Tuesday, July 28, 2015

The hard business of trying to specify allomorphs in FLEx

While a substantial part of my research is on the phonetics and phonology of different Otomanguean languages, I have been working on the morphophonology of the Itunyoso Triqui language for many years. Ever since I first started my work on the language, I was fascinated by the many ways in which a single verb root, for instance, could have a multitude of forms when one includes aspectual prefixes and personal enclitics.

One of the most notable things about Triqui morphology is just how much tone plays a role in marking different distinctions. Take the verb /a³chi³/ 'to peel', for example. There are four possible tonal shapes of stems, shown below (note "j" is /h/, "h" is /ʔ/, and a post-vocalic "n" in the final syllable marks contrastive vowel nasality):

Table 1: Stem shapes of verb /a³chi³/ 'to peel.'
This particular paradigm displays some common patterns in Triqui morphology. First, the 1st person singular is marked by a change in tone (to /5/) and involves the insertion of a coda "j" /h/. Second, the 2nd person singular is marked by tone raising to /4/ before the clitic. Third, the perfective prefix on vowel-initial stems is just /k-/. Fourth, the potential prefix involves prefixation of /k-/ and a change of tone on the initial syllable of the root. 

The result of these processes is five possible stem shapes: /a³chi³, a³chij⁵, a³chi⁴, a²chij⁵, a²chi³/, marked in bold above. Each of these morphological processes can be described well enough. However, things start to get rather messy when we wish to include additional verbs. Note the verb /a³chinj⁵/ 'to request' below.

Table 2: Stem shapes of verb /a³chinj⁵/ 'to request.'
We notice different patterns here. Instead of inserting a coda "j" /h/ to mark first person, we delete it from the root and change tone /5/ to /43/. Since the verb stem already has a high final stem tone, we do not observe any tone raising before the 2S clitic /=reh¹/. However, the form of the potential is rather different. Like in the habitual or unmarked form of the verb, we find that the coda "j" /h/ is deleted, but the entire stem changes its tone to /2/. This change is not particular to the 1S either - it occurs with all other persons in the potential, as the example with the 3SM clitic demonstrates. As a result of these processes, we have four possible stem shapes for the verb in Table 2: /a³chinj⁵, a³chin⁴³, a²chin², a²chinj²/.

I won't begin to provide a full analysis of the tonal morphology in Trique here (but see DiCanio, forthcoming). Rather, I wish to focus on two particular patterns and to discuss how they might be analyzed from a practical point of view. The first pattern is the marking of the 1S. This involves either the insertion of a coda "j" if it is not present on the stem or its deletion if it is present. Such a process is called a morphological reversal or exchange rule (see Inkelas, 2014). Tonal changes co-occur with this process for verbs with upper register tones (DiCanio, forthcoming), but we will not focus on these here.

The second pattern involves the way in which the potential aspect is marked. For certain verbs, it is marked by a change to tone /2/ on the syllable to which the prefix is attached, as in Table 1. On other verbs, it is marked by a change to tone /2/ on every syllable of the stem, as in Table 2. In such cases, the 1st person clitic no longer involves a tone change since the tone on the stem is now /2/, which belongs to the lower register. (Incidentally, one might describe this as a case of morphological opacity, where stage 1 prefixal/aspectual morphology bleeds the conditions for the application of clitic tone raising.)

At least segmentally, the 1S clitic is easy enough to characterize, though how might one go about marking such forms in a digital lexicon/dictionary like FLEx? One procedure might be to mark each and every 1S form, e.g. include /a³chij⁵/ 'peel.1S' as a variant of /a³chi³/ 'peel.' While certain of the morphological patterns are motivated by phonological well-formedness constraints (DiCanio, forthcoming), listing the variants in a table or paradigm as above provides a useful framework for describing the morphological patterns within the Triqui lexicon. 

This "listing" approach is the one that I currently use. However, doing this is rather time-consuming, as all words in the Triqui lexicon undergo this very regular alternation (though the tonal processes are rather complex). Doing this also loses the broader generalization of the rule. Moreover, there is currently no neat way of including paradigms within FLEx; one must specify additional forms as variants or allomorphs derived via a rule.

Another approach might be to create a phonological rule within FLEx's phonological grammar. However, the only available way to encode such rules is via a classical rewrite rule. This would produce rules of the form: Vh > V /_# ; and V > Vh /_#. Yet, there is no way to connect this particular rule with the set of morphological processes that it affects. It is an alternation that is primarily used for marking the 1st person singular (though similar alternations also mark previously-mentioned 3rd person discourse referents and derive nominal forms from quantifiers).

The same possibilities seem to be relevant for the potential aspect marking. It is either specified in a paradigm or it can be derived via a rule. However, a new problem presents itself when one considers the latter possibility. For those verbs, as in Table 2, which undergo an entire stem change to tone /2/ with the potential aspect, what is the phonological environment for a rewrite rule? It is the entire word's tonal melody. FLEx currently provides no way of separating the stem's tonal shape from the stem itself as one might do with an autosegmental representation. Thus, FLEx is unable to make sense of a string like /ka²chin²/ 'request.POT.1S.' when it comes to morphological parsing.

This problem is compounded by the nature of Triqui morphology when one considers the interaction between the potential aspect and 1S marking mentioned above. If there are a specific set of rewrite rules for the 1S clitic, one must specify that the tonal part of the alternation does not apply if the stem has undergone a change to the potential aspect. I currently know of no solution as to how one might resolve these issues within a FLEx lexicon.


DiCanio, C. (forthcoming) Tonal classes in Itunyoso Triqui person morphology, in Tone and Inflection, Empirical Approaches to Language Typology series, Mouton de Gruyter, Palancar, Enrique and Léonard, Jean-Léo (eds).

Inkelas, Sharon (2014) The interplay of Morphology and Phonology. Oxford Surveys in Syntax and Morphology. Oxford, UK.

1 comment:

  1. Have you tried using Dekereke and Flex together?