying forms.
(ii) They are accounted for by a rule of Deletion such as the one given below.
Under traditional assumptions, option (i) would be ruled out, because it misses the generalization that all four prefixes are patterning alike. From these data alone a choice between (ii) and (iii) is not possible. Both Epenthesis and Deletion rules are widespread among the world’s languages; most commonly Epenthesis rules will also affect CCC and CC# clusters. Both rules have clear natural motivation; Deletion avoids VV sequences, which are harder to pronounce or perceive than CV sequences, and Epenthesis avoids hard-to-pronounce (and hard-to-perceive) tautosyllabic CC clusters. Thus neither rule is a priori preferable on grounds of universal availability or naturalness. There is an alternation pattern involving o’s deleting prevocalically in reflexive markers (exemplified below in (4)), which could be accounted for by the Deletion rule. (e and a systematically do not delete prevocally; it is for this reason that the rule above only refers to high vowels.) However, there are exceptions to Deletion: it predicts that there should be no VV sequences, but such sequences (including iV sequences) do exist in both CN and TN. Epenthesis, however, has no clear cases of such exceptions that I know of; there are no instances of #CC (or of CCC or CC#) in either CN or TN. These facts could lead one to prefer an Epenthesis analysis for these forms.
1.2 The 3rd Person Singular Object Marker
The 3rd person sg. object marker in both CN and TN is k-/ki-. k- occurs preceding a vowel, or following a vowel when itself followed by a CV. Otherwise (i.e. initial preceding a consonant, preceded and followed by consonants, or preceding two consonants), ki- is used. These patterns are illustrated in (2).
As the forms for ‘he reaches it’ and ‘he wants it’ show, this morpheme patterns like the subject markers discussed in 1.1 when initial. These alternants could be accounted for by either Deletion or Epenthesis; Deletion taking ki- as basic and Epenthesis taking k-. We could also account for the ‘I reach it’ form by deriving it through Deletion, and a Deletion solution would correctly predict that no vowel would be deleted in the ’you pl. want it’ and ‘I see it’ forms. However, a Deletion solution would not be able to account for forms like ’I want it’, where k- is used before a consonant. Epenthesis, however, can, if properly adjusted, account for all the forms. No Epenthesis takes place in ‘he reaches it’ nor after the k- in ‘I reach it’, nor (crucially) in ‘I want it’, because no improper consonant clusters obtain. But in ‘he wants it’ Epenthesis would break up an unpronounceable #CC cluster, and in ‘you pl. want it’ and ‘I see it’ it would break up CCC clusters.
Note that in ‘you pl. want it’ and ‘I see it’ the CCC clusters are broken in different places.
In the first case the i is epenthesized between the last two C’s,
and in the second case it is between the first two C’s.
Apparently the Epenthesis is occurring only at morpheme boundaries,
preferring to occur between the last two C’s where possible.
We could write the rule as follows:
Once again, the alternation could in principle be accounted for suppletively. But, under traditional assumptions, such an analysis would be an unwarranted loss of generalization: given that the alternation can be accounted for by the independently motivated rule of Epenthesis, it should be accounted for by that rule. Suppletion and production by rule are an either/or proposition: if forms are produced by rule they are not recorded in the lexicon, and if they are recorded in the lexicon they are not produced by rule.
1.3 CC-initial verbs and i-Deletion
We have already seen an example above with the verb meaning ‘to see’, which I treated as being CC initial. Actually, many analysts have treated such verbs as being iCC initial. The following common forms are relevant.
Ignoring for the moment the form tē-itta, which will be dealt with in section 2, notice that all the other forms can be accounted for by Epenthesis. In the forms for ‘I see it/him’, ‘he sees me’, ‘he sees you’, and ‘ye says it’, Epenthesis introduces an i to break up a CCC cluster. (This i is arbitrarily given as part of the stem in the forms, but actually it is, under this analysis, not part of either morpheme. Note the alternative parsing of ‘I see him/it’ in (2).) In the other forms, Epenthesis does not take place because there is no CCC cluster for it to break up. Notice that under this analysis it is no accident that all these stems have two consonants after the i; it is the only reason the i appears at all.
Alternatively, we could posit that the i is part of the prefixes involved
and try to get rid of it some way.
This seems clearly wrong (except possibly in the case of k-/ki-,
which we already decided was best accounted for by Epenthesis);
it would involve either multiple suppletions for those prefixes,
following the same pattern (clearly losing a generalization),
or positing some sort of rule deleting i after only those prefixes
before a V or one but not two C’s.
Or we could posit that the i is underlyingly part of the stems,
as do many analysts, and do a Deletion of some sort to account for these forms.
The previously posited Deletion would not work, for it removes the first of two contiguous vowels,
and here it would be the disappearance of the second that we would want to account for.
Rather we would need a rule like i-Deletion:
One objection to this analysis is this: it must be viewed as a total accident that only iCC-initial stems undergo i-Deletion. We would expect to find iCV initial stems with a CV-initial allomorph, and the fact that we do not is totally surprising. Note that we cannot even make an ad hoc adjustment to the rule to make it operate / V __ CC: to do so would prohibit its application in the derivation of šošō-wi, destroying its independent motivation.
Further support for an Epenthetic analysis for these verbs can be gotten from the following facts.
As previously mentioned, the reflexive morphemes (mo-,
no-, and to-)
lose their final o before V-initial stems.
This process is illustrated in (4).
Second person honorific forms are regularly formed by second person reflexive markings on a causative or applicative stem, as in the last two examples above. Consider then the following TN forms:
Consider also the CN forms in (6):
Under an i-Deletion analysis we would need to say that the forms for ’you hon. forget it’ and ‘they fight each other’ do not undergo i-Deletion but rather have the o of the reflexive delete instead. This contrasts with forms like ‘you hon. remember it’, o ‘he sees himself’ (in (3) above), where the i deletes but the o does not. The only way to distinguish the cases is by ad hoc lexical markings suppressing one of the rules for one of the forms (e.g. marking il-kāwa and ihkali [- i-Deletion]). Under an Epenthesis solution we can give these verb stems underlying initial i’s, but leave the other verbs like l-nāmiki and htoa as CC-initial. All the allomorphies then fall out by rule: Epenthesis inserts an i in the CCC cluster in ‘I remember it’ and ‘I see him’, but not in ‘I forget it’ nor ‘he fights him’, which have no CCC clusters. The o of mo- does not delete in ‘you hon. remember it’ because it is not prevocalic: it does delete in ‘you hon. remember it’ and ‘they fight each other’, because it is prevocalic there.
Thus, a solution involving Epenthesis is preferable to an i-Deletion solution for these verbs on three counts: (i) i-Deletion has exceptions; Epenthesis does not. (ii) Under i-Deletion it is accidental that only iCC-initial stems show this alternation pattern, whereas under Epenthesis it is predicted. (iii) The i-Deletion solution would predict that all iCC-initial stems should undergo the rule, and it would have to resort to ad hoc lexical markings to account for the forms in (5) and (6). In contrast, the Epenthesis solution would be able to let the rules in question apply generally.
One final fact: occasionally one of these verb stems we are analysing as CC-initial will occur with
In such a case it begins with i.
Examples are CN ihtawi ‘it becomes known’ (Anderson 197:443),
and TN ilpi-tika (tie-dur) ‘it is tied’.
This means that, under the Epenthesis model, i is being inserted before
rather than between an initial CC.
However, this can be predicted to happen whenever there is no morpheme boundary between the C’s.
As before, Epenthesis is occurring only at morpheme boundaries.
We formulate as follows:
1.4 -iƛ ‘absolutive’
Consider the data in (7):
Several possibilities exist for accounting for the i in forms like ‘tree’, ‘leaf’, etc.
(i) The i could be treated as part of the stem (Andrews 1975:151, Sullivan 1976:47). It then must be deleted in all the forms except the absolutive. This cannot be done by a general rule; there are many nouns (exemplified above by ‘flower’) which do not lose their stem-final i. Often it is assumed that the i is deleted by the same rule responsible for the absence of the final a in a large class of nouns represented by ‘mat’ above (Andrews, loc. cit., Sullivan, loc. cit.). That it is not the same thing is indicated by the fact that the a’s are deleted only in possessed singular forms (almost always word final), whereas the i’s must also be deleted in the other forms with post-stem elements other than the absolutive, which is the only suffix consisting of a single C. Thus this analysis must posit a new, minor and relatively unnatural rule to account for these data. Further, when these nouns started to become pluralized in the historical development of TN, there would be no reason for this rule to apply to the new forms: one would predict such incorrect forms as kwawime and šāmi-me. The only advantage to this solution is that it permits us to predict the distribution of the most common allomorphs of the absolutive: -ƛ will be used on V-final and -ƛi on C-final stems.
(ii) The i could be considered part of the absolutive suffix. (Siméon 1902:17 may approximate this analysis; cf. also Sullivan 1976:35.) At first this looks like an undesirable addition of another allomorph to the absolutive morpheme. One could, however, claim that all -ƛ absolutives are underlyingly -iƛ, and that the i goes by the i-Deletion rule discussed (and tentatively rejected) above. This would imply that the basic form of the absolutive shows up only on a relatively quite small subset of nouns, though that is a perfectly acceptable proposal to many phonologists. This solution also gives up the ability to predict the distribution of the allomorphs of the absolutive suffix: -ƛi would occur only after C-final stems and -iƛ would occur after all V-final stems, but -iƛ would also occur after some C-final stems. There would also be no explanation of why -iƛ is the only nominal suffix to show this alternation. One would tend to expect an iCV suffix or two showing the same alternation. However, this is not a strong consideration because the number of suffixes is so small.
(iii) The i could be treated as part of neither the stem nor the suffix.
By the already motivated rule of Epenthesis it would be inserted in the absolutive forms
and not elsewhere, breaking up a final CC cluster.
This would involve a natural adjustment of Epenthesis as previously given to include final C+C
On balance, I would judge that either (ii) or (iii) is preferable to (i) but that (iii) is slightly preferable to (ii), as Epenthesis is more strongly motivated than i-Deletion and keeps the extremely common absolutive ending -ƛ as one of the basic allomorphs, and because it predicts the lack of any iCV suffix with the same allomorphy. (The case for (iii) is strengthened by consideration of the relationship between -ƛ and -ƛi as discussed in 1.7.) To the extent that these judgments are correct, these data give further motivation for Epenthesis.
1.5 -k and -ki ‘adjective’ in TN
there is a pair of endings -k and -ki
(probably ultimately derived historically from participial endings)
which appear on most TN singular and plural adjectives, respectively.
They are illustrated in (8).
There is a small class of adjectives (consisting of the last four above) which shows an alternation involving the presence or absence of an i. Analyses parallel to the three given in the previous section are possible. Again, the Epenthesis solution seems preferable, for much the same reasons. To the extent that this is the case, these data provide further motivation for Epenthesis.
1.6 CC-initial possessed stems
There are a number of cases of alternations involving possessives on noun or postposition stems
which can be accounted for by Epenthesis. Some are given in (9)
The initial i on ‘foot, abs.‘ and ‘victuals, abs.‘ and the medial i in ‘their feet’, ‘their lunch’, ‘in them’, and ‘in the house’ can all be accounted for by Epenthesis, and thus provide further support for it.
These data could also be accounted for by i-Deletion. However, under i-Deletion it would be accidental that on (i)CC-initial noun or postposition stems show these alternations. You would expect to find iCV-initial stems undergoing i-Deletion, but you don’t. Epenthesis, on the other hand, clearly requires CC-initial stems in order to apply.
1.7 Epenthesis in final position
We have so far had Epenthesis ocurring, always at morpheme breaks,
before two initial consonants or between them,
between the first two or between the second two of three medial consonants,
or between two final consonants.
We would expect it to occur also after a final CC,
filling out the pattern.
We would expect to be able to write the rule as follows,
collapsing this final case with the first subpart of the rule.
The underlying forms of ‘sandal, abs.‘ and ‘male, abs.‘ are, on this analysis, CC final; the final i is contributed by Epenthesis.
Another example involves certain CN nouns ending in CCa.
A rule has been alluded to which deletes a
from the end of noun stems when possessed.
This rule affects these CN forms as exemplified in (11).
The final i’s on the forms ‘my jewel’ and ‘my breechcloth’ are best viewed as arising from Epenthesis (cf. Andrews 1975:151). (The alternative would be to posit that the stems underlyingly end in ia, with the i disappearing except where the a does.)
Note that there is a problem here. The rule as written claims that all cases of CC#, including those with C+C#, should receive final Epenthesis, and we are claiming that forms like ‘sandal, abs.‘ and ‘male, abs.‘ do have a C+C# and do receive final Epenthesis. However, in sections 1.4 and 1.5 we claimed that Epenthesis occurred between the consonants in such configurations. Several responses to this dilemma are possible; they all would require some sort of lexical marking on the nouns which take -(i)ƛ (section 1.4). I will here opt for claiming that the forms in sections 1.4 and 1.5 are lexically marked not to undergo the first (applicable) subrule of Epenthesis but only the second. This solution is preferable to the others because it accounts for the close phonological relationship of the allomorphs. Under any analysis utilizing suppletion, the fact that the allomorphs are -ƛ, -ƛi and -iƛ is essentially accidental; they could as easily be -ƛ, -a, and kwo. Also, under suppletive solutions the very existence of the alternations is ad hoc; there would be no reason for there to be more than one allomorph. However, under Epenthesis the existence of underlying Cƛ# sequences motivates the derivation of both the -iƛ and the -ƛi variants for -ƛ; some sort of allomorphy is to be expected, and the two types predicted are those that occur.
1.8 Irregular cases (TN)
There are two irregular cases which lend some support to Epenthesis.
One is that of the TN verb wal-ika (com-be)
’bring’, some of whose forms are listed in (12).
The surprising forms are the ‘they brought it’ forms. Assuming an Epenthetic analysis, the underlying form of all the regular forms will have a k followed by a w. In particular, the expected ‘they brought it’ form, *ki-walika-ki, would be underlyingly k-walika-ki. This k-w would be easily confusable with a kw. Apparently some people so confused it; this is the source of the irregular forms. If the Epenthetic analysis is not followed, the k and the w will be underlyingly separated, and it will not be so easy to see how the confusion would arise.
The other case involves verbs such as lpia ‘tie’, which has a funny reduplicated form mo-ho-lpia (refl-rdp-tie) ‘they get tied up’ (cf. Andrews 1975:119,120). Under an Epenthetic analysis, the underlying form for the unreduplicated mo-lpia ‘it gets tied up’ is simply mo-lpia. Reduplication normally reproduces the first (h)(C)V of the stem, with an h after it, yielding (C)Vh(C)V. In this case there is no stem initial (h)(C)V to reproduce. So it is easy to see why people would cast around for something else to reduplicate, and would latch on to the o, positing, in effect, that the underlying form was mo-olpia. The rest would follow by normal rules. However, if an Epenthetic solution is not followed, the underlying form would be mo-ilpia, and there is no reason why that should not reduplicate normally, giving (by Vowel Deletion) *m-ih-ilpia or perhaps (by i-Deletion) *mo-h-ilpia.
In this section some potentially serious counterexamples will be examined and alternative analyses for them discussed. Most of these counterexamples involve positive exceptions. A negative exception is the ordinary kind; it consists of a case where a rule fails to apply when it should. Positive exceptions are more serious; they consist of cases where a rule should not apply but apparently does so anyway.
2.1 ki- before embedded elements (TN)
Occasionally, with some speakers of TN,
when the 3rd person singular object marker is used before certain V-initial verb stems
whose first mofpheme is an embedded noun or a borrowed (Spanish) verb,
it will surface as ki- instead of the expected k-.
Examples are in (13).
In the unexpected forms Epenthesis is apparently taking place prevocalically, directly against its natural motivation. I do not know if any such thing happened in CN: it has earmarks of an innovation in TN. It could be accounted for in at least three ways. (i) We could posit some sort of consonant on the beginnings of the stems involved, giving us an initial CC into which we can Epenthesize the i. This approach seems clearly wrong. The consonant wouldn’t always be there: i.e. we would be making the several verb stems suppletive instead of the one morpheme ki-. Then an ad hoc rule would be needed to delete the consonant everywhere. (ii) We could posit a strong boundary (effectively a #) before the stem, giving us an initial C# to condition Epenthesis. This seems more plausible: the incorporated nouns and borrowed verbs are somehow intuitively foreign to the verb. However, we would still be positing that the boundary was not always there, even with the same speaker. And we would be positing Epenthesis in yet a new (though not unexpected) context: /#C__#. (iii) We could give in and say that it isn’t Epenthesis after all. The forms above would only be negative exceptions to Vowel Deletion, which is preferable to having positive exceptions to Epenthesis. But that would mean we would have to make the underlying form be ki- and the morpheme would then have to be suppletive.
2.2 tē-, ne-, and ƛa- unspecified object markers
In (3) we saw the form tē-itta ‘he sees someone (CN)/him hon. (TN)’. If the i on itta is to be accounted for by Epenthesis, as we have claimed, this form is anomalous. Again we have an apparent positive exception: Epenthesis is occurring after a vowel, clean contrary to its natural motivation. The somewhat less common ne- ‘unspecified human reflexive/reciprocal’ (CN) or ‘hon. reflexive/reciprocal’ (TN) exhibits the same patterning. These forms may be related etymologically in a sort of backwards way: see Langacker 1976:133. It is a general fact that all these verbs we have analyzed as being CC-initial have an i when in construction with tē- or ne-, though those are the only V-final suffixes with which this happens. Again, there are at least three ways to try to deal with the situation. (i) Claim that tē- and ne- are not really V-final after all: posit some sort of consonant on the end of them. y seems to be the best candidate: tē-itta shows up phonetically in TN as [tyeyÍhta], and ne-esōƛa-lo (hon.refl-vomit-hon) ‘he hon. vomits’ turns up as [neyesuƛálo], and one could claim that the y’s are underlying. The cost would be, apart from poor motivation (the y’s mentioned can be fairly easily accounted for in other ways) that some rule would be needed to get rid of the y’s before C-initial stems (where they would be expected to become š’s in CN or perhaps TN, and h’s in TN) and before stems beginning in back vowels (e.g. ne-ah-ahwa-lo (hon.refl-rdp-scold-hon) they argue loudly’, is pronounced [neahahWálo], not *[neyahahWálo]). (ii) We could put a strong boundary between tē- and ne- and the stem, claiming in effect that tē- and ne- are not really prefixes. This is quite ad hoc: whatever evidence there is suggests that the unspecified object markers are more rather than less tightly bound to the stem than other prefixal elements (they follow the specified subject and object markers and frequently merge with the stem to form a new stem). (iii) We could give in and say that it wasn’t Epenthesis after all. The stems have initial i’s and the allomorphies elsewhere are to be accounted for by i-Deletion. i-Deletion will have to be suppressed after tē- and ne-, but it is better to have negative exceptions to i-Deletion than positive exceptions to Epenthesis.
An interesting apparent further development is that in TN the postposition -htek ‘inside of’, which we analyzed as involving Epenthesis, has a form with the non-human unspecified object marker ƛa-. Rather than the expected ƛa-htek (cf. no-htek ’inside me’), it is ƛa-ihtek ‘on the inside’. Apparently the behavior of the semantically related tē- and ne- is beginning to carry over to ƛa-.
2.3 ne-/na- second person plural subject marker (TN)
In TN the form of the second person plural subject marker shifted as follows: first (apparently) a reduplication changed an- to nan-. Then two overlapping changes occurred: one changing the a to e and the other removing the final n from the prefix. The results are summarized below:
The forms from nan- on down are still attested in TN, but nan- and nen- are limited to the speech of a few elderly. ne- is the most common form. One result of all these changes has been to make the otherwise deviant an- look more like a subject marker. An additional side-effect of this change is that the environment for one case of Epenthesis is destroyed. One would expect Epenthesis to cease to apply in that environment, but it doesn’t. The crucial forms are ones like ne-ki-neki ~ na-ki-neki (you.pl-it-want) ’you all want it’, in which the ki- parallels the semantically similar CN an-ki-neki rather than the phonologically similar ni-k-neki (dial. ne-k-neke) ’I want it’. Again, we have an apparent positive exception. As usual, there are three ways to handle it. (i) We can claim that the form is still underlyingly nan- or nen-), and let that account for the occurrence of Epenthesis. A later rule, totally restricted to this form, would be needed to delete the n. This is not totally implausible, as those forms still exist; however most speakers never use them, and many younger speakers may almost never hear them. They would better be treated as archaic than as underlying forms. (ii) We could try positing a boundary after the prefix again, perhaps replacing the n historically when it deleted. This would be equivalent in many respects to positing some sort of an abstract consonant. Either one seems hopelessly ad hoc. (iii) We could give in, and say that it is not Epenthesis after all. We could claim that the distribution of the allomorphs of the third person sg. object marker was governed suppletively rather than phonologically: k- occurred after n(i)-, t(i)-, t(i)-, and š(i)-, and/or before a vowel (except sometimes, as exemplified in section 2.1), and ki- occurred elsewhere (i.e. following ne(n)- ~ na(n)- except before a vowel, initially, and in the cases described in 2.1). Under this analysis there would be no reason to expect the i to disappear simply because a consonant did.
2.4 Historical changes
As a final consideration, we have seen several cases where words have historically shifted from one class with respect to Epenthesis to another. One case involved CN n-ihtek ‘in me’ which shifted in TN to no-htek, making the initial i Epenthetic (1.6 and footnote #25). This sort of shift was not uncommon: another example is CN m-ihtōtia vs. TN mo-htōtia ‘dance’. Similarly in CN there apparently was a shift in most forms from ihtoa ‘say’ to htoa (Andrews 1975:443) (though perhaps the change went in the opposite direction). Under an Epenthetic analysis, there is no reason to expect any i to suddenly become Epenthetic. The CN forms with stems like ihtek or ihtōtia never had any illegal consonant clusters in them, and therefore had no reason for Epenthesis to apply to them. However, if an i-Deletion analysis is adopted, these changes can be explained as extensions of the rule; forms that were formerly marked not to undergo it lose that marking. Thus an i-Deletion analysis accounts for the data better than an Epenthetic analysis.
However, the change sometimes went the other way. For instance, the TN forms il-kāwa ‘forget’ and l-nāmiki ‘remember’ (presented in (5)) begin with the same morpheme meaning something like ‘memory’ (memory?-leave vs. memory?-meet). In both cases the vowel behaved Epenthetically in CN; the form for ‘remember’ deviated historically. As another example, CN had the form mo-sƛakmēya ‘salivate’, whose TN counterpart is m-isƛakmēya. This looks bad for i-Deletion; one would have to posit that the forms gained rather than lost a marking. It is not much better for Epenthesis, at least if the model is supposed to represent psychological reality. If the underlying form is what is psychologically real and the differing allomorphs are produced blindly by automatic rules (in this case Epenthesis), there is no reason for a stem like l-kāwa or sƛakmēya to suddenly acquire an initial i. The model that handles contradictory changes of this sort best is one that would have the two forms as suppletive, allowing either one to generalize at the expense of the other.
So we are faced with a dilemma. Do we abandon Epenthesis, which accounts for so much so neatly, and take up with a rag-tag band of suppletions and deletions? Or do we keep Epenthesis in spite of the positive exceptions? Or is there some way we can have our cake and eat it too?
Almost all of the above argumentation has depended on a set of crucial assumptions which have underlain most linguistic practice for many years. These assumptions include the following: (A) There is only one right solution for accounting for any set of data. (B) The way to tell that one right solution from all the others is by some sort of simplicity metric. The solution that entails the least complication (whether complication of formal mechanism or of posited mental computation on the part of the speaker) is the right one. These assumptions have been questioned by many analysts in recent years. Hankamer (1977) argues for permitting multiple analyses of syntactic phenomena, and Langacker (to appear) argues that the rule/list antithesis should go. I think both these analysts are correct. Multiple analyses seem necessary to account for certain historical shifts, as well as synchronic facts such as those arising from a shift in progress, and it seems too obvious to need argument that making a generalization (rule) need not involve forgetting the particulars on which that rule is based. It is obvious that linguists come to different analyses of their data and juggle incompatible models around: why should we expect that speakers (who are all potential linguists) do any differently? In the model Langacker has developed (and applied mostly to syntax so far), generalizations are represented as schematic units which abstract away from the differences in individual content units which instantiate them. Both kinds of units coexist in speakers’ minds, and must be listed in a grammar. They will differ in terms of their relative salience; in one case the schematic unit may be very salient and the content unit’s independent existence quite marginal (e.g. the schematic NOUN-s construction is more salient than the non-schematic polytheism-s, and in another the content unit may be salient and the schematic unit marginal (e.g. Lightner’s (1976) schema uniting morphemes within eleven, twelve, lend, and eclipse). And a given content unit can be ranked under more than one schematic unit, even if the specifications for those units are contradictory: a bat may be at once considered to be a mammal and a bird. But to get rid of the content units since they are in some sense and to some extent accounted for by the schematic units is as much an abuse of the simplicity criterion as it would be to get rid of the generalizations because they wouldn’t exist without the actual facts. Similarly to get rid of one schematic unit because all or some of its instantiating content units are also ranked under another schematic unit is an abuse of simplicity.
From this point of view, it seems clear that rule-governed vs. suppletive alternation are not strict alternatives. Individual alternations are content units, and rules are schematic units abstracting away from many such alternations. The different alternations between k- and ki- are content units, many of which can be instantiations of the schematic unit Epenthesis. Explaining them by Epenthesis need not be taken as explaining them away; speakers may well still learn them in their own right. In fact, with such a very common morpheme, they are virtually certain to learn them in their own right. Thus, when a historical shift like that discussed in section 2.3, of nen- changing to ne- and destroying an environment for Epenthesis, takes place, speakers are left with more than one option. If the schematic unit Epenthesis were all-salient and the individual alternations had no independent existence, we would have to predict that the alternation would cease. However, if the individual alternations do exist on their own, there is the possibility of their continuing even though their connection to the schema is severed. I would claim that that is what happened (or is happening) in TN: the alternations of k- and ki- were instances of Epenthesis, but they also existed in their own right, being drilled in over and over again in using the language. When the change came along, forcing Epenthesis to stop sanctioning certain of those alternations, the alternations were so salient that they continued anyway. Doubtless the fact that they helped carry the burden of disambiguation between e.g. ‘I’ and ‘you pl.‘ forms added to their salience, and using them to carry the total burden of disambiguation emphasized them still further. But the net result was that the alternation did not cease when it ceased to be Epenthetic.
The other counterexamples to Epenthesis are, I believe, to be accounted for in similar ways. The fact that ki- surfaces before vowels in a few forms is to be attributed to its existence as an underlying form, perhaps as the underlying form for those cases. This presupposes either a suppletive analysis or a Deletion analysis (or both) for those forms. Cases like tē-itta or ƛa-ihtek where an i surfaces following a vowel, should be accounted for by a suppletive or by an i-Deletion analysis or both. I would claim that both Vowel Deletion and i-Deletion are real rules and function as alternate schemata under which the data may be ranged by speakers of Nahuatl. Different schemata may predominate in different speakers for different cases to differing degrees. Andrews is probably right to speak sometimes of introducing, sometimes of deleting, ‘supportive /i/’. The i in words like kwawiƛ may be correctly assigned simultaneously in varying degrees to Epenthesis, a suppletive stem kwawi, which serves as the underlying form for i-Deletion in other forms, and a suppletive absolutive -iƛ. Also I would claim that the form is learned as a whole and entered as such in the lexicon. The absence of i in kwawme is probably due in differing degrees to the (expected) failure of Epenthesis to apply, the application of a class-specific Deletion rule, and a couple of suppletions (i.e. the suppletive stem kwaw is used, besides which the form is memorized as a whole and entered as such in the lexicon). Andrews and Sullivan are probably right to claim that the absence of the final i in i-kwaw should be linked to the absence of the final a in i-peƛ, though it is doubtless also right to claim that it should be accounted for by the non-application of Epenthesis and, again, by suppletion. The historical shifts of section 2.4 can be accounted for by the suppletiveness of the allomorphs and, in the first case, by unmarking the form for i-Deletion, without denying that once the change has been made it should be accounted for by Epenthesis.
I share with Hankamer mixed feelings, of conviction and almost of horror, at the picture that emerges when we permit such multiple and overlapping explanations of our data:
... in giving up [this] strong (albeit rather covert) claim about the fundamental nature of linguistic knowledge we reduce the explanatory power of our theory from nothing to less than nothing, it seems ... This result will probably cause many linguists, rightly concerned for the preservation of our already pitiful supply of valid means of argumentation, to recoil in horror. Nevertheless, you cannot make a mode of argumentation valid by refusing to contemplate the possibility that the assumptions on which it rests are unsupportable.
(Hankamer 1977:601, 600)
And I think abandoning the old assumptions is the only way to account for what actually happens in language, that when dialects split developments in one dialect will point to one analysis of data in the mother dialect while developments in the other daughter will point to another analysis, that synchronically within any dialect the picture is quite confused, with some evidence pointing to one analysis (e.g. Epenthesis) and other evidence pointing to a logically incompatible one (e.g. i-Deletion), that functional considerations can enter into the picture, enhancing an analysis in some places without forcing that analysis everywhere, that diachronic developments such as that of k- and ki- show alternations attributable to a rule taking on a life of their own. We linguists have for too long been cutting our throats with Occam’s razor, and it is time we stopped.
CN was spoken by the inhabitants of Tenochtitlán (Mexico City) and the surrounding areas in the sixteenth century. Primary sources for it include Molina 1571, Rincón 1595, Carochi 1645, and various collections of texts. TN is a modern dialect spoken by the inhabitants of Tetelcingo, Mpo. Cuautla, Morelos, about 60 miles from the outskirts of Mexico City. Sources for it include Pittman 1948, 1954, 1961; Brewer and Brewer 1962, Brewer 1969, and Tuggy 1979a. For most purposes TN can be treated as a direct descendent of CN. The vocalic system for CN had eight vowels: i, e, a, and o, plus a long version of each. They can be differentiated by the features [high], [front], and [long]; note that o functions as a high vowel. In TN the feature of length changed to something like tenseness, causing the vowels to shift their phonetic values as follows: CN ī > TN i, CN i > TN ɪ, CN ē > TN ie, CN ā > TN ɔa, and CN ō > TN u. Examples are cited in an alphabet reflecting the historical values of these sounds rather than their present phonetic shapes. Similarly, I have sometimes given a posited underlying form for one dialect or the other where the difference is not relevant, in order to avoid citing two forms or obscuring the point at issue. ‘Saltillo’ (h or ʔ in CN, h in TN) is given as h; sometimes (where irrelevant) it is omitted from CN forms, again to avoid citing two forms.
Among previous analysts Carochi (see footnote 8) and Andrews speak of some of these alternations in terms of an epenthesis. Andrews refers repeatedly to ‘supportive /i/’, explaining that ‘any illegal consonant cluster is lifted in a pronounceable sequence by the introduction of an /i/, either (a) before, (b) between, or (c) after the consonants’ (1975:8). But then almost immediately he adds that ‘Under certain conditions, ... supportive /i/ ... is dropped’, leaving one to wonder if he is positing epenthesis or deletion. In the text he usually refers to it as dropped. The burden of this paper is that such ambivalence may well be justified.
Unless specifically given as CN or TN, cited forms are common to both dialects.
The only exceptions to this rule are verbs with the prefix or proclitic ō- (TN o-) ‘antecessive’, which marks past tense verbs, and which in several respects (Andrews 1975:14, 33, 43) does not behave like a real prefix. It would seem reasonable to posit a stronger boundary between it and the rest of the verb, which would act like a word boundary and would therefore be able to condition Epenthesis.
An example is TN ki-soki-aki-(y)a (it-mud-fit-caus) ‘he fits it (tamps it) into the mud’. Verb endings in i-(y)a like that above are very common: it is debatable whether they should be posited to have an underlying y (or some other consonant, perhaps h) or not (Çanger 1980, Vázquez Soto 1977, Tuggy 1979b). If the traditional view is accepted that they do not, then they constitute a large class of exceptions to a rule like Deletion.
Exceptions which are probably to be discounted include a few Spanish loan words and a few fast-speech forms in TN.
Once again, the ō- ‘past’ marker must be excluded. The only other vowels the 3rd pers. sg. object marker ever follows are those of the subject markers given in 1.1.
It is interesting that Carochi (1645:412) accounts for the k-/ki- alternation in almost exactly these terms: ‘Pero en terceras personas de singular y plural; y en la segunda del plural la c. se buelue en qui, cuando el verbo empeçare por consonante, porque la c. no se pudiera sin mucha difficultad pronunciar ... Mal se puede pronunciar cmictia, o ancmictia; y por esto la c. se buelue en qui, en las personas dichas.’ k- is basic, ki- is derived to facilitate pronunciation, since e.g. initial km and medial nkm cannot well be pronounced. Notice that Carochi still lists the forms in which the change takes place: the thesis of this paper is that he is right to do so. See also Andrews 1975:43.
CN also has the form m-ihtoa with this meaning; see Andrews 1975:443.
There is no evidence for any ordering between the cases included in the [second pair of embedded] curly brackets: the two subrules cannot interact. would work as well as .
In CN inanimate nouns were only rarely pluralized.
The final resonants parenthesized in these forms are deleted in TN by a regular rule.
The y is usually not written in CN in this and similar words. This is a special case of a larger problem: y is hard to hear next to i. I would claim that the y surfaces, however, in CN forms with -e ‘possessor of’; e.g. kwēy-e ‘possessor of a/the skirt’ (Andrews 1975:216, Sullivan 1976:34,35). Note that this analysis implies a new rule deleting y word finally instead of changing it to š as expected. This would fit in with the historical development to such a rule in TN.
Molina (1571:154) gives wapal-li as well as wapali-ƛ
In CN they are also deleted before the suffix -e ‘possessor of’ (which doesn’t exist in TN), so perhaps the semantic environment is more important than the phonological one. Or perhaps this is another reflex of Deletion?
Again, there can be no necessary ordering between the subrules joined by the curly brackets: will do as well as .
The fact that sometimes both variants apparently occur on a single stem (e.g. CN wapal-li < wapal-ƛi ~ wapal-iƛ, ƛakēm-iƛ ~ ƛakēn-ƛi ‘clothing’) can be taken as at least weak evidence that speakers were not sure how to form the absolutive of every stem and thus that it was not totally predictable. Also, assuming that -(i)ƛ is used after C-final stems, some degree of functional explanation is available. Except for a few cases with š, mostly related to kaš-iƛ ‘bowl’, the -(i)ƛ absolutive occurs only after resonants, which are the most vowel-like consonants and thus would be a natural class for the ending associated with V-final stems to intrude into. Also, especially in TN, it occurs almost always after consonants whose identity would otherwise be lost through neutralization. Perhaps the largest group of nouns in this class ends in m. In CN m becomes n word finally; in TN it is deleted. In both CN and TN all nasals assimilate to a following consonant. Thus m can be distinguished from n only prevocally. Except for the very limited CN suffix -e ‘possessor of’ or the no-really-suffixal CN -é ‘hey, you’, and -(i)ƛ, all suffixes are C-initial. Thus, were the absolutive ƛi to be assigned to a noun in this class, the final m would lose all evidence of being an m but would become an n. Apparently this sometimes happened historically: both CN and TN have ƛakēm-iƛ ~ ƛakēn-ƛi. In Nahuatl of the state of Guerrero (Brewer 1962) apparently all these nouns have come to take -ƛi; the form for ‘waterpot’ is kon-ƛi and the form for ‘leaf’ is ših-ƛi, instead of kōm-iƛ and šiw-iƛ as in CN and TN, so there is now no form of ‘waterpot’ with an m in it. The evidence is good that these were originally m’s: in Pochutec (Boas 1910[1917?]:14,15) several nouns from this class are given with an m before the absolutive -t (e.g. atom-t ‘louse’, keškem-t ‘huipil’, and kum-t ‘water-pot’; atem-iƛ (TN atim-iƛ, kečkēm-iƛ, and kōm-iƛ respectively for CN and TN). this contrasts with other nouns, from the class taking -ƛi in CN and TN and apparently ending in an n, which in Pochutec end in n (e.g. ¢on < ¢on-t ‘hair’, = CN and TN ¢on-ƛi. -t ‘absolutive’ was deleted finally after alveolars.) Or, again, both CN and TN have īš-pan ‘before’, but Huasteca Nahuatl (Beller and Beller 1979:238) has a form ƛa-īš-pam-iƛ (unspec-eye-on-abs) ‘altar’, which apparently preserves an original m. Similarly, in TN w and y delete word-finally and neutralize with h preconsonantally. (In CN y is expected to neutralize with š in both positions.) Using -ƛ and Epenthesis instead of -ƛi in these cases thus avoids fairly large-scale neutralizations of contrast. However, these are far from predictive explanations. w in CN and l in both CN and TN would not neutralize but only devoice. And not all resonants take -iƛ; one finds minimal or near minimal pairs such as CN kwaw-ƛi ‘eagle’ vs. kwaw-iƛ ’tree’, or wapal-iƛ ‘board’ vs. ƛapal-li ( < ƛapal-ƛi) ‘paint’. And, as pointed out, in Guerrero apparently the forms simply went ahead and neutralized anyway.
Note that it would seem even more anomalous in this case to claim that the i is part of the ending in that the basic form would appear on only 4 forms in the language.
There is a rather large class of adjectives which shows the presence/absence of an a in much the same manner as these. an example is čikāwa-k, čikāh-ki ‘strong’. It might be thought that both classes should be accounted for by the same rule. However, almost all these adjectives are fairly transparently derived from verbs, and this alternation can be accounted for by a rule of stem-formation independently needed for the proper verbal forms.
Other examples of final Epenthesis may include the suffix -wi of e.g. no-oh-wi ‘my road’, the suffix -ki of e.g. CN iw-ki ’such a thing’, and perhaps others. See Andrews 1975:8,150,348.
Among the options would be to adopt an
i-Deletion analysis for the forms in 1.4 and 1.5,
or to declare -ƛi as suppletive for -ƛ.
Using i-Deletion would necessitate a suppletion anyway:
-ƛi cannot be derived from -iƛ,
nor vice versa, without ad hoc rules.
The second expedient doesn’t try to predict the allomorphs either,
and would require rewriting Epenthesis to not collapse the CC_# cases with the C(+)C+__C cases,
The solution adopted in the text predicts all three allomorphs, at the cost of marking exactly the same forms which must be marked in the other solutions, and keeps Epenthesis formally simpler.
Footnote (17) may give some sort of a functional explanation for why this should happen. Notice that kwaw-iƛ ‘tree’ and kwaw-ƛi ‘eagle’ would be neutralized if all CC# Epenthesis were final, as would wēl-ik ‘delicious, sg.‘ and wēl-ki ‘delicious, pl.‘ (TN).
It is certainly relevant to note that no such confusion arose in the ‘he brings it’ form, and that if it had, it would have caused neutralization of that form with the very common kwal-ika (good-be) ’that’s good, OK’.
It is also possible that such a derivation should indeed be posited, with mo-h-ilpia producing mo-h-olpia by an independently needed rule of Vowel Harmony. However, the Harmony rule elsewhere works regressively rather than progressively as it would have to here. Positing Harmony would not argue against Epenthesis in any case; mo-h-ilpia could be produced by having Reduplication insert only the h before CC-initial stems, yielding mo-h-lpia, to which Epenthesis would apply.
Occasionally such forms show up without the i; in CN this was ‘frequent’ with ne- according to Andrews (1975:225,228). For example, CN tē-ihšili-sƛi or tē-hšili-sƛi (unspec.hum-prick-nmlzr) ‘act of stabbing someone’ (Andrews 1975:443), or ne-htō-lli (unspec.refl-say-nmlzr) ’promise, vow’ (Andrews, loc. cit.—Andrews claims this is from a (related) stem ihtoa ‘be civil, volunteer’, which never elsewhere loses the initial i: m-ihtoa, not mo-htoa for ‘he volunteers’!); TN tē-iknīwa or (fast-speechy?) tē-knīwa (hon.possr-siblings) ’his hon. brothers’.
Andrews (1975:243) has ihtek and its more common variant ihtik as having a non-’supportive’ (i.e. non-Epenthetic) i: one would say n-ihtek ‘inside me’. Perhaps ƛa-ihtek in TN is simply a carry-over from a time when the i of ihtek was so analyzed. It still might have retained that form because of its semantic affiliation with tē- and ne-.
In fact, many younger speakers have merged I with e (presumably under the influence of Spanish), with the result that ne- is ambiguous between first person sg. and second pl. when preconsonantal.
Were k- to be used instead of ki- in the ‘you pl.‘ forms, a very large number of very common verbal forms would become ambiguous: ne-k-neke would mean either ‘I want it’ or ‘you all want it’. This is quite certainly involved in why these forms retained the i. But note that such ambiguity is tolerated in rather large doses elsewhere: ti-k-neki in both CN and TN means either ‘you sg. want it’ or ’we want it’, and TN dial. nekase means either ‘I grab it’ or ‘you all grab it’.
These references are incomplete and imperfectly reconstructed. As originally published they were included in a volume-wide list, and were not included in offprints such as the one this edition was copied from.
Andrews, J. Richard. 1975. Introduction to Classical Nahuatl. Austin: University of Texas.
Beller, Richard, and Patricia Beller. 1979. Huasteca Nahuatl.. In Langacker, ed., 1979, pp. 199-306. Dallas: UTA and SIL.
Boas, Franz. 1917. “El dialecto mexicano de Pochulta, Oaxaca.” IJAL 1.9-44.
Brewer, Forrest, y Jean G. Brewer. 1962. Vocabulario mexicano de Tetelcingo. Vocabularios indígenas “Mariano Silva y Aceves” 8. México: Instituto Lingüístico de Verano.
Brewer, Forrest. 1962. Field notes on Guerrero Nahuatl.
Brewer, Forrest. 1969. “Morelos (Tetelcingo) Nahuatl verb stem constructions”. In Dow F. Robinson, ed., 1969, pp. 33-51. Norman, OK: Summer Institute of Linguistics.
Çanger, Una. 1980. Five studies inspired by Nahuatl verbs in -oa. Travaux du Cercle Linguistique de Copenhague Vol. XIX.
Carochi, P. Horacio. 1645. Arte de la lengua mexicana. Edición facsimilar 1983. México: Editorial Innovación.
Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper & Row.
Hankamer, Jorge. 1977. “Multiple analyses.” In Charles Li, ed., Mechanisms of syntactic change. Austin: University of Texas Press.
Harms, Robert T. 1968. Introduction to Phonological Theory. Englewood Cliffs: Prentice-Hall.
Langacker, Ronald W. 1976. Non-distinct arguments in Uto-Aztecan. University of California Publications in Linguistics 82. Berkeley: UC Press.
Langacker, Ronald W., ed. 1979. Modern Aztec grammatical sketches. Studies in Uto-Aztecan grammar, Vol. II. Arlington TX: UTA and SIL.
Langacker, Ronald W. (to appear). (The referenced point is made in many of Langacker’s subsequent writings, of which the most prominent may be 1987 (pp. 29, 42, etc.) Foundations of Cognitive grammar Vol. I Theoretical prerequisites. Stanford: Stanford University Press.)
Lightner, Theodore M. 1976. “A note on McCawley’s review of SPE.” IJAL 41.236-239.
Molina, Alonso de. 1571. Vocabulario en lengua mexicana y castellana. Editado en la casa de Antonio de Spinosa, México.
Rincón, Antonio del. 1595. Arte mexicana. México: Pedro Balli.
Pittman, Richard S. 1948. “Nahuatl honorifics”. IJAL 14:236-39.
Pittman, Richard S. 1954. A grammar of Tetelcingo (Morelos) Nahuatl. Language Dissertation 50 (supplement to Language 30).
Pittman, Richard S. 1961. “The Phonemes of Tetelcingo (Morelos) Nahuatl”. In A William Cameron Townsend en el XXV aniversario del ILV, 643-651. México: Instituto Lingüístico de Verano.
Robinson, Dow F., ed. Aztec Studies I. Norman, OK: SIL.
Siméon, Remi. 1977. Diccionario de la lengua náhuatl. México: Editorial Siglo XXI.
Sullivan, Thelma D. 1976. Compendio de la gramática náhuatl. México: UNAM.
Tuggy, David. 1979a. “Tetelcingo Nahuatl”. In Langacker, ed., 1979, pp. 1-140. Dallas: UTA and SIL.
Tuggy, David. 1979b. “The Phonology of -iya and -owa Verbs in Tetelcingo Aztec.” Unpublished ms. [2003. Electronic (scanned) edition available on this website. URL: www.sil.org/~tuggyd/1979 tetelcingo verb phonology/85i-verbphonology-nhg.htm]
Vázquez Soto, Verónica. 1977. Fonología generativa del náhuatl clásico. México: tesis profesional de la E.N.A.H.
Back to the top
To David Tuggy’s home page