1. The linguistic problem.

The problem to be investigated in this study concerns the position of so-called clitic pronouns in Spanish. During the documented history of these elements in Old Spanish many positional arrangements were possible; clitics, however, could not be clause initial. In contrast, in Modern Spanish the so-called clitic pronouns are predominantly proclitic to the verb and their position is strictly fixed regardless of the position within the sentence of their host verb, leading to an inflectional affix analysis in some linguistic studies (cf. Givón, 1976).

However, the prosodic-based principles used in attempts to account for clitic placement in Old Spanish do not explain the kind of variation observed during the transition to proclisis that takes place during the 15^th and 16^th centuries. For instance, Meilán (1991:85), upon observing the variation of these structures in his 15th century data, considers that the order "seems to be completely free." Similarly Keniston (1937:94) observes that in variable contexts, "it does not seem possible to discover any functional principles which would explain the practice of the sixteenth century." Traditional explanations that invoke the presence of some preceding stressed element find unsolvable challenges in the 16th century data in which verbs in identical syntactic environments exhibit both proclisis and enclisis. For instance, in both examples (1) and (2) the verb is preceded by a conditional clause with a stressed subject NP. Nevertheless we have enclisis in the former and proclisis in the latter. Likewise, pronoun position cannot be linked to the presence of a stressed subject NP in the preceding clause (cf. examples (3) through (6)).⁽²⁾
 

(1) si el confesor la tray a más perfeción, aprétala tanto [50-7]
if the confessor brings her to more perfection, presses-her so-much
 

(2) si éste quiere dar lugar a sus vanidades, lo hace todo. [53-12]
if this-one wants to yield to his vanity, he-does-it everything
 

(3) en 0 quiriendo algo más, se perderá todo. [119-5]
upon (one) wanting something else, everything will be lost
 

(4) porque como ha tanto que 0 tratan de virtud, paréceles que… [II52-14]
because (they) have been dealing with virtue for so long, it seems to them that…
 

(5) Cuando el Señor quiere tornar el alma así, pónela … [140-14]
when the Lord wants to turn the soul like this, He puts it…
 

(6) Y vosotras, si advertís en ello, lo entenderéis [20,10]
And you-pl, if pay attention to it, you will understand it
 

The explanation for the regularization of pronoun position to be developed in this study proposes that during the 16^th century, the former Old Spanish evolves into an affixal object marker attached to the verb. This allows its appearance in clause initial position. Consequently, in explaining the general changes affecting the position of the emerging object agreement markers illustrated in examples (1) through (6), it is important to consider the universal mechanisms that govern morphological processes.

The hypothesis advanced here is that the patterns of diffusion of proclisis are connected to morphologization and the processing consequences it brings about. From a language processing point of view, affixation comes into being through what has been described as a "flirting process;" that is to say, the process through which language learners tentatively reanalyze the combination of the semantically and phonologically decayed word and its potential host as word + affix (Hall, 1992). The reanalysis may be rejected or accepted based on, among other factors, whether affixation entails some type of processing difficulty.

In the case at issue, we have seen that unstressed pronouns occur both in front or after the verb. As a result, these elements approach verb stems as both potential prefixes and suffixes. I will argue that the prefixation of the emerging object agreement markers has a processing advantage. It avoids additional morphological material attached to already existing tense, mood, aspect and subject agreement inflections in the Spanish verb paradigm. This favors the shift to proclisis in those verb forms that are especially complex to process. These typically involve forms that are textually less frequent and have more inflectional suffixes.

The idea that the patterns of positioning of object agreement markers during the morphologization of object agreement are sensitive to the morphological complexity of the verb form finds support in the fact that the resulting object agreement markers in Modern Spanish exhibit a complementary distribution of prefixes in finite verb forms and suffixes in non-finite forms (cf. Rini 1995 for a similar idea).⁽³⁾

In fully testing this hypothesis, this study adopts several perspectives. First, I examine the formal realization of verbal inflections from a typological perspective and use psycholinguistic, semantic, and historical factors to account for the form of agreement categories. Then, a statistical treatment of actual data from the period of morphologization of object agreement in Spanish is used as a micro-analysis of language change. This enables us to observe how the proposed principles actually function in a natural language situation.

The corpus used in the typological study consists of three samples collected by Leon Stassen, Revere Perkins and Gary Gilligan totaling 203 languages (cf. Hawkins and Gilligan 1988: 249-257). The verbal inflections investigated are tense, mood, aspect, and subject and object agreement. In order to complete the insufficient data on object agreement not recorded in the samples collected by Stassen and Gilligan, data from Dryer's data base (personal communication) has been used. This combination of samples will be referred to as the "combined sample."
 

2. The data and universals.

Various cross-linguistic studies of morphology have observed that suffixes are more frequent than prefixes (Greenberg 1957; Hawkins and Gilligan 1988; Hawkins and Cutler 1988, Hall 1992). This distribution seems to be in close relation with head-ordering typology. While verb final languages are almost exclusively suffixing, non-verb final languages allow for both prefixes and suffixes. The correlations between basic word order and tense, aspect, and mood morpheme order in the sample are displayed in Table 1.⁽⁴⁾
 

TABLE 1

Tense, mood and aspect affix order correlation with word order
 

TENSE

MOOD

ASPECT

Prefixes

Suffixes

Prefixes

Suffixes

Prefixes

Suffixes

VO	20	18%	27	25%	8.5	15%	13.5	23%	28.5	25%	30.5	26%
OV	2	2%	61	55%	2.5	4%	33.5	58%	8.5	7%	48.5	42%
TOTAL	22	20%	88	80%	11	19%	47	81%	37	32%	79	68%

 

Table 1 demonstrates that while VO languages allow both prefixing and suffixing, OV languages are skewed in favor of exclusive suffixing with only 2%, 4% and 7% of the languages exhibiting tense, mood or aspect prefixes respectively.

Subject agreement and object agreement in verbs, however, do not follow these general tendencies. Table 2 shows that agreement categories are consistently more resistant to the suffixing preference and that there is no correlation between word order and the forms of agreement markers.
 

TABLE 2
Agreement affix order correlations with word order
 

SUBJECT AGREEMENT

OBJECT AGREEMENT

Prefixes

Suffixes

Prefixes

Suffixes

VO	41	32%	26	20%	16.5	27%	17.5	30%
OV	17.5	14%	44.5	34%	17.5	30%	8.5	14%
TOTAL	58.5	46%	70.5	54%	34	56%	26	44%

In the combined sample, subject agreement is realized as a prefix in 46% of the languages, while 54% have suffixes. Object agreement exhibits 56% prefixing and 44% suffixing. Other language samples confirm these results, namely that agreement categories, and especially object agreement, do not show a suffixing preference.⁽⁵⁾
 

3. Explanations for the form of verbal inflections.

A number of different (although not conflicting) accounts for the form of verbal inflections has been given in the literature. The first one, originally formulated in Greenberg (1957) and developed by Hawkins and Cutler (1988) and Hawkins and Gilligan (1988), invokes principles of lexical processing to specify a universal suffixing preference. The second, the Head Ordering Principle (HOP), stems from some generative work in morphology (Aronoff 1976; Williams 1981). The HOP predicts that the order of affixes with respect to their host words will align with the order of the heads in the syntax, thus predicting suffixes in OV languages and prefixes in VO. However, as we have just seen in tables 1 and 2, while such explanations account for tense, mood and aspect verbal inflections, the data shows that agreement morphemes do not show a suffixing preference nor exhibit a correlation with syntactical order (cf. Siewierska and Bakker 1994 for similar conclusions).

A third explanation appeals to principles of language change (Givón 1979, also discussed in Bybee Perkins and Pagliuca 1990, Hall 1992, Siewierska and Bakker 1994), and advocates a preference for morphemes to be located in the positions of the erstwhile independent words that gave rise to them. Since it is assumed that the syntax of the language determines the order of free lexical morphemes, in Givón's explanation, the order of "today's morphology" is determined by the order of "yesterday's syntax."

However, if Givón's principles were the only motivation for these correlations it would be necessary to show that in all of these morphologization processes the original syntactic order of the elements involved has been preserved. That is not the case, as can be seen from the readily attested development of Romance object clitic pronouns. Cliticized objects have not preserved their original syntactic order: while object noun phrases regularly follow finite verb forms, cliticized objects are preverbal. This preverbal position cannot be a result of the morphologization of these bound pronouns during the OV stage of Romance, since the preverbal position of these clitics is a relatively recent development established only long after the VO order had become dominant. This means that, even though Givón's assumptions concerning the origin and evolution of inflections are confirmed in some attested historical derivations, it is clear that they do not allow us to make universal generalizations about the position of affixes.

I consider that there are two main weaknesses in these attempts to explain the form of grammatical markers. First, they usually argue for a single principle while ignoring competing theories. Second, they fail to notice that the extent to which the principles proposed for explanation correlate with affix position varies considerably with the semantic category of the affix. As seen in Tables 1 and 2, agreement categories, and tense, mood, aspect categories behave quite differently in regards to affix position. I believe that the answer may be found in an account that combines all of the factors that determine the position of affixes with special attention to the processing consequences of the morphologization of agreement.
 

4. A revised explanation
One of the factors that determines the form of grammatical markers is the frequency with which the words candidate for morphologization occur contiguous to the verb stem (cf. Bybee 1985: 38-43). Following the principle that words that function together in the sentence tend to occur together in the sentence (Vennemann, 1973: 41), a word whose scope is primarily the verb such as a tense auxiliary, will appear closer to the verb and become a ready candidate for desemantization and fusion to the verb. On the other hand, a word like a pronoun, whose scope may include the whole proposition and whose meaning is related to the participants in the action rather than the action described by the verb, will occur further from the verb, and a process of desemantization and fusion to the verb will be less likely to occur. Moreover, even if both tense auxiliaries and object pronouns occur contiguous to the verb, the former may be more likely to attach to the verb than the latter if the combination verb+tense occurs in every clause but the juxtaposition of verb+object-pronoun does not occur in intransitive clauses or clauses containing a nominal object.

There are two consequences of this principle. First, the expression units of agreement reflect a lower degree of fusion to verb stems. While tense, mood and aspect categories are often signaled by means of stem changes, reduplication of the stem or vowel harmony, agreement remains more easily segmentable and loosely fused to the stem. Crosslinguistically it is far more common for agreement categories to be signaled by means of clitics or detached markers than it is for tense, mood or aspect (cf. Bybee 1985). This makes agreement inflections more vulnerable to a change in their position due to the pressure of pragmatic or processing factors.

Second, this principle explains why agreement inflections are less frequent crosslinguistically. As observed in Greenberg (1966: 93) the presence of inflectional subject agreement in a language implies the presence of either aspect, tense or mood and, as observed in Siewierska and Bakker (1994), no language has object agreement that does not also have subject agreement. This distribution is confirmed in the analysis of the combined sample. Based on the morpheme distribution suggested by all these samples an implicational universal emerges, which may be summarized as in (7):
 

(7) tense/mood/aspect > subject agreement > object agreement
 

In order for a language to have inflections for a given category, it must have inflections for at least one of the categories in each of the positions to its left in the hierarchy.
 

In addition to the frequency of the juxtaposition of words candidate for morphologization and verb stems, there is a second factor that may influence the outcome of a morphologization process: the type of source from which inflections develop. Tense, mood and aspect affixes derive from verbs, more specifically auxiliaries, which tend to reflect a position consistent with word order typology. For instance, in the great majority of the SOV languages, auxiliaries follow the main verb (cf. Greenberg 1966, Dryer 1992). As a result, when the auxiliary loses its free lexical status in a head-final language it will typically result in a suffix. Similarly, affixes deriving from heads in a head-initial language will likely be realized as prefixes, yielding a correlation between the position of affixes and the position of heads within the syntax. In contrast, agreement categories derive from subject and object unstressed pronouns which, contrary to auxiliaries, are more vulnerable to factors like topicalization, prosody, and other constraints such as verb second. Since these factors often influence the order in which clitics are placed with respect to the host, the position of agreement affixes may reflect an order that does not correlate with syntactical word order.

These historical and semantic factors would explain why the position of agreement inflections does not correlate with syntactical order. However, further explanation is needed to account for why agreement inflections are more resistant to the suffixing preference. I believe this may be explained by invoking processing factors in interaction with universal (7) above.

As pointed out in section 3, a tradition of research has linked the suffixing preference to principles of language processing. Prefixing entails greater representational complexity and hinders the access to the stem-based lexicon. Consequently, prefixes are disfavored crosslinguistically. Despite the lack of a suffixing preference for agreement inflections, this argument seems to be basically correct. However, I consider that there is another factor, not covered in the literature, namely polymorphemicity, that also results in difficulties in the processing of verbal morphology. The reason for this difficulty is that affixing (i.e both suffixing and prefixing) always entails some degree of processing difficulty (cf. Hall, 1992) and consequently, polymorphemicity increases the processing demand even more. As a result, prefixing may be preferred in languages in which there are already many suffixes, since new suffixed inflections may lead to additional performance difficulties, such as an awkward stress group. This means that, under certain circumstances, such as high polymorphemicity in exclusively suffixing languages, the general processing advantages associated with suffixing may cease to apply, and prefixing may then constitute a relatively more efficient form of morphologization.

Due to the fact that the inflectional expression of agreement depends implicationally on the inflectional expression of other grammatical categories, and not viceversa (as seen in universal (7) above), agreement categories are the verbal inflections most likely to emerge in already highly polymorphemic hosts. This fact, in combination with the observation that the suffixing preference may be invalidated in forms in which there is already a lot of suffixing, leads to an explanation for the non-preponderance of suffixes in agreement inflections.⁽⁶⁾

This hypothesis predicts that, in a representative sample of the languages of the world, polymorphemicity should influence the distribution of inflections. As shown in tables 3 through 5, this prediction finds wide support in the data from the combined sample. In these tables, numbers one to five in the upper row of the tables indicate the number of inflectional grammatical categories present in the languages in question (the categories considered are tense, mood, aspect, and subject and object agreement). The columns indicate number of languages and column percentages.
 

TABLE 3
Type of affixation correlations with polymorphemicity in OV languages (N=76)
 

No of grammatical categories

1

2

3

4

5

only suffixes	3	100%	11	92%	19	57%	7	34%	2	29%
only prefixes	0	0%	0	0%	2	6%	0	0%	0	0%
both suf/pref	0	0%	1	8%	12	36%	14	66%	5	71%

TOTAL

3

12

33

21

7

Table 3 presents the distribution of affixes in the 76 OV languages in the combined sample. As first observed in Greenberg (1966) prefixes are most dispreferred in head-final languages. However, the distribution of prefixes exhibits a clear correlation with polymorphemicity in the sense that prefixing tends to occur more in languages that have more inflections. Languages with one verbal inflection do not present any prefixing and languages with two inflections have some prefixing in 8% of the languages (1 out of total 12). As the number of inflectional categories increases the percentage of languages with some prefixing also increases. Languages with three, four and five categories of inflections exhibit some suffixing in 36%, 66% and 71% of the languages respectively.
 

TABLE 4
Type of affixation correlations with polymorphemicity in VO languages (N=75)
 

No of grammatical categories

1

2

3

4

5

only suffixes	3	30%	11	58%	3	19%	3	18%	0	0%
only prefixes	5	50%	4	21%	0	0%	1	6%	2	15%
both suf / pref	2	20%	4	21%	13	81%	13	76%	11	85%

TOTAL

10

19

16

17

13

While OV languages favor suffixing, languages with VO order allow both prefixing and suffixing on just about equal terms (cf. Table 4). Consequently, there will be cases in which, like OV languages, polymorphemicity will result in an accumulation of suffixes leading to prefixing as an alternative morphologization type. However, in head initial languages high polymorphemicity in the form of a lot of prefixing is possible too. In those cases in which prefixing is the dominant tendency, suffixing may be the alternative in morphologization in highly polymorphemic hosts. As a result, unless we look at individual languages, it is not possible to determine whether a high affix to word ratio in VO languages will result in more suffixing or prefixing, since this will depend on the dominant position of affixes in each language. All that can be said is that, according to the hypothesis here, both exclusive suffixing and exclusive prefixing should be tolerated in languages with few inflections and dispreferred in languages with a lot of inflections. Table 4 provides some support for this prediction, since languages with one and two inflections have both affixation types in roughly 20% of the cases while languages with three, four and five inflections exhibit both prefixing and suffixing in percentages that range between 76% and 85%.
 

TABLE 5
Type of affixation correlations with polymorphemicity in the total sample (N=151)
 

No of grammatical categories

1

2

3

4

5

only suffixes	6	46%	22	71%	22	45%	10	26%	2	10%
only prefixes	5	38%	4	13%	2	4%	1	3%	2	10%

only suf or pref	11	85%	26	84%	24	49%	11	29%	4	20%
both suf and pref	2	15%	5	16%	25	51%	27	71%	16	80%

TOTAL

13

31

49

38

20

Finally, Table 5 includes the total 151 languages with some type of verbal affixation in the combined sample. In consonance with the hypothesis that polymorphemicity results in both prefixing and suffixing, languages with only one or two categories of inflections have both types of affixation in 15% and 16% of the cases respectively. Languages with three verbal inflections have suffixes and prefixes in roughly half of the languages (51%); and languages with four and five categories of verbal inflections have both types of affixation in 71% and 80% respectively.⁽⁷⁾

In summary, the crosslinguistic data presented here provides support for the hypothesis that in languages with many inflectional suffixes, additional affixes are more likely to be prefixes than in languages with fewer affixes. The effect of this is that since agreement morphemes attach to verbs generally only when other categories exist as affixes, agreement morphemes will be prefixes more often that other categories.
 

5. A micro-analysis of historical change
The purpose of this last section is to reconstruct the historical scenario of the morphologization of object agreement markers in Spanish in order to illustrate how the principles proposed to explain the form of agreement markers actually take place in a natural language situation. By the second half of the 16^th century, with the increase in textual frequency brought about by the obligatoriness of clitic doubling, the already semantically and phonologically weak unstressed object pronouns became routinized, and thus, available to fusion to verb stems. Since unstressed pronouns could occur both before and after the verb, they were liable to become suffixes or prefixes. The preceding sections have shown that factors associated with language processing may determine the choice of position for the emerging agreement inflections. One obvious difference between the combinations of suffixed and prefixed unstressed pronouns is that, given the rich suffixal morphology exhibited in the Spanish verbal paradigm, additional suffixes create a complex sequence of morphemes with processing consequences. Quite possibly this articulatory difference favored the choice of prefixed object markers.

In order to measure quantitatively the correlation between degree of processing difficulty and proclisis, examples of verb-clitic combination in a variable environment have been coded. Three complete works Camino de perfección and Las moradas by Teresa de Jesús and La gitanilla by Cervantes have been coded in their entirety. The examples selected correspond to unstressed pronouns carrying argument structure of verbs occurring in the following linguistic environments: a) clause initial verbs in main sentences conjoined with y `and' (cf. (8) and (9)); and b) verbs in main sentences after the pause that usually follows an embedded clause (cf. examples (10) through (13)).

(8) . . . obra Dios tanto en el alma y hácela tantas mercedes, que . . . [105,3]
works God so much on the soul and does it so many favors, that

(9) . . . lloraba con el Señor y le suplicaba remediase tanto mal. [12,1]
I would cry with the Lord and beg Him to remedy so much evil

(10) aunque no sea en toda perfeción, vese que . . . . [118-13]
although not in all perfection, it is seen that…

(11) aunque sea en sí honrado, le tienen en poco [21:14]
eventhough he is honest, he is not highly regarded

(12) porque no va libre la vista, ciéganos el polvo [168-1]
because eyesight is not free, the dust blinds us

(13) porque os guardéis de ellas, las pongo aquí [97-11n]
in order for you to keep away from them, I put them down here

The data contain a total of 314 examples of object markers occurring with verbs in the specified contexts, with a distribution of proclisis and enclisis close to 50-50. As examples (1-6) and (8-13) illustrate, position of pronouns cannot be predicted from the type of embedded clause preceding the verb: both proclisis and enclisis are possible after identical syntactic environments.

As it follows from the theory developed in this investigation, the position of agreement markers should be sensitive to morphological features of the verb. The general prediction is that verbal forms that are harder to process should favor proclisis. First we must rate the relative processing difficulty of any verb form. In order to do so, I have established a markedness scale (see table 6) based on the features that are typically associated with markedness (Eckman, Moravcsik and Wirth 1986): forms that a) are textually less frequent, b) carry more marked grammatical meanings and/or c) exhibit additional morphological material, receive higher markedness values.

TABLE 6
Markedness values in the Spanish verb
 

Category	0 (unmarked)	1 (marked)	2 (most marked)
aspect	present  preterite future	imperfect
tense	present	preterite  imperfect	future
number	singular	plural
person	3rd	1st	2nd

To measure the markedness level of a given form we simply sum the numerical values in Table 6 (zero, one and two for unmarked, marked and most marked respectively) depending on what grammatical meanings are represented in the verb form in question.⁽⁸⁾

Table 7 shows the results of the cross-tabulation of clitic position and markedness. As the hypothesis presented in this investigation predicted, less marked forms (i.e. more easily processable) such as third person singular of the Present Indicative retain enclisis with greater frequency. As markedness values increase, so does processing difficulty, and proclisis is more frequent. This correlation is statistically significant at p < .0001.

TABLE 7
Crosstabulation of clitic position and markedness values
 

Markedness

0

1

2

3/more

TOTAL

Preverbal	36	30%	63	59%	39	63%	20	74%	158	50%
Postverbal	83	70%	43	41%	23	37%	7	26%	156	50%

TOTAL	119	106	62	27	314
	37.9%	33.8%	19.7%	7.6%	100%

The results in Table 7 illustrate the hypothesis developed in this investigation, in the context of a linguistic change in progress. There is a close relationship between the position of agreement markers and the processing difficulty of the verbal morphology. However, given that marked forms typically exhibit more morphological material than non-marked ones, it could be possible that the observed correlation is just the effect of word length. That is to say, greater word length alone would trigger proclisis. Table 8 shows the percentages of proclisis by word length controlling for tense to demonstrate that less marked present forms produce proclisis less often, regardless of word size. Take for instance all verb forms with 3 syllables. While 3 syllable forms overall account for 49% of proclisis, present tense forms account for 43.8% of the cases, with significantly higher percentages for preterite with 48.1% and future with 70%. The same principle, that within verb forms with the same number of syllables, less marked tenses retain more enclisis, holds for the remaining word-length types.

TABLE 8
Percentages of proclisis by word length controlling for tense
 

Word length	1 syllable	2 syllables	3 syllables	4 syllables
Present	32.3%	35.4%	43.8%	50%
Preterite	50%	64.3%	48.1%	71.4%
Future		60%	70%	91.7%
Imperfect		100%	58.3%	100%
Overall proclisis	35.9%	47.1%	49.5%	78.9%

The correlations shown in the statistical analysis make clear that the spread of proclisis with finite verb forms happened first in marked forms, that is to say, in forms that are semantic and morphologicaly more complex. These diachronic facts and the cross-linguistic evidence support the hypothesis that psychological principles of language processing can influence the position of affixes during the stage at which language change is implemented. Especifically, a universal psycholinguistic principle prevents the fusion of grammatical markers to highly polymorphemic sequences of inflections. This explanation provides an alternative to the traditional prosodic-based explanations.

REFERENCES

A. TEXTS

Aguado, J. M. (ed.). 1929. Santa Teresa de Jesús. Camino de perfección (vol. I). Madrid: Clásicos Castellanos. Ediciones de La Lectura.

Aguado, J. M. (ed.). 1930. Santa Teresa de Jesús. Camino de perfección (vol. II). Madrid: Clásicos Castellanos. Ediciones de La Lectura.

Navarro Tomás, T. (ed.). 1910. Santa Teresa de Jesús. Las moradas. Madrid: Clásicos Castellanos. Ediciones de La Lectura.

Rodríguez Marín, F. (ed.). 1914. Miguel de Cervantes. La gitanilla, en Novelas ejemplares (vol. I:3-136). Madrid: Clásicos Castellanos. Ediciones de La Lectura.
 

B. STUDIES

Aronoff, M. 1976. Word formation in generative Grammar. Cambridge: MIT Press.

Bybee, J.L. 1985. Morphology. Amsterdam/Philadelphia: John Benjamins.

Bybee, J.L., W. Pagliuca and R.D. Perkins 1990. On the asymmetries in the affixation of grammatical material. In W. Croft, K. Denning and S. Kemmer (eds.), Studies in typology and diachrony: papers presented to Joseph H. Greenberg on his 75^th birthday. 1-42. Amsterdam/Philadelphia: John Benjamins

Cowgill, W. 1966. A search for universals in Indo-European diachronic morphology. In J.H. Greenberg (ed.), Universals of Language. Cambridge: MIT Press.

Dryer, M.S. 1992. The Greenbergian word order correlations. Language. 68. 81-138

Eckman, F., E. Moravcsik and J. Wirth (eds.). 1986. Markedness. New York: Plenum Press.

Enrique-Arias, A. 1993. La distribución de los pronombres de objeto en español: consideraciones históricas, tipológicas y psicolingüísticas. Lingüística, 5. 41-75.

Givón, T. 1976. "Topic, pronoun and grammatical agreement". In Li, C.N. (ed.), Subject and topic. New York: Academic Press.

Givón, T.1979. On understanding grammar. New York: Academic Press.

Greenberg, J.H. 1957. Order of affixing: a study in general linguistics. In Essays in linguistics. Chicago: University of Chicago Press.

Greenberg, J.H. 1966. "Some universals of grammar with particular reference to the order of meaningful elements". In J. H. Greenberg (ed.), Universals of language. 73-113. Cambridge: MIT press.

Hall, C.J. 1988. "Integrating diachronic and processing principles in explaining the suffixing preference." In J. A. Hawkins (ed.), Explaining language universals. 321-349 Oxford: Basil Blackwell

Hall, C.J. 1992. Morphology and Mind. London: Routledge.

Hawkins, J. and A. Cutler. 1988. Psycholinguistic factors in morphological asymmetry. In J. A. Hawkins (ed.), Explaining language universals. 280-317. Oxford: Basil Blackwell.

Hawkins, J. and G. Gilligan, G. 1988. Prefixing and suffixing universals in relation to basic word order. In J. A. Hawkins and H. K. Holmback (eds.), Papers in universal grammar: generative and typological approaches. Lingua Special Issue, 74 2/3. 219-259.

Keniston, H. 1937. The Syntax of Castilian Prose. The Sixteenth Century. Chicago: The University of Chicago Press.

Meilán García, A. 1991. La oración simple en la prosa castellana del siglo XV. Oviedo: Departamento de Filología Española.

Nichols, J. 1986. Head-marking and dependent-marking grammar. Language. 62. 56-119.

Rini, J. 1995. The evolution of the nature and position of the Spanish clitic pronoun. La Corónica. 24.1. 173-195.

Siewierska, A. and D. Bakker. 1994. The distribution of subject and object agreement and word order type. In A. Siewierska (ed.), Eurotype Working Papers, 6. 83-126.

Vennemann, T. 1973. Explanation in syntax. In J. Kimball (ed.), Syntax and Semantics II. New York: Academic Press.

Williams, E. 1981. On the notions "lexically related" and "head of a word". Linguistic Inquiry, 12. 245-74.
 

The research in this paper was partially supported by a Dissertation Award granted by the Del Amo Foundation. I would like to acknowledge useful discussions with Mayrene Bentley, Jack Hawkins, Stephen Matthews, Carmen Silva-Corvalán, Enric Vallduví and Gayle Vierma, and comments from three anonymous referees from Studies in Language on a previous version of this paper.

¹ The examples are from Teresa de Jesús, Camino de perfección, Madrid: Clásicos Castellanos, 1929 edition. The numbers between brackets correspond to page and line respectively.

² As far as I know Rini (1995) is the only study on the position of Spanish clitic pronouns that considers (albeit vaguely) factors of morphological processing in connection with the affixation of object agreement markers. Rini developed his theory without knowing of Enrique-Arias (1993), where I first formulate the hypothesis presented in this paper. Likewise, I learned of Rini's article only after finishing the present investigation.

³ The aggregate proportions of prefixing to suffixing are arrived at by calculating the ratio of prefixing to suffixing in all the languages that have entries for a given category. Consider the example of the 116 languages with aspect inflections (59 VO and 57 OV) distributed as follows: within the 59 VO languages 25 have suffixes, 23 languages have prefixes and 11 languages have both. The 57 OV have 47 languages with suffixes, 7 languages with prefixes and 3 languages with both. For the purposes of calculation, languages with both suffixes and prefixes will count as 0.5 prefixing and 0.5 suffixing, making the ratio of suffixing to prefixing for aspect inflections 30.5 to 28.5 in VO languages, 48.5 to 8.5 in OV languages, and 79 to 37 or 68% / 32% overall.

⁴ In all of the available samples agreement categories are the most resistant to the suffixing preference. The distribution of subject agreement markers ranges from 61% prefixing in Perkins' sample to over a third of prefixes (36%) in Stassen's sample. Siewierska and Bakker (1994) and Gilligan's sample exhibit values close to a 50-50 distribution of prefixes and suffixes. As for object agreement we have a strong preference for prefixes in Perkins' data (75%), and again values close to 50% in both Siewierka and Bakker (1994) and Dryer's sample (cf. Hawkins and Gilligan 1988:225).

⁵ The psycholinguistic argumentation applied here to the problem of affixation has not been explored in accounts of affix position in the literature. However, the data discussed in some studies seems to support the claim presented here that even though there is some fundamental difficulty about prefixes compared to suffixes, in words with lots of suffixes, prefixing may be a preferable affixation type. For instance, Cowgill (1966) has observed a correlation between polymorphemicity and prefixing in a corpus of data from Indo-European languages. He concludes that "a high morpheme-to-word ratio appears to go with a high prefix to suffix ratio" (p. 131). Bybee, Perkins, and Pagliuca (1990:8) observe that the only V-final languages with a high ratio of prefixing are those that have a high degree of affixation. Those `exceptional' prefixes are agreement markers. Hall (1988:332) observes that prefixing is more common in verbs than it is in nouns (39.3% vs. 18.4%). Again this yields a correlation between prefixing and polymorphemicity: while affixation in nouns is usually restricted to a maximum of two markers, the number of affixes a verb can have is virtually unlimited (cf. Nichols 1986:105).

⁶ Since the values in some of the cells in the first and fifth columns are less than 5, which may decrease the validity of the chi-square test, I performed tests collapsing the five groups of `number of grammatical categories' into four and three groups; this procedure yielded higher numbers. All such tests were statistically significant at p < 0001.

 ⁷ Let us apply this procedure to the Spanish third person singular forms of the Present Indicative (canta `he sings') and second person plural forms of the Future (cantaréis `you-plural will sing'). The levels of markedness for third person singular forms in Present Indicative (e.g. canta) are 0 for aspect and 0 for tense. Singular is 0 marked for number. Third person is 0 marked for person. The total correspondent markedness value for this form is 0. As for cantaréis the levels of markedness for second person plural forms in Future tense are 0 for aspect and 2 for tense. Plural is 1 marked for number. Second person is 2 marked for person. The total correspondent markedness value for this form is 5.