Latin verb inflection presents an interesting problem in morphological analysis. Beginning with data from the four regular conjugations and limiting ourselves to the present active indicative, we have the following matrix or paradigm, in which the forms of a representative verb from each conjugation occupy a column, and each of their person-number combinations a row. A gloss for the meaning constant to each column and row is given at their heads. The problem can be viewed as one of segmentation between stem and suffix -- how much of each form correlates with the meaning of its row, and how much with its column, and how can the forms be optimally disassembled along these lines and restored by algorithm. Note 1.
Conjugations: I II III IV
Glosses: 'love' 'advise' 'cover' 'hear'
1s amo: moneo: tego: audio:
2s ama:s mone:s tegis audi:s
3s amat monet tegit audit
1p ama:mus mone:mus tegimus audi:mus
2p ama:tis mone:tis tegitis audi:tis
3p amant monent tegunt audiunt
(Infinitives: ama:re mone:re tegere audi:re )
One solution has invariant stems and variant suffixes (although there is partial identity of the suffixes for 1s in I and III, and for 3s in III and IV):
I II III IV
Stems: am- mon- teg- aud-
Suffixes: 1s -o: -eo: -o: -io:
2s -a:s -e:s -is -i:s
3s -at -et -it -it
1p -a:mus -e:mus -imus -i:mus
2p -a:tis -e:tis -itis -i:tis
3p -ant -ent -unt -iunt
We can refer to this as a morphological solution. One must know to which morphological class (or conjugation) each verb stem belongs in order to determine which set of suffix variants to attach to it. If we were to analyze a number of additional members of each of these four classes in the same way, we would find that their stems all end in various consonants, and that there is nothing distinctive about their phonological shape that could be used to identify their class. The class membership of each verb must be learned in addition to the shape of its stem. It is one type of Item and Arrangement (IA) solution (Hockett 1954), one that consists of items (the stems and the suffixes) and their arrangements (an indication of which suffix variants go with which stems).
Another solution would be one with invariant suffixes and variant stems. For such a solution, the segmentation is generally done further to the right in each form, so that the varying vowel material making for suffix variants in the morphological solution above is included in the stems instead-the variability is transferred to the stems.
Stems: Suffixes: am- mone- teg- audi- -o: 1s ama:- mone:- tegi- audi:- -s 2s ama- mone- tegi- audi- -t 3s ama:- mone:- tegi- audi:- -mus 1p ama:- mone:- tegi- audi:- -tis 2p ama- mone- tegu- audiu- -nt 3p
An additional possible step in this analysis is to permute the rows of this matrix so as to make identical stem variants contiguous insofar as possible. One possible such arrangement is the following:
am- |mone-| teg- |audi-| -o:
| | | |
|ama-| |mone-| |tegi-| |audi-| -t
| | | |
|ama-| |mone-| tegu- audiu- -nt
ama:- mone:- |tegi-| audi:- -s,-mus,-tis
Identical stems are included between vertical lines. Note that there is no possible two-dimensional arrangement in which all identical variants are juxtaposed. In this arrangement, the form tegi- is interrupted by tegu- in the third column.
This can be called a Variant Stem solution. It too is an IA solution in that one needs to know which of the several stem variants for each verb to arrange with each suffix. What must one know about a given Latin verb in order to determine which of the above four patterns of stem variation it follows-the counterpart to the knowledge of conjugation membership required in the preceding solution? Is it possible to choose one variant for each verb to serve as its citation stem, from which one could determine the pattern of variation and predict the other variants? The answer is that the forms of either the third or fourth rows could be chosen for unambiguous listing. Two of the first-row variants end in consonants (am- and teg-), and thus could not unambiguously identify their patterns, while two of the second-row forms (tegi- and audi- ) end in short i and thus could not be chosen for the same reason. But if one uses as the citation stem for each verb its variant in the fourth row (which occurs with three of the suffixes), one can determine its pattern of stem variation simply by looking at its final vowel, and the stems fall into four groups: those in a:, e:, i, and i:. This is thus a phonological solution; the material built into the end of each stem identifies its class. From the point of view of the language learner (and presumably also the native speaking language user), a variant stem solution should require somewhat less memory burden than a variant suffix or conjugational solution. One listed form, properly chosen, can be used to determine the full pattern of each verb, whereas in a conjugational solution, it is necessary to know conjugation membership in addition to the shape of the only stem form available for listing. Note 2
A theoretically possible solution of a third type would have both invariant stems and invariant suffixes. This could be achieved if we were to use the citation stems from the variant stem analysis with all suffixes, and then use rules to adjust any infelicitous combinations that result. This is the sort of solution that Hockett (1954) termed an Item and Process solution-the items being the basic stems and suffixes, and the processes the necessary rules. (Although this type of solution was not favored by the Neo-Bloomfieldians of Hockett's day, it had been employed earlier by Bloomfield and Sapir themselves in the grammars they wrote of Native American languages, and it was again given legitimacy in early generative phonology.) Note 3 For example, one might adopt the following invariant stem forms for the Latin verbs under consideration, and combine them with the same invariant suffixes of the Variable Stem solution above, with the following results:
'love' 'advise' 'cover' 'hear' ama:- mone:- tege- audi: ama:o: mone:o: tegeo: audi:o: -o: 1s ama:s mone:s teges audi:s -s 2s ama:t mone:t teget audi:t -t 3s ama:mus mone:mus tegemus audi:mus -mus 1p ama:tis mone:tis tegetis audi:tis -tis 2p ama:nt mone:nt tegent audi:nt -nt 3p
Incorrect portions of the resultant forms are given in underlined boldface. They might be corrected by rules like the following, given here informally:
These changes are sketchy and informal, and there are several variations possible on the invariant stem forms that might be assumed and the resultant changes that would be required, but it is not our purpose here to pursue their details. Our main aim is to exemplify in broad outlines the chief types of solutions to the problem we have posed.
As noted above, this type of solution, with invariant stems and affixes, and rules to straighten out any infelicitous resultant combinations, is what has been termed an Item and Process (IP) solution. Note 5
To summarize, we have examined three types of solutions:
SOLUTION CLASSES STEMS AFFIXES RULES TYPE
"Conjugational" morphological Invariant Variant No IA
"Variant Stem" phonological Variant Invariant No IA
"Phonological" phonological Invariant Invariant Yes IP
Before one can decide which solution is preferable, it is necessary to clarify the purpose to which the solution is to be put, or what in the real world it is intended to model and illuminate. It turns out that there are at least as many possible purposes as there are solution types -- if not more. Note 6 Briefly, to mention several of the main purposes that have motivated research to date, (1) we may be interested in modeling the idealized native speaker-hearer, or (2) we may wish to assist the second language learner by constructing an optimal pedagogical grammar, or (3) we may wish to plumb to the depths what there is to be learned about the history of the language through the method of internal reconstruction, or (4) we may wish to predict future directions apt to occur in a language.
At this point, we have only some preliminary suggestions to offer. There is probably most agreement among linguists concerning the relevance of phonological or Item and Process solutions to internal reconstruction. Less commonly known or acknowledged is the relevance of Variant Stem solutions to the prediction of change resulting from the process known as paradigm leveling. Note 7 For this to be true, it would seem to be necessary for such processing of language data to also have some relevance for native speaker-hearers, those individuals who are ultimately responsible for effecting morphological change. Yet, in possible contradiction, one of the strong implications of generative grammar is the relation of Item and Process treatments to the everyday processing of the native speaker-hearer. Lastly, in spite of their seeming lack of economy, conjugational solutions should not be dismissed too readily, at least for the second language learner, in view of their lengthy tradition in pedagogical grammars. This, incidentally, keeps alive the question as to the relation of second to first language learning.
We do not mean to imply that we will answer any of these questions definitively in the course of this work. We do hope to suggest, however, that the answer to some if not all may depend ultimately upon the domain of morphology for at least part of their solution, and that this should be sufficient reason for those interested in such questions to approach the study of morphology with enthusiasm.
1. As to the purposes of such an exercise, we will have more to say. [back to text]
2. There is another question as to pattern or conjugation membership with respect to which the two solutions are equivalent. If one is presented with a Latin verb in context (and knows only that it is a regular present, active, indicative verb), the pattern followed by its other forms (and its conjugation membership) can be predicted except when it is a 3rd person singular form ending in -it (which is ambiguous between the tegi- and audi- patterns and between Conjugations III and IV) or when it is a 1st person singular form ending in Co: (which is ambiguous between the amo:- and tego:- patterns and between Conjugations I and III). [back to text]
3. Much has been written on this subject. See especially Hockett 1954 and Pearson 1977 (109-25). [back to text]
4. One might legitimately ask why e is chosen as the stem vowel for tege-, since within the scope of this problem none of the e's remain. We have only to look as far as the infinitive to find part of the answer, as well as a reason for excluding r from the consonants before which the change to i takes place. A less obvious reason is that this is the vowel chosen by Householder 1971 in two of the three possible solutions he discusses (218 ff.), and for which he is able to say that "As it happens, almost everything in these rules corresponds closely to one line of assumed historical development from Indo-European." Although there are several slightly different possible invariant stem-invariant suffix solutions that might be chosen for these data, one of the points we hope to make in this discussion is that in general they will tend to mirror actual historical developments. (This is as good a place as any to acknowledge my debt not only to Householder 1971 [especially Chapter 12], but also to many personal communications with him in my days as his student and since, for many of the ideas underlying this discussion. Those concerned with the history of ideas may be interested to know that Householder, in turn, attributed a number of them to his former teacher Roman Jakobson.) [back to text]
5. Again, credit Hockett 1954. [back to text]
6. This essay was written in the early eighties. Since then it has become increasingly clear to me that all three of the solutions that it explores assume, or may be interpreted to assume, that the morphemes into which the words are segmented are semantic primes, and that the words are in that sense compositional. A fourth alternative, in many ways radically different in its assumptions, is the Word and Paradigm (WP) solution mentioned by Hockett (1954) and discussed by Robins (1959) and Matthews (1991). Its view of the morpheme is more attractive than those of the other three, and its full implications need further study. Also, it has been impressed on me recently that morphological solutions, although seemingly requiring greater memory resources, have more than their persistence in pedagogical grammars to commend them, and seem to have greater reality for native speakers than phonological solutions. It should be obvious, then, that this essay does nothing more than introduce and scratch the surface of this most intriguing topic. [back to text]
7. See Jeffers and Lehiste, 1979. [back to text]
Hockett, Charles F. 1954. Two models of grammatical description. Word 10.210-231. [back to 1st ref.] [back to 2nd ref.] [back to Note 3] [back to Note 5] [back to Note 6]
Householder, Fred W. 1971. Linguistic speculations. Cambridge University Press. [back to Note 4]
Jeffers, Robert J., and Ilse Lehiste. 1979. Principles and methods for historical linguistics. Cambridge, MA: The MIT Press. [back to Note 7]
Matthews, P. H. 1991. Morphology. 2nd ed. Cambridge: Cambridge University Press. [back to Note 6]
Pearson, Bruce L. 1977. Introduction to linguistic concepts. New York: Knopf. [back to Note 3]
Robins, R. H. 1959. In defence of WP. Transactions of the Philological Society1959:116-144. Reprinted in Diversions of Bloomsbury: Selected writings on linguistics, by R. H. Robins, pp. 47-77. Amsterdam: North-Holland. [back to Note 6]