Assessing coherence in the Spanish PYTA morphome

1 Introduction

Regular verbs in Spanish are often defined as those whose stems do not change across verb conjugation (Cuenca and Jegerski 2023; Real Academia Española; Rivera et al. 2010). Irregular verbs on the other hand, often display verb allomorphy, yet always contain a matching stem alternant for various tenses: the preterit, two imperfect subjunctives, and the obsolete future subjunctive.^[1] Though these tenses share no cohesive function, their stem is identical for every Spanish verb, with allomorphy in some verbs, e.g., Sp. hice, hiciera, hiciese, hiciere (hacer) ‘to do’. Although the aforementioned tenses show no cohesive semantic function, they are the only remaining survivors of the Latin perfectum stem-explaining their formal coherence, and their irregular stems become autonomously morphological^[2] (Esher 2015; Maiden 2018). Abstract morphological templates are known as morphomes, an idea first proposed by Aronoff (1994).

The four previously mentioned tenses are said to comprise a morphome known as the PYTA morphome, perfectos y tiempos afines or ‘perfects and related tenses’ (O’Neill 2014, pp. 34–35). The PYTA morphome is distinguished by the loss of common function of a natural class (perfectivity), along with changes in the function of the Latin perfectum reflexes (Herce 2020; Herce and Cathcart 2024). Table 1 displays the loss of function, with Latin verb forms that were previously perfective becoming preterit or subjunctive forms in Spanish.

This PYTA morphome is the oldest emerging morphome of the four or five^[3] morphomic patterns that have been proposed thus far, established before the diversification of Proto-Romance into the different Romance languages (Herce 2020; Herce and Cathcart 2024). Some morphomes, such as the L, U, or N morphome, arise from sound changes that happened during or after the time of Proto-Romance, namely, palatalization of velars before front vowels in the case of the L and U morphomes,^[4] and diphthongization in stressed mid-vowels in the case of the N morphome. These patterns are shown in Table 2 for Spanish decir ‘to say’ and Table 3 for Spanish dormir ‘to sleep’, with members of the morphomes underlined.

One glaring question that remains in the discussion of morphomes is how psychologically ‘real’ they are (Maiden 2009, 2013). Although such patterns seem useful for L2 learners, teachers, or linguists, it is not clear to what extent native speakers internalize these patterns (Luís and Bermúdez-Otero 2016; Maiden 2013, 2018). Little research has been done to remedy this lingering question; one primary issue being that speakers of Latin, Proto-Romance, and Old Spanish have long since passed. One notable experiment conducted to assess the psychological reality of L-morphomes found that speakers of Romance languages were not able to generalize the pattern (Nevins et al. 2015). The L-morphome is defined by matching allomorphic stems in the 1sg present indicative and the entirety of the present subjunctive. Participants were given two morphophonological alternants in either the 1sg present indicative and 2sg present indicative or the present subjunctive and 2sg present indicative. They were asked to fill in the blank verb form for whichever form was not seen in the example, either the present subjunctive or 1sg present indicative. Nevins et al. found that participants preferred using the 2sg indicative form as the stem to form the corresponding present subjunctive or 1sg form regardless of being presented with an alternation in the examples. The authors refer to this as using the ‘natural’ form, i.e., a pattern that shows stem uniformity within mood between the 1sg and 2sg present indicative (Nevins et al. 2015).

Maiden’s analyses primarily focus on Romance verb paradigms, particularly stem distributions, and Maiden often attempts to assess the reality of the morphome in the mind of a speaker, though he admits that this psychological task is difficult, with some serious limitations (Maiden 2018). The idea of the morphome is particularly important in the evolution of the Romance languages, which contain many paradigms that appear to defy the laws of regular sound change or have completely innovative forms which bear little resemblance to their ancestral Latin etyma.

Maiden considers these ‘irregular’ diachronic innovations to be a key concept in proving the reality of morphomic patterns in the mind of a speaker (Maiden 2005). Such irregular innovation works to supplant the inherited forms that are transferred from speaker to speaker: If younger generations are merely memorizing alternant stem forms from an older generation, it is possible that they may never need to make an abstract generalization of morphomic patterns. However, if an innovated form is extended into the paradigm of a new lexeme, i.e., if morphomic abstraction is real, this pattern should behave ‘coherently’, affecting every cell of the pattern into which it has entered, with members of the morphome sharing the same stem. Strong forms of preterit verbs and the other PYTA verb paradigms in Spanish are then the perfect testing grounds for the coherence of morphomic patterns, as they undergo both analogical and morphological change diachronically (Maiden 2005; O’Neill 2014).

According to morphomes as laid out by Maiden, if some change to morphological structure occurs in one cell within a morphome, it should affect all other cells simultaneously. Crucially, rather than memorizing each individual form in a verb paradigm, a speaker is making an abstract pattern to help them memorize irregularity. Maiden says, “The diachronic maintenance of morphomes is the way in which that predictability is achieved if allomorphy is not levelled out. It might be seen as an alternative ‘Plan B’, if ‘Plan A’, namely levelling, should fail” (Maiden 2013, p. 519).

This concept is known as ‘coherence’, in which change that affects one member of the PYTA morphome affects all other members simultaneously. Through coherence, a change that is phonologically motivated in one cell of the verb spreads to all others, even without the prerequisite conditioning environment, causing more regularity within the verb paradigm, e.g., the alternation of e ∼ i in the stem of Old Spanish 1sg preterit fize and 2sg feziste levelling to i alone in Modern Spanish hice and hiciste (Maschi 2007).

As Maiden says, coherence is a type of analogical levelling process in which any change “that affect[s] any one of the paradigmatic cells implicated in the alternation pattern …always affect[s] all the others” (Maiden 2018, p. 3).

The use of corpus data can help elucidate the status of morphomes in linguistic theory, examining the theoretical basis via thorough historical examination, made possible with modern corpora (Rini 2020). To test Maiden’s morphomic theory, I undertake a corpus analysis of diachronic changes seen in Spanish by examining texts from the 12th to 17th century to assess the use of PYTA forms and to determine how consistently speakers of Old Spanish generalize the pattern of ‘coherence’ in the morphomic structure.

In this paper, I will first discuss Spanish strong verbs and the diachronic changes that occur within the PYTA morphome, these changes being the focus of the corpus analysis. I then provide a brief overview of the notion of the morphome in contemporary morphology, highlighting a couple key experiments that have been carried out with morphomes in Romance.

This is followed by a discussion of my methodology for the corpus analyses, and the results of these analyses, with both a holistic and individual author analysis undertaken. A discussion of both these analyses will come after presenting their results, finishing with some concluding remarks.

2 Spanish strong verbs

The verbs that belong to the PYTA morphome are those that have ‘strong’ forms in the preterit, imperfect subjunctives, and future subjunctive. ‘Strong’ verbs are those that have rhizotonic stress, i.e., those that have stress on the root of the verb (see 1a). ‘Weak’ verbs are those with arhizotonic stress, i.e., stress on the ending of the verb rather than the root (1b).

Modern Spanish verb infinitives with their strong preterit stems are given in Table 4.^[5]

Several other verbs that had strong preterits in Old Spanish have been regularized, e.g., Old Spanish vivir ‘to live’ had 1sg preterit visque, now modern Spanish viví ‘I lived’.

The reason for these verbs’ synchronic irregularity can be attributed to their inheritance of a specific form which was regular in Latin but becomes opaque after phonological changes, or which maintains irregularity already attested in Latin (2). Those forms which are most frequent are typically most resistant to regularization and retain their idiosyncrasy (Lieberman et al. 2007).

Curiously, there are a handful of Spanish verbs (andar ‘to walk’, tener ‘to have’, estar ‘to be’) whose modern preterit forms are strong, but which have no Latin antecedent (3).

Though Old Spanish estide is a phonologically expected outcome of Latin stetī, Modern Spanish estuve is not. The stem estuv-is said to have been built on analogy with the verb haber ‘to have’ (whose preterit stem is hub-) (Lloyd 1987), yet why a verb should look to haber rather than regularizing or maintaining their already strong forms is not well understood. Though the PYTA stem hub-of haber is synchronically irregular (4), it is a phonologically expected outcome from Latin through the process of metathesis and the raising of mid vowels from Old Spanish to Modern Spanish in verb forms (Blaylock 1965; Esher 2015; Lloyd 1987).

The same cannot be said for the PYTA forms of verbs estar and andar, which are thought to come from stetī and either reconstructed *andedi , a strong form innovated in the Latin of Spain and Italy (Fulk 1990; Meyer-Lübke 1895), or reconstructed *andāvī.^[6] No regular sound changes can cause *andedi or *andāvī to become Modern Spanish anduve and stetī to become Modern Spanish estuve. Proponents of morphomic theory state that this -uv- segment becomes a phonological segment that marks the preterit tense, and causes ‘convergence’, a process by which verb forms change to resemble other verb lexemes, such as the verbs andar and estar acquiring phonological information from haber (Esher 2015; Herce 2020; Maschi 2007). The verb tener ‘to have’ follows a similar process, with the Modern Spanish PYTA root as tuv-. However, unlike andar and estar, there is no attested form inherited from the Latin perfect, tenuī ( Lloyd 1987 ; Penny 2002 ). ^[7]

Alongside the analogical change caused by convergence, one other change that affects PYTA morphomic patterns is of particular note to this analysis: the alternation between mid-vowels in older forms and high vowels seen in modern forms caused by sound change, e.g. Old Spanish tove ‘I had’ and Modern Spanish tuve ‘I had’. This alternation is caused by the raising of middle vowels to high vowels in rhizotonic verb forms, with the mid-vowels e and o raised to i and u in preterit verbal stems, respectively. Old posieron ‘they put’ (poner) gives way to modern pusieron, old fezist ‘you did’ (hacer) to modern hiciste. This raising happens throughout the PYTA cells, even when there is no phonological motivation for it to raise, as an example of paradigmatic levelling (Blaylock 1965; Lloyd 1987). As Blaylock states, “This closed vowel was then extended to the rest of the paradigm of metaphonized verbs, and the pressure of the predominant pattern eventually forced the non-metaphonized forms into the mold” (Blaylock 1965, p. 263).

This raising of mid vowels is an example of the aforementioned concept of ‘coherence’, a powerful mechanism for the uniformity of verb morphology within the paradigm. With a newer verbal stem with a high vowel replacing the older preterit stem, a high vowel would then replace the older stems in every other member of the PYTA morphome, namely, the imperfect subjunctives and future subjunctive (Maschi 2007).

Though synchronically, coherence is ‘real’ in the context of Spanish morphology (there are no verbs that lack coherent stems in PYTA members), the question of whether coherence is psychologically real is more debatable, without any resolution on the matter. The use of morphomic theory to describe the allomorphy seen in Romance languages is hotly contested, and not without a number of detractors.

3 The psychological ‘reality’ of the Morphome

One persisting question that remains in the discussion of the morphome is its psychological reality (Luís and Bermúdez-Otero 2016; Maiden 2013, 2018). While some consider the concept of the morphome to be a revelation, others have referred to it as an idea that is ‘notoriously uncomfortable for many formal models of morphology’ (Herce 2020, p. 45) ‘undermines’ morphological study (Luís and Bermúdez-Otero 2016, p. 309), and understates factors important to diachronic language change such as phonologically conditioned allomorphy (Anderson 2011) and phonologically conditioned syncretism (Steriade 2016).

The idea of the morphome was first introduced by Mark Aronoff. As the progenitor of the concept of a morphome, Aronoff’s argument is that morphology has an exponence function that associates syntactic exponenda and arbitrary phonological exponents. Aronoff says that there is a purely morphological level of representation that is hidden, i.e., the morphomic level (Aronoff 1994).

According to Aronoff (1994, p. 28), the morphomic level exists between morphosyntax and morphophonology as an independent category, and so this morphomic level must be taken into account before reaching the level of phonology. However, some supporters of morphomes have proposed a framework which places them in a postsyntactic setting, interpretable using postsyntactic rules, underspecification, and prespecification with particular vocabulary items.^[8]

Past participles in Western European languages are often touted as models of the morphome in context (Aronoff 1994). Past participles have two functions: They are used both in perfect tenses and the passive voice. Though these are two different grammatical functions, they are typically expressed using the same form, e.g., It. Ho lavato il cuscino ‘I washed the cushion’, Il cuscino è stato lavato ‘The cushion is washed’. While these functions are not the same (passive and perfect), they use the same form, here, lavato ‘washed’.

Once a morphomic connection is established, as suggested for past participles, the passive and perfect become mentally linked, and if one changes, the other may be expected to change alongside it, as the forms ‘cohere’ to one another (Maschi 2007).

Another preeminent example used to validate the idea of the morphomic level is that of the ‘third-stem’ of Latin verbs, also known as the ‘supine’ or the ‘fourth principal part’. The third stem occurs in a set of highly heterogeneous forms. Examples of this third stem are showcased in Table 5.

These forms lack a single semantic or syntactic category (Luís and Bermúdez-Otero 2016). As such, Aronoff (1994) and Maiden (2013) say that this third stem is purely morphomic-its base lacks any semantic meaning. Maiden goes on to say that “the notion that speakers of Classical Latin were unconscious of a distributional reality which is valid for every single verb in their language… seems downright implausible” (Maiden 2013, p. 497). Yet given that there are no extant native speakers of Latin, it is impossible to test whether such a hypothesis is true. In other words, there is no way of knowing whether this morphome would have been ‘psychologically real’ for a Latin speaker.

In light of this, Maiden turns to modern Romance languages, paying special attention to Romanian. However, Maiden himself points out that the past participle (the descendant of the Latin third stem) does not always cohere in passive and perfect contexts, with some languages showing multiple forms, such as Portuguese perfect acendido and passive aceso ‘lit’ (acender), perfect prendido and passive preso ‘arrested’ (prender), perfect gastado and passive gasto ‘spent’ (gastar), etc. Compare, A Maria tem acendido muitas luzes. ‘Maria has been lighting many lights.’, and Muitas luzes foram acesas pela Maria. ‘Many lights were lit by Maria.’ Similar splits are seen in Sicilian dialects and Neapolitan (Maiden 2013). Spanish shows occasional deviation from this-ostensibly- consistent pattern, with forms such as frito/freído ‘fried’ (freír), and impreso/imprimido ‘printed’ (imprimir), e.g., he freído el tomate ‘I have fried the tomato’, and el tomate está frito ‘The tomato is fried.’

Much of Maiden’s argument for the third stem morphome hinges on the consistency between the Romanian past participle and supine, though the idea that the supine of Romanian is a direct continuant of the Latin supine is disputed (Maiden 2013). By showing the supposed uniformity between the supine and the past participle, Maiden stresses that the Latin third stem must be morphomic, and that the exceptions seen in Romanian, Portuguese, Neapolitan, and Sicilian (he does not mention those seen in Spanish) are due to semantic shift; one form of the past participle obtains an idiosyncratic meaning that is so far removed from the verb infinitive that the two forms are no longer thought to belong to the same lemma, and no longer belong to the same morphome. However, Maiden’s Portuguese examples do not show wide semantic differentiation between them. Indeed, he gives the same English translation for both the perfect and passive of the verb acender ‘to light’: acendido/aceso, ‘lit’. It is also unclear how semantically distant the lighting is in sentences such as ‘He had lit the candles’ and ‘The candles were lit’.

The idea of using the Modern Romanian supine forms to assess Latin morphomic patterns in the third stem is not without concerns, the most primary of which is the dubious connection between the Romanian and Latin supines (Maiden 2013). Yet for all the impact morphomes have had on the field of morphology and the lingering questions about their existence (psychological reality) in the mind of a speaker, surprisingly little research has been undertaken to verify their status.

Two notable exceptions to this lack of experimentation include Nevins et al. (2015) and Cappellaro et al. (2024). Nevins et al. concern themselves with the L-pattern morphome, a common pattern found in Romance verb paradigms. The L-pattern is characterized by a 1sg indicative form that serves as the base for the entire present subjunctive (Table 6).

Their research finds that this pattern is ‘underlearned’, and while generalizations such as the ‘L-pattern’ may be useful for grammarians, linguists, and second language learners, native speakers do not productively apply these patterns to novel verbs. Speakers of European Portuguese, Spanish, and Italian were unable to recreate the L-pattern, which was once ‘thriving’ in Romance, despite its ‘unnaturalness’. Nevins et al. say that in the 16th century, the pattern was just as unnatural but better represented in the lexicon, and therefore more salient. The researchers state that if the same task were run in the 16th century it is possible that participants would have generalized the L-pattern, a bold musing.

However, the more recent Cappellaro et al. (2024) finds the exact opposite results in Italian with the L/U-pattern when replicating and expanding upon Nevins et al.’s experiment. The authors state that the previous study by Nevins et al. (2015) is flawed, as it tested off-line rather than on-line production of morphomes. The authors attempt to fill this gap by measuring reaction times and doing an eye-tracking test, using native speakers of Italian. The authors also note that the morphophonological alternations used in the experiment conducted by Nevins et al. (2015) bear no resemblance to verbs seen in Romance, a flaw which does appear very serious indeed: ‘The paradigmatic distributional pattern is totally abstract and not conditioned by any extramorphological factors, even though extension to novel lexemes may be a concrete phonological replication of an existing pattern of alternation’ (Cappellaro et al. 2024, p. 41).

They point out that Nevins et al. (2015) also ignore inflection class in Italian, stating that the 1sg and 2sg in Italian are identical in the indicative for all verb classes (-o, -i, respectively). The L/U pattern extends across the entire present indicative and present subjunctive, and in Italian the 2 pl indicative forms are unambiguous for verb class (-ate, -ete, -ite for 1st, 2nd, and 3rd conjugation, respectively). They believe that Nevins et al. introduce bias in the experimental design by ignoring these factors.

Lastly, Cappellaro et al. believe that Nevins et al. do not use specific enough instructions in their experimental design. They believe that participants may not have recognized that two alternate morphophonological forms belonged to the same lexical item, and that the form-to-meaning connection may have been lost due to underspecific directions.

Cappellaro use both two-alternative forced choice (2AFC) and eye tracking methods measuring reaction time to assess the psychological reality of the morphome. In contrast to Nevins et al. (2015), they find that the L-morphome is a psychological reality in the minds of Italian speakers. They reach this conclusion via two principal findings. Firstly, speakers of Italian do extend the L/U pattern into novel words. The authors state that they find these results in contrast to Nevins et al. (2015) due to three main differences in the study’s design.

Speakers in Cappellaro et al. (2024) were forced to choose between only two options; instructions were explicit in that participants were told that there was only one lexical verb in question in the examples; and pseudo verbs were marked as not belonging to the first conjugation (-are). The 2AFC experiment showed a preference for target items that are consistent with L/U-patterns and faster decision-making when L/U-items are chosen. Eye-tracking measures found longer fixations on L/U patterns, indicating that the participants were “choosing” those forms over non-L/U patterns.

Maiden has acknowledged that on the surface, morphomic patterns make little sense in an extramorphological context (Maiden 2013). It is not clear to what extent speakers internalize morphomic patterns across a grammar, though Carstairs (1987) asserts this is the only plausible reality. Maiden considers diachronic innovation to be an opportunity to prove the reality of the morphome’s status in the mind of the speaker (Maiden 2005). If an innovated form is extended into the paradigm of a new lexeme, i.e., if morphomic abstraction is real, this pattern should behave ‘coherently’, i.e., affecting every cell of the pattern into which it has entered.

Strong forms of preterit verbs and PYTA verb paradigms are the perfect testing grounds for the psychological reality of morphomic patterns (Maiden 2004; O’Neill 2014), with both analogical change caused by convergence 1) andido > andovo ‘he/she went’ (andar) and morphological change (Rini 1999, p. 65) 2) andovo > anduvo ‘he/she went’. According to the principle of morphomic coherence, if one cell of the morphomic pattern is affected by a change, whether analogical or morphological, then all other members should be affected. The idea of morphomic coherence posits that it is unlikely one speaker would use the innovative preterit anduvo and the conservative future subjunctive andidiere concurrently, since younger generations are abstracting morphological generalizations rather than memorizing and transmitting individual forms. In Maiden’s own words, 3.

Maiden discusses the previously discussed raising of vowels as well, in which metaphony (where /e/ raises to /i/ and /o/ to /u/) should affect all other members of the PYTA morphome (Maiden 2001). O’Neill says, “the hallmark of the morphome seems to be identity of form across a semantically and syntactically heterogeneous set of cells in the inflectional paradigm” (O’Neill 2014, p. 75). Therefore, we should not expect to see the synonymous presence of preterit innovative v-stem estovo ‘He/she was’ (estar) and an imperfect subjunctive conservative d-stem estidiera ‘that he/she was’ (estar) in the grammar of one speaker.

The use of corpus analysis in the period when these innovative forms displace older forms could elucidate morphomic theory. If morphomic coherence requires all cells be affected simultaneously, we should expect each lexeme to have similar development across members of the PYTA group, regardless of which member is examined. Looking at data from the 13th to 17th centuries can provide an in-depth analysis of the historical development of modern PYTA verb members over time. In this study, the verbs andar ‘to go’ and estar ‘to be’ are examined for analogical replacement of d-stem forms (andid- and estid-) with modern v-stem forms (anduv- and estuv-), along with PYTA verb members which show a shift from mid to high vowels as discussed above. First, we will examine the quantitative analysis of the corpus as a whole.

4 Methodology

The corpus study was designed as a quantitative analysis of PYTA verb forms in Spanish. The verbs that were selected all fall under the category of ‘strong’ verbs, i.e., those with root-stress, and comprise (1) verbs which contain morphomic innovative forms in which a d-stem form is replaced by v-stem forms thought to have arisen from analogy with haber ‘to have’, namely, andar ‘to go’ estar ‘to be’, (2) verbs which show some diachronic change from /o/ to /u/ in the PYTA stem: poner ‘to put’, poder ‘to be able’, saber ‘to know’, caber ‘to fit’, haber ‘to have’, traer ^[9] ‘to bring’, or (3) verbs which have an analogical form from haber, but no d-stem form attested in Old Spanish: tener ‘to have’ (Lloyd 1987). Following Herce (2016), prefixed verb forms were not considered relevant to this study as it is unlikely that they are conceptualized by Spanish speakers as anything but the base verb paradigm with a prefix added to the front, given their morphological transparency.

Both the ‘Corpus diacrónico del Español’ (CORDE) and the ‘Corpus del Español’ were originally considered as viable options for this study, given the large amounts of historical data they possess. Ultimately, CORDE was chosen as it preserves paleographic spellings much more consistently and also provides concordance information necessary to eliminate homographs that are semantically unrelated. The Corpus del Español has also been criticized for its accuracy in a more general manner, with numerous chronological errors (Rini 2020).

After verb selection, the relevant forms of each verb were chosen, i.e., the PYTA forms of the verbs in both conservative and innovative forms. For example, for the PYTA forms of the verb andar, five stems were identified using Fulk (1990) as a benchmark: andid-, andod-, andud-, andov-, and anduv-. These forms were searched for in both their 3sg and 3 pl forms. Other persons and numbers were considered, but 1st and 2nd person forms occurred so infrequently in the corpus that they were ultimately excluded from the analysis. For example, only one case of andide was found in the entire corpus (from a book written in 1942 about historical Spanish linguistics) compared to 167 cases of 3sg andido and 94 cases of andidieron.

The years included in this analysis ranged from 1,200 to 1,699. The year 1,200 was chosen as a starting year as the 13th century is the first century in which there is a robust enough sample size to draw from, e.g., only one token of 3sg andido ‘he/she went’ was found in the two-century-period from 1,000 to 1,199 CE, while 144 tokens of andido were found in the single-century-period from 1,200 to 1,299 CE.

The year 1,699 was chosen as an end year for this analysis because the 17th century is the first century in which more conservative forms such as andido or andovo approach a frequency of zero for all lexemes analyzed and modern forms such as anduvo are used nearly 100 % of the time.

A high level of consideration was devoted to account for all spelling variants of a given form. Even forms completely unattested in Fulk (1990), Lloyd (1987), and Penny (2002) were searched to ensure that any hypothetical form caused by spelling variation was accounted for. As an example, the verb hubieron was searched as: hubieron, huvieron, huuieron, hubyeron, huvyeron, huuyeron, hvbieron, hvvieron, hvuieron, hvbyeron, hvvyeron, hvuyeron, ubieron, uvieron, uuieron, ubyeron, uvyeron, uuyeron, vbieron, vvieron, vuieron, vbyeron, vvyeron, and vuyeron, and then repeated using capital letters, which are considered a separate character in CORDE search tools.

Results were coded by lexeme, the century in which they occurred (CORDE allows searches to be conducted by year), their tense/mood (TAM) (preterit, imperfect subjunctive -se-, imperfect subjunctive -ra-, and future subjunctive), their stem class, in the case of andar and estar (d-stem e.g., andido or v-stem e.g., anduvo), their stem group (andid-, andod-, andud-, andov-, anduv-), and whether they were o-stem (andovo) or u-stem (anduvo) forms. These results were entered into R software version 2023.09.0+463, where they were analyzed, sorted, and filtered (R Core Team 2022). Proportions of stem class and stem groups were created in R by century, with 3sg and 3 pl forms analyzed both separately and together, to test for the effect of number, if any.

In the case of syncretic forms in the imperfect subjunctives e.g., anduviera ‘that he/she went’ which can be either 1sg or 3sg, concordance options were used to manually check whether the verb form in question was 1sg or 3sg. Any verbs that were in 1sg forms were not included in the total, although such instances of 1sg verb forms were extremely scarce.

One challenge encountered in this study was the difficulty in accounting for all spelling variants of a particular form. Though Fulk (1990) tries to list as many variants as possible, his focus is primarily on form rather than historical spelling variation. While I tried to be exhaustive in searching for different spellings, it may be the case that there are other historical spellings of the verbs analyzed in this paper or other extremely infrequent analogical forms that went undetected.

5 Results

Nearly half a million tokens were collected overall: the raw frequency is given in Table 7, with totals for century in bold in the bottom row, and totals for verb lexeme across centuries in bold in the rightmost column.

A quantitative analysis of the corpus results indicates that for many verbs, there is a disparate rate of development from conservative to innovative forms. These numbers are particularly pronounced in the 14th and 15th centuries for several verbs. Both the verbs estar ‘to be’ and andar ‘to walk’ (Figure 1) show innovative preterit forms in the 1300s and conservative future subjunctive forms in the 1300s. Moreover, estar shows higher innovative forms in general in earlier centuries (Figure 2).

Figure 1:

Andar V-stems by TAM.

Figure 2:

Estar V-stems by TAM.

The overall shape of these two verbs mirrors one another in the 14th century, with preterit innovative rates far exceeding those of the future subjunctive. The future subjunctive appears to be the least innovative form in the 14th century in other verbs as well. Where the verbs andar and estar show analogical change with the addition of -uv-, the other verbs surveyed show morphological change alone in raising from mid to high vowels. This innovative preterit in the 14th century is seen in other verbs as well, with an innovative preterit seen for the verb saber as well(Figure 3).

Figure 3:

Saber u-stem proportions by TAM.

A Poisson regression analysis was conducted in R to examine the occurrences of o-stem and u-stem forms across all verbs that contrast solely o-stems and u-stems. Multiple occurrences of statistical significance between TAM and Year were found (p < 0.05). In order to assess the overall impact of TAM, an analysis of deviance was done using the ANOVA function in R, applied to a chi-square test. This test found that TAM was statistically significant (p < 0.05), and that the interaction between TAM, Year, and Stem type was also statistically significant (p < 0.05) (For more information, see Table A1 and A2 in the appendix).

Such results, in which the TAM is statistically significant in the overall model, indicate that different tenses can affect the development of verb morphology at different rates, even within a morphomic set. However, one unfortunate possibility remains that the disparate rates seen across time may be explained due to differences in the internal grammar of an individual. Though we see that the verb saber ‘to know’ had u-stem proportions of 11.63 % for the future subjunctive in the 14th century, and preterit u-stem proportions of 54.53 % for the same century, it is possible that there is no variation at all on an interspeaker basis, and that it just so happens that nearly half of people use an older o-stem form (sopo) in all cases whereas the other half use a newer form (supo) in all cases. Moreover, it is possible that those who used o-stem forms just happened to use the future subjunctive much more often in their texts, accounting for the disparate rates seen across centuries. Perhaps there truly is no mismatch in the internal grammar of a speaker, and they use one form across the entire PYTA morphome, never alternating.

6 Individual analysis

To examine such an idea, a second corpus analysis was performed looking at the data from individuals. Authors were chosen based on the availability of material and the time period to which they belonged, so as to have a balanced sample that spans from 1,200 to 1,600 at regular intervals. When this process was completed, 14 authors in total (Table 8) were selected.

After author selection was completed, the verbs to be analyzed were chosen. Although initially the same verbs were considered as those used in the previous analysis, in many cases verbs such as caber ‘to fit’ or traer ‘to carry’, had very few occurrences in PYTA forms among these authors. Only four verbs had sufficient appearances across authors to facilitate a viable analysis, namely, estar ‘to be’, andar ‘to go’, saber ‘to know’ and tener ‘to have’. Similarly to the first analysis, only 3rd person forms showed sufficient occurrences for worthwhile comparison, and as such, only 3sg and 3 pl forms were ultimately examined within the corpus.

For the verbs andar and estar, six stem forms were searched, the inherited -id- stem, (andido, estido), the later -ud- stem, (andudo, estudo), the marginal -od- form, (andodo, estodo), the -ov- stem, (andovo, estovo), the modern -uv- stem, (anduvo, estuvo), and weak stems (e.g., andó, estó).

In the case of saber and tener, these verbs do not have d-stem forms from Latin but do both show o and u stems. As such, three stem forms were searched for across these two verbs: o stems, (sopo, tovo), u stems, (supo, tuvo), and innovative regularized stems, (sabió, tenió).

Given the large number of spelling variations across Old Spanish texts, verb forms were searched for in several ways, with a verb such as tuvo searched for in the corpus using tuvo, tuuo, tubo, Tuvo, Tuuo, and Tubo, as uppercase letters are treated differently than lowercase letters by the CORDE search engine. The frequency of these stem forms for each member of the PYTA morphome was manually recorded form by form by author name, after which the results were analyzed.

7 Secondary results

The results of the secondary analysis suggest that few authors utilize morphomic coherence in their works. Specifically, the first question concerns the idea of coherence overall, i.e., does a speaker use one stem consistently for all cells of a particular verb lexeme?

Shows these results; the column ‘Total Coherence’ refers to whether a speaker used one consistent stem within all forms of one particular lexeme. The second column displays whether a speaker was coherent (used only one stem) for at least one of the four verbs chosen. If an author was coherent within a certain verb, that verb is listed in parentheses afterward. The results are given as Yes/No form in Table 9.

Across all 14 authors surveyed, none used a consistent stem form for all verbs analyzed. Among the 56 different verb uses examined (14 authors, 4 different verbs) only 3 instances showed cohesion within a verb lexeme, two instances of tener, and one instance of saber. Among the two tener examples, one coherent pattern comes from 14th century author Fernán Sánchez de Valladolid, in which he only uses o-stem forms (e.g., tovo), although he does use many u-stem forms in the verb saber (e.g., supo). The other author who uses tener with only one stem is 16th century author Martín Pérez de Ayala, who similarly uses only o-stem forms for tener. The last instance of coherence is by 16th century author Gonzalo Fernández de Oviedo y Valdés, who uses only u-stem forms for the verb saber.

One may suggest that perhaps the authors’ internalized grammars are changing over time, and that when they do eventually adopt an incumbent phonological or morphological change to their idiolect, they then extend it to all forms, coinciding with the idea of morphomic coherence as proposed by Maiden. Yet, there are no indications from the corpus analysis that this is the case and conversely, authors alternated between stems in a seemingly random fashion. For example, in his translation of the Aeneid, written in 1,427–1,428, Enrique de Villena uses 3 pl forms of estar such as estodieron, estudieron, estovieron, and estuvieron.

With these data, we are then left with two possibilities: 1) Enrique de Villena is behaving in a morphomically coherent manner, and adopts innovative forms of the verb estar rapidly, adopting three new forms within the span of approximately one year, or 2) Morphomic coherence is not being adhered to by de Villena, that there is a large level of variation for him, and he alternates between different forms as he desires.^[10]

8 Discussion

There are few if any generalizations that can be made about stem patterns as they related to tense, mood, or number. If a speaker is using different stems in different verbs across different distributions, and crucially, is not using the same stem within the same verb, how can morphomes be said to be an abstract template for predictability?

The data collected from the CORDE corpus indicate that coherence is diachronically complicated. The overall analysis finds wide levels of disparity within centuries. Moreover, authors rarely used verbal forms consistently across the PYTA morphome, with only three authors using coherent stem forms and only for one verb each. Given that these stems seem to change differently based upon their tense/mood, it may be useful to invoke other reasons for this change over time. Frequency could be a key factor in explaining why PYTA stem forms underwent analogy at different rates. Previous research has found that high-frequency words tend to undergo sound change first (Hooper 1976) whereas less frequent words tend to undergo analogical change more often diachronically (Lieberman et al. 2007).

In both the overall corpus analysis and the individual author analysis, the preterit appears to innovate the earliest, whereas the future subjunctive appears to innovate the latest. We may expect then that the preterit is the most frequently occurring of these forms. The data collected in this study corroborate that notion. In the second analysis, preterit forms for estar are both the most frequently occurring, and the stem most used in the preterits is the uv-stem, percentage wise, when examining singular and plural forms separately. Conversely, the imperfect subjunctive -ra- stem forms are the least frequently occurring and have the lowest percentage of uv stems from overall usage (Table 10).^[11]

Two Fischer’s exact tests were conducted to assess the statistical significance between stem-type and the number of occurrences in the verb estar. In the first test, singular and plural forms were kept separate, as shown in Table 10. In the second version singular and plural forms were combined by TAM. In both cases, p-values found a statistically significant relationship between TAM and stem type (p < 0.05).

These trends may be further reflected by looking at authors on an individual level. For those who use both o and u forms in the 13th and 14th centuries, there appears to be an emergent pattern in which the innovative u form is used more often (sometimes exclusively) in the preterit but in no other member of the PYTA morphome pattern. For example, the works of Alfonso X’s scriptorium, were the earliest works analyzed and use very few u forms in general. However, as Table 11 displays, u forms are used sparingly in the preterit sg. and pl. with the verb saber, though in no other tense.

This pattern, in which an innovative form is used only in the preterit and no other tense, is repeated numerous times. Consider the data from Juan Manuel, in which only the preterit of estar uses innovative ov stem forms, whereas other tenses use older d-stem forms (id, od, ud). Likewise, Manuel’s use of saber u-stems aligns with that of Alfonso X’s scriptorium. These trends are reflected in Table 12 and Table 13, respectively.

Such examples can be seen even within the same edition of a work, take for example, Juan Manuel’s alternating vowels in the use of preterit sg. saber, seen in close proximity to one another.

Such a pattern is not restricted to only the earlier centuries of this corpus research and is seen in 15th century as well, with writers García de Salazar and Alfonso Martinez de Toledo’s use of tener, where u-stem forms are reserved for the preterit, as displayed in Table 14 and Table 15.

These verbs do not show coherence, with certain tenses adopting innovative forms at earlier points in time. More curiously, the trends seen for verb stem choices are not consistent even when comparing forms used by the same author. For example, while author Hernando del Pulgar strongly prefers u-stem forms in the verb saber, this is not consistent with his other verb form patterns, and he strongly prefers o-stem forms with the verb tener. These data indicating u-stem percentages are presented in Table 16 and Table 17.

These tables indicate that even general trends in PYTA stem forms are not generalizable across different verb forms, even those used by one author.

One final observation relates to the use of regular forms. Although the PYTA morphome was said to be extant before the diversification of the Proto-Romance branch, writers show varying use of both weak (verb stress on the endings) and strong (verb stress on the verb stem) verb forms. The largest user of weak forms among this corpus is Fernández de Heredia. Examples of his weak forms with the verb andar are shown in examples 11 and 12.

Yet not all his uses of pret pl. andar are weak, as shown in example 13.

While Heredia uses 60 different forms of andar in total, 47 of those forms are weak, with and(a)- as the stem, rather than andid-, anduv-, etc. Moreover, though Heredia uses andar in the preterit sg. ten times, he does not use any weak forms, yet he uses andar in the preterit pl 11 times, and ten of those occurrences are weak forms (Table 18).

This use of weak forms and strong forms is highly disparate, even within the same tense (preterit). An analysis using four Fischer’s exact tests finds statistical significance between stem and tense, both when combining singular and plural forms and leaving them separate, and both when leaving weak forms in or omitting them (p < 0.05). It would be expected that a native speaker of Spanish would understand the connection between the preterit 3sg and preterit 3 pl form, but despite this, the verb forms are different. Compare example 14 to example 11, (Caualgaron et andaron toda noche … ) from the same piece of writing.

9 Conclusions

How ‘real’ a morphome is psychologically speaking remains a large question. So far, very little research has been undertaken to fill this gap, and discussions of morphomes remain largely theoretical rather than experimental.

The corpus analysis undertaken in this paper looked at PYTA verb tenses in two analyses: The first of these was an overall quantitative analysis in which highly disparate rates of innovation can be seen across the different tenses of the PYTA morphome. In order to examine the possible bias of such an analysis, a more qualitative individual analysis was conducted for four frequently occurring verbs: estar, andar, saber, and tener as used by 14 different writers in the 12th-16th centuries. Although morphomic theory posits that a speaker will use forms coherently across different cells of the morphome, the evidence collected indicates that romance verb allomorphy may be more multifaceted than morphomic theory can account for.

The disparate rates of innovation were particularly pronounced in the 14th century, with the preterit often displaying the highest rates of innovation, and the future subjunctive with the lowest. This may be due to frequency effects, with the preterit the most frequent of the four members of the PYTA morphome.

Among the authors and verbs examined, it is challenging to find clear patterns of consistent stem usage. Only three authors showed coherence in PYTA cells, and even then, those three authors showed coherence only in one of the four verbs selected.

This suggests that the evolution from Latin to Old Spanish was marked by diverse factors that may intersect with morphology, but that cannot be explained through morphomic abstraction alone.

Similarly, PYTA morphomes appear to be unrecognized or non-abstractable in Old Spanish. The role of morphomes in understanding language change may need fine-tuning in its interaction with diachronic variation.

This study sheds light on the complexity of language evolution and the wide amount of variation that can exist in older forms of a language during the time before standardization. Such variation poses a challenge in establishing that there is a neat morphomic template that speakers adhere to. If PYTA patterns showed so much unpredictable allomorphy in Old Spanish, that might then raise question about other proposed morphomes from even older periods, such as the Latin third stem.

Detractors of morphomic theory opine that it undermines the search for linguistic truth (Luís and Bermúdez-Otero 2016). This study may open up opportunities for further research, for those who support or contend with morphomic theory. Future research may consider factors such as frequency, transmission, diatopic variation, and phonetics as potential causes of this uneven diachronic innovation rate in Spanish PYTA forms.