Pico de Orizaba from Atlewaya
    Jeff Burnham &
David Tuggy     

A Spectrographic Analysis of Vowel Length in Rafael Delgado Nahuatl    

This paper was presented at the June 1979 meeting of the Friends of Uto-Aztecan. The present version reproduces the text that was presented at that conference, with a few error corrections, updated phonetic symbols, and other minor changes. Appendices include scans of the (pretty horrible) spectrograms on which the analysis was based. Rafael Delgado Nahuatl is part of the Orizaba Nawatl dialect area (ISO code nlv).

I. Introduction

The purpose of this study is to investigate the phonetic status of vowel length in the Nahuatl language spoken in Rafael Delgado, Veracruz, Mexico (RD). In Nahuatl generally there are the short vowels /i e a o/  and their long counterparts /iː eː aː oː/. Vowel length was clearly present in Classical Nahuatl. The earliest grammars noted the phenomenon but most did not treat it orthographically. An important exception is the grammar of Carochi (1645), in which long and short vowels are distinguished, though not always consistently. A short list of minimal pairs follows his grammar. The relative brevity of this list indicates that vowel length carried a very low functional load in the classical language.

Several modern Aztecan dialects have lost the vowel length contrast altogether. These include Pipil (Campbell 1976), Michoacán Nahual (Robinson 1969:31), Nauzontla (McQuown 1942), and Pochutec (Boas 1911). In the dialect of Tetelcingo, Morelos (Pittman 1961) vowel length has been converted into a difference in quality, with the historically long vowels becoming tense /i ʲe ɔ u/ and the short vowels lax /ɪ e a o/. Knab (1976) claims that in several dialects in the state of Puebla vowel length has either been lost altogether, is only maintained in “deliberate” speech, or carries the concomitant feature of low tone. In the dialect of the Sierra of Puebla the two available sources (Robinson 1966, and Key and Key 1953) give conflicting accounts of vowel length. Robinson (1966:162) noted this, saying, “the marking of long and short vowels does not conform to the marking as shown in Key’s vocabulary. … Although there is an emic contrast between short and long vowels, the factors of stress, of preceding voiced consonants, and of following voiced consonants cause neutralization of the length contrast.”

In RD vowel length seems to be present in the expected handful of minimal pairs but in other words with historically long vowels the length distinction doesn’t always appear to be present phonetically. To complicate matters, elicited minimal pairs may have a clear length contrast, but in normal speech the contrast appears to be neutralized or at least becomes very difficult to perceive. No other feature such as pitch or vowel quality seems to maintain the contrast. Interestingly, investigators in the neighboring dialect of Zongolica (Goller et al. 1974) note that “the length phoneme has been noted infrequently in Orizaba Nahuatl [their term for the dialects spoken to the north and south of Orizaba, Veracruz]. Length does not appear to carry the same functional load that it carries in other Nahuatlan languages. This load may prove to be carried by stress in Orizaba Nahuatl.”

It should be noted here (for what it’s worth) that neither informant in this present study could consistently identify long and short vowels except in minimal pairs. In other cases they would contradict each others opinions or reverse their own previous judgements about vowel length. This suggests that the contrast may not be entirely “psychologically real” for them, and might be used as evidence that the contrast is well down the road to neutralization.

In sum, vowel length appears to be losing ground in Nahuatl in general, and its status in RD is problematical. This study is intended to examine data bearing on the hypothesis that vowel length is still phonemically present in RD but is neutralized (optionally or obligatorily) where it makes no difference to the meaning.

II. General Procedure

The spectrograms that form the basis for this study were made from recordings on magnetic tape made by Jeff Burnham in Orizaba. Among other things he had recorded were minimal and near-minimal pairs representing all the vowel pairs that contrast by length except /e /. These, along with a number of other words that historically had long vowels, were analyzed spectrographically.

The principal informant was Victor Hernández de Jesús, age 41. Supplemental and somewhat sketchy data was provided by Juan Carrasco, age 25. Both men are natives of Rafael Delgado, Veracruz. Both are fully bilingual in Nahuatl and Spanish. Victor also speaks English and gives private classes in the nearby city of Orizaba. He learned English as a teenager and has since become quite fluent through his association with an American who spends a few months each year in Orizaba. Victor’s speech is more rapid than Juan’s, though this is most noticeable in Spanish. Likely as a result of this long vowels are more perceptible in Juan’s speech.

The chosen words from the tapes were re-recorded on the Series 700 Sound Spectrograph made by Voice Identification, Inc., and spectrograms were made of the tokens. Length of the vowels was determined by measuring the length of the vowel formants (including transitions) in millimeters, and multiplying that length by 7.9 milliseconds.

Tokens of two words with putative long vowels were also taken from a recorded narrative and analyzed to determine if the vowel length which is registered in the elicitation forms is maintained in more normal speech. Our hypothesis suggests that it might not be maintained when it does not directly contrast with the opposing member of the minimal or near-minimal set.

We encountered several problems in interpreting the spectrograms. Spectrograms from both speakers (but especially Victor) often showed only faint or completely absent third formants. Since we were primarily interested in vocalic duration and not vowel quality, this wasn’t a crucial problem. When the vowels were preceded or followed by a nasal consonant, /w/, or /l/, the boundaries between consonant and vowel were often difficult to determine.

III. The Data

On the next page is summarized the list of words used and the results from measuring vowel length. The words are listed with long vowels marked where there were long vowels in the classical language. Then, under the speaker’s name, is listed the calculated length of the vowels in the word in milliseconds. Where more than one set of lengths is given there is represented more than one token of the same word. Where more than one form occurred of a single verb, the different forms are listed under one number. The numbers correspond to those marked on the spectrograms, in the appendix to this paper. The traditional digraphs tz and tl are used to represent the phonemes // (ts) and /ƛ/ () respectively.

Word Vowel lengths Gloss
Victor Juan
1. šiwitl 71-40 grass
2. šiːwitl 87-87 comet
3. čičik 103-55 bitter
4. čiːči 134-95 suck
5. šiktlati

Burn it!
6. šiktlaːti
Hide it!
7. kitlahtlani

asks for it
Ask for it!
8. kitlaːni 55-119-87
earns it
9. kalli 79-55 house
10. tlaːlli 79-71 87-47 earth
11. oːniktokak 55-55-55-79
I followed him
12. oːniktoːkak 79-55-79-79
I planted it
13. oːčoːkak

he cried

we cried
14. tekoloːtl 55-110-102 owl
15. tzohpiːloːtl ?-87-79 buzzard
16. tlaːkatl 63-63 man
17. toːnaltzin
18. meːtztli 79-55

IV. Predictable Length

The vowels that appear on the spectrograms range in length from about 16 to approximately 198 msec. It would appear, at least from these data, that there are some positions in RD words where length is predictable to some extent from factors other than historical length. These positions are as follows.

In stressed (penultimate, usually) position, vowels tend to be lengthened, as the following chart of average lengths of (historically) short vowels indicates. The length numbers are in msec., and the number in parentheses indicates the number of examples that were averaged to get the length figure.

Unstressed Stressed % Longer when Stressed
i 47 (38) 87 (2) 47%
a 62 (11) 91 (10) 85%
o 55 (3) 111 (1) 102%
e 48 (2)
Total 51 (54) 92 (12) 80%

As can be noted above, the vowel /i/ tends to be shorter than the others. This tendency is especially noticeable in verbal prefixes. Following /š/ in this position (in the prefix ši- ‘imperative’) /i/ even disappears entirely in 6 out of 9 occurrences.

/i/ Range Average Length
Prefixal 0-63 30 (18)
Other unstressed 16-95 48 (19)
Stressed 71-103 87 (2)

There were no stressed prefixal /i/’s in these data. It is perhaps noteworthy that these prefixal vowels can be analyzed in many Nahuatl dialects as epenthesized. 

A good number of quite short /i/’s were also noted in final position: 8 (of 17) were under 35 msec. long. Many final /i/’s (excluding those on /tlani/ or /tlaːni/, which had the two longest tokens: 87 and 95 msec.) can be analyzed as epenthesized.

There also seems to be some tendency to lengthen a (historically) short vowel when it occurs in the same word as a long vowel, especially in the ultimate syllable following a stressed long vowel. Examples are /šiːwitl/ ‘comet’ (87-87) as contrasted with /šiwitl/ ‘grass’ (71-40) (#’s 1 and 2). Or #3 /čičik/ ‘bitter’ (103-55) as opposed to #4 /čiːči/ ‘suck’ (134-95). But this does not always hold true: the same informant (Victor) gave #5 /štlati/ ‘burn it’ (63-40) and #6 /štlaːti/ ‘hide it’ (87-32), which go against the pattern.

V. The Historically Long Vowels

It does appear from the measurements of vowel length on the spectrograms that the historically long vowels in minimal or near-minimal pairs are appreciably longer than their historically short counterparts. Following are charted the lengths of the long vowels in the elicited pairs and their short partners. In the third column are the differences in msec. between the long and the short vowel. In the last column are the short vowel/long vowel (V/Vː) ratios.

  /i/     //    Difference  V/Vː ratio
1-2 grass/comet 71 87 16 82%
3-4 bitter/suck 103 134 31 77%

  /a/     //    Difference  V/Vː ratio
5-6 burn/hide Victor
(avg. 67)
24 74%
Juan 119/198
(avg. 158)
198 79/0 80%/0%
7-8 ask/earn Victor 71 119 48 60%
Juan 71/47
(avg. 59)
28 68%
9-10 house/earth Victor 79 79 0 0%

  /o/     //    Difference  V/Vː ratio
11-12 follow/plant Victor55 79 24 70%

(avg. 55)
(avg. 146)
91 38%

The picture is not totally clear: Juan gave one variant of the short /a/ in ‘burn’ which was 198 msec. long, as long as the long // in ‘hide’, or indeed any vowel in the whole data. If we ignore this occurrence, the upper limit for the short vowels is brought down to 119 msec., which is reached in Juan’s other token of ‘burn’. The average duration for the short vowels (all stressed) is 81 msec. The 13 long vowels range from 79-198 msec., with an average duration of 110 msec. The average ratio of short vowel/long vowel is then 74% (81/110). The vowels in the tokens from Victor or the near-minimal pair ‘house/earth’ are of the same length. In all other cases the long vowel of the pair is longer than the corresponding short vowel. Differences range from 16 msec. to 91 msec. (averaged when there are two tokens of a word). Short vowel/long vowel ratios range from 38% (follow/plant) to 82% (comet/grass). The 38% ratio from Juan’s ‘follow/plant’ is much lower than the other ratios. If we throw it out, the lowest ratio is 60% from Victor’s ‘ask/earn’.

Thus it would seem clear that there is an increase of vowel length where there were long vowels historically in these pairs (with the possible exception of ‘house/earth’, the near-minimal pair). In no case was a historically short vowel pronounced longer than its long counterpart: in two cases they were even, and in the rest of the cases there was a substantial increase. The 74% V/Vː ratio (38%-80% range) in these minimal pairs is minimal but just within the limit of what would be a plausible length distinction. Lehiste (1970:33) reports that “In languages in which stressed vowels have two contrastive degrees of quantity, the V/Vː ratio is close to 50% but may vary a great deal.” He gives the following examples to illustrate the range of V/Vː ratios found in languages:

V/Vː Ratio
Danish (Fischer-Jørgensen, 1955)  50.5%
Serbo-Croatian (Lehiste and Ivic, 1963)    67%
Thai (Abrahamson, 1962)  28-50%
German dialects (Zwirner 1959, 1962)  51% (western dialects)
  90.3% (eastern dialects)

Concerning the Eastern German dialects, Lehiste observes that “the near equality in the duration of long and short vowels in some dialects raises the question whether in these dialects duration is the primary distinctive factor”. Various experiments have shown that differences in duration on the order of 10% (90% V/Vː ratio) are the threshold of perceptibility. The V/Vː ranges found in the RD data are well below this threshold, but still quite a bit higher than the more typical 50% ratio usually found in languages with a vowel length difference.

In addition to the elicited tokens listed in Section III, tokens of two of the minimal pair members with long vowels were taken out of a recording of a discourse by Victor. They are as follows:

19. seːkitoːka87-40-79-79‘we plant it’
20. tlaːlli79-55‘earth’

It is interesting to note that in both cases the long vowels are the same length as in the elicitation tokens given above (79 msec.). The short vs. long contrast wasn’t significant in the elicitation pair ‘house/earth’, and the discourse token of ‘earth’ does not change that at all, as it has the same values for its two vowels (79-55) as the elicited token of ‘house’. It is possible, however, that the two elicitation pair members are distinguished by the length of the second (historically short) vowel of the words, as suggested in the last paragraph of Section IV.

In the words which do not have a minimal or near-minimal contrasting partner (#’s 13-19) vowel length seems to be a much more sporadic thing. These words were all taken from recorded texts by Juan. In general the stressed vowels seem to be longer than their neighbors, irrespective of whether they were historically long or short. The longest vowel in these tokens is the // in ‘he cried’; it would appear possible that Juan maintains length on that vowel (the ‘past’ prefix). Otherwise the longest vowels in each word are stressed vowels: in two words this stressed vowel is historically long (#’s 15 and 18). In one case (#16) the stressed historically long vowel has the same duration as the following short vowel.

In sum, there seems to be some support for the hypothesis that length is preserved where it signals meaning (the minimal pairs), but tends to be lost elsewhere.

VI. Miscellaneous

One thing that stood out clearly on the spectrograms is the way that the double /ll/ in words like /kalli/ ‘house’ or /tlaːlli/ ‘earth’ was consistently long. In six occurrences, including two in unstressed position discourse, it averaged 150 msec in length, with a range of 79-221 milliseconds. And this is in a consonant, which are typically shorter than vowels. The speakers of RD can make a long segment if they want to!

VII Spectrograms

Following are appended the spectrograms on which this analysis was based, in the order of their presentation in Section III.


Boas, Franz. 1917. “El Dialecto Mexicano de Pochutla, Oaxaca.” IJAL 1.9-44.

Campbell, Lyle. 1976. “El Pipil de El Salvador.” América Indígena.

Carochi, Horacio. 1645. Compendio del Arte de la Lengua Mexicana. Edited version by Ignacio de Paredes, 1759. Mexico: Imprenta de la Biblioteca Mexicana.

Goller, Theodore R., Patricia L. Goller, and Viola G. Waterhouse. 1974.
“The Phonemes of Orizaba Náhuatl.” IJAL 40.126-131.

Key, Harold, and Mary Key. 1953. Vocabulario Mejicano de la Sierra de Zacapoaxtla, Puebla. Mexico City: Summer Institute of Linguistics.

Knab, Tim. 1977. “The Long and Short of Aztec Dialects.” BLS 3.74-84.

Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, Massachusetts: MIT press.

McQuown, Norman A. 1942. “La Fonémica de un Dialecto Olmeca-Mexicano de la Sierra Norte de Puebla.” El México Antiguo 6.61-72.

Pittman, Richard S. 1954. A Grammar of Tetelcingo (Morelos) Nahuatl. Supplement to Language 30. Language Dissertation No. 50.

Robinson, Dow R. (ed.) 1969. Aztec Studies I, Phonological and Grammatical Studies in Modern Nahuatl Dialects. Publications in Linguistics and Related Fields, 19.
Norman, Oklahoma: Summer Institute of Linguistics.

Return to David Tuggy’s main page