The Galician definite article is an unstressed function word that determines nouns and nominalized units (Álvarez & Xove, 2002). Its morphophonemic intricacies have attracted the interest of numerous scholars: Álvarez Blanco (1983) and Dubert (2004) working in classic Structuralism, Dubert (2001) in Lexical Phonology, Fernández Rei (2002, 2005, 2012) in Lexical and Prosodic Phonology, and Kikuchi (2006) in Optimality Theory. As Kikuchi has posited (2006, pp. 41–42), the present-day Galician definite article has three allomorphs /o, lo, no/ (see (1)) conditioned by the ending of the preceding word, although the article and the preceding word are not syntactically immediate constituents.

The allomorph /lo/ is selected when the preceding word ends in /ɾ/ or /s/; after selecting /lo/, the trigger of the selection is deleted, so the presence of /lo/ becomes opaque; the allomorph /no/ is selected when the preceding word ends in a nasal and it is manifested by the alveolar feature of the nasal; the allomorph /o/ is selected elsewhere (see Section 2.2). Evidence shows that the selection of the allomorphs is made in the syntax, it is phonologically conditioned and triggers morphophonemic alternations. Dubert (2001) therefore introduces the current definite article in what Kaisse (1985) calls the postlexical 1 component; Fernández Rei explains it as the outcome of precompiled phonological rules (Hayes, 1990).

This article will study the linguistic conditions that made possible the phonological changes that created the present-day alternations. I will assume that the linguistic conditions reflected the grammar at the time when the changes took place. The formalizations needed to describe the current alternations explain neither the changes, nor the conditions existing when the changes took place. This paper is historical since it describes changes and old states of language, but it is also synchronic since it tries to describe the original conditions that triggered the changes. Thus, my concern will be to examine the processes and conditions that produced the present-day alternations, not the present-day alternations per se.

The conditions of the present-day alternations differ somewhat between the standard written variety and vernacular dialects. The contexts of the alternations in the former are an arbitrary subset of those typical of the latter, which are lexically less restricted. The data for contemporary Galician are taken from the dialect of Santiago de Compostela (Dubert, 1998, 1999), from my own annotations as a dialectologist and from the Atlas lingüístico galego. Vol. II. Morfología non verbal (ALGa2; García, Santamarina, Álvarez Blanco, Fernández Rei, & González González, 1995).

In Section 2 I present the properties of the present-day allomorphs of the article and their alternations; in Section 3 I present the origins of the alternations in Old Galician; in Section 4 I analyze problematic consonant sequences; in Section 5 the data are discussed in the light of usage based premises; finally, in Section 6 I present some conclusions.

I will follow the following graphic conventions: (.) syllable boundary, (#) morpheme boundary, (##) word boundary, (=) clitic boundary, (ω) phonological word, (&) contracted element, (V) any vowel, (C) any consonant, (X) any segment. Small capital letters will be used in glosses and double quotation marks (“ ”) with translations.

In this section I will present the present-day Galician definite article, its allomorphs and their distribution; I will also classify the clitic based on its properties.

In its transformation into an article, beside stress Lat. illum, illam lost its initial vowel: illum > lu > lo. In the course of the ninth and tenth centuries Galician lost Latin word-internal intervocalic /l/, e.g., Lat. uolare > voar “fly” (Williams, 1962, pp. 68–74). When it appeared in intervocalic position (6a), the initial /l/ of the article lo underwent the same processes as word-internal intervocalic /l/s (6b) and was also lost. This development affected both the definite article and the 3.acc clitic. Other words starting with /l/ (6c) were not affected, despite the similar context:

Note that at the time when /l/ dropped the article (6a) must have had the aforementioned special phonological relationship with the preceding word, since its /l/ was treated as a word-internal intervocalic /l/ (6b): thus, prosodic and syntactic bracketing didn’t match. In (7) the square brackets [ ] represent phonological units, while the curly brackets { } represent syntactic units.

However, the /l/ of the definite article assimilated a preceding /ɾ/ and /s/ (8a, b): come/ɾlo/ pan > come/lːo/ pan eat-inf the bread “to eat the bread”; come/slo/ pan > come/lːo/ pan eat-2sg the bread “you eat the bread”. The /l/ of the definite article assimilated to a preceding nasal (8c): come/nlo/ pan > come/nːo/ pan eat-3sg the bread “they eat the bread”. Long /lː/ and /nː/ were later simplified to /l/ and /n/, respectively (Ferreiro, 1995; Williams, 1962). This was not the case in (8d, e, f), where no assimilation has taken place. Here the word-internal clusters /sl/ and /nl/ remained unchanged, while the word internal /ɾl/ either remained unchanged or underwent metathesis:

Williams (1962) was aware of this situation and speaks about an intertonic context:

Speaking of medieval varieties of (Galician-) Portuguese, Mattos e Silva (1989) also says regarding 2FA:

One may wonder why the /l/ in intervocalic position of both the definite article and the 3.acc clitic behaved like a word-internal intervocalic /l/ (como lo pan > como o pan like moler > moer), yet when preceded by a consonant it behaved differently (comer lo > comel lo > come-lo, but merlo > melro ~ merlo). Why were there assimilations only in the cases of (8a, b, c), but not in the others? Note that in (7) /l/ behaved as if word-internal, but in (8) the word-internal /l/ of merlo, garlopa, canle behaved differently than the “word-internal” /l/ in comen lo, comer lo, comes lo. Secondly, in /n.l/, why was the assimilation progressive, /n.l/ > /nː/, while in /s.l/ and /ɾ.l/ it was regressive, /s.l/ and /ɾ.l/ > /lː/?

In this study I will only deal with the origins of the alternations referred to here as 2FA and 3FA in (2) and (8), but it was necessary to mention all these facts about intervocalic 1FA in (7) to show that the definite article had special, phonological relationships with the words that preceded it. These phonological relations are not only a feature of contemporary Galician but must already have been present in the present-day alternations originated. Let us now consider the nature of the phonological relations that ended up producing 2FA and 3FA in present-day Galician. First, I will present some strong tendencies in syllable structure; then, I will look at the Galician clusters /ɾ.l/, /s.l/, /n.l/ in the light of those laws.

As the outcome of a regular phonetic law, Latin /l/ and /n/ might have disappeared anywhere in intervocalic position, in which case the loss of intervocalic /l/ and /n/ would have occurred not only within words but also between words. Instead, word-initial or word-final /l/ and /n/ were kept: este lado “this side”, esta nube “this cloud”, non é “it isn’t”. In terms of current prosodic phonology, the elision of /l/ and /n/ was applied within a certain prosodic domain, probably the phonological word. If so, elision may have been the outcome of a domain span rule (Nespor & Vogel, 2007, p. 15) such as (15):

Since the /l/ of the definite article was lost when preceded by a vowel, we may assume that (15) also refers to environments containing the article; since the phonological word is a domain established on phonological grounds, and (15) operated within it, it would seem that the elision rule must be postlexical. All we need is to formulate the algorithm that defines the prosodic word in such a way that it includes the article (for present-day Galician, see Fernández Rei, 2002).

However, if we assume that every definite article constitutes a phonological word in conjunction with its preceding host, we have seen that, in the case of /s.l/, /n.l/ and /ɾ.l/, different changes took place depending on whether /l/ was part of the definite article (where assimilation occurred) or not (where there was no assimilation). It seems odd that, where the article is concerned, /V##l/ should have constituted part of one prosodic domain but /C##l/, a different one.

Intervocalic deletion or postconsonantal assimilation occurred even when the article was preceded by unstressed function words, giving rise to the contractions in (4):

Thus, [a ## lo], [de ## lo], [en ## lo] must form part of the same prosodic domain as [como ## lo], [comen ## lo], [comer ## lo], where the article was preceded by a stressed word.

Nespor and Vogel (2007, pp. 27–60) make a distinction between two kinds of rules: those which operate only with phonological information (prosodic domains are phonological information) and those which depend on morphological or syntactic information. Both kinds of rules operate both inside and between words: “A distinction must be drawn between those phonological processes that are to be handled by purely phonological rules and those that are to be handled by rules that also take other than phonological information into account” (p. 33).

They only consider processes of the first kind to belong to Prosodic Phonology. The early assimilations /s.l/, /ɾ.l/ > /lː/ and /n.l/ > /nː/ seem to have been morphosyntactically conditioned from the start. This is not surprising since the definite article and the 3.acc clitic are the only unstressed monosyllabic word forms starting with /l/. Thus, according to Nespor & Vogel (2007), the rules responsible for the postconsonantal assimilations must be included in a component other than Prosodic Phonology and hence in a different component from the intervocalic elision of come lo pan or de lo pan. However, we saw that the changes that occurred in the context of the postconsonantal definite article had a clear phonetic motivation based on strength/sonority scales and optimal syllabic configurations: this appears to be what Pierrehumbert (2002) calls word-specific phonetics, but across syntactic boundaries. Locating the assimilation rules in a different component does not seem to explain why the assimilations took place only with definite articles (and the 3.acc clitics). The special component in which these special assimilation rules need to be located is an outcome of the assimilation, not an explanation for the change; indeed, the reason why we need this special component is due to the morphological conditioning of the assimilations.

For the present-day morphology of the Galician article Dubert (2001), Fernández Rei (2002), and Kikuchi (2006) propose an allomorph /lo, la, los, las/, i.e., 2FA, which is selected by the ending of the preceding word; once 2FA is selected, the consonants /s/ and /ɾ/ drop out: comer lo pan > come-lo pan. Fernández Rei and Dubert think that the elision is morphophonemically conditioned. Kikuchi (pp. 45–46) attempts to make the rule more phonological by suggesting that words such as desleal “disloyal” or merlo “blackbird” keep the groups /s.l/ and /ɾ.l/ in present-day Galician because Domain-Contiguity (the faithfulness constraint against morpheme-internal deletion) is ranked higher than OCP[+cont] (an OCP constraint that impedes a sequence of two continuous segments). In the case of 2FA, however, the faithfulness constraint MaxC is ranked lower than OCP[+cont], producing the elision of /s/ and /ɾ/.8 His phonology, however, needs to recognise the morphological boundaries of words. In much the same way, if we were to apply the same reasoning to the time at which the old assimilations took place, we might say that the faithfulness constraints blocked the change in desleal, but not in comer lo pan; in the case of the article, the constraints based on strength produced the assimilation because they were ranked higher than the faithfulness constraints. But again, why was this the case? Why were the faithfulness constraints stronger inside the morphological word but weaker when the /l/ of the article was involved? In islam, merlo and fontenla, /s.l/, /ɾ.l/, and /n.l/ are permanently stored in the lexicon, continuously violating certain constraints and containing non-optimal lexical syllabic configurations, while in the case of the article, the sequences /s.l/, /ɾ.l/, and /n.l/ must be constructed on the fly each time they are needed, disappearing just after its use (at least in a generative framework where the constructions with the article should not be memorized, but only exist momentarily, when used).

Probably, one of the explanations of the assimilations comes from the fact that the frequency of use of the sequences /s.l/, /ɾ.l/ and /n.l/ when /l/ forms part of the definite article is much greater than the frequency of use of these sequences in word-form-internal position: the use of the construction [Verb-/s##l/-Def-Art] is overwhelmingly more common than the individual use of isolated words like esluír, desligar, eslora, which contain the same sequences. Most word forms corresponding to the 2sg, 1pl and 2pl of verbs end in /s/; most noun phrases begin with a definite article. To the frequency of [Verb-/s##l/-Def-Art] we may add other common constructions such as those in (16):

All these constructions may be reduced to a more abstract one, [/s##l/-Def-Art], which is even more usual (in fact, [Verb-/s##l/-Def-Art] is a more specific instantiation of both [/s##l/-Def-Art] and [Verb /l/-Def-Art]). It seems that it is frequency of use that helps to maintain the contraction polo(s), pola(s) by=the-m, by-the-f “by the” and the sequence tódolos, tódalas all=the-m.pl, all=the-f.pl “all the” which are today probably lexicalized (see Section 2.2), while variation between comer o pan and come-lo pan is possible.

We do not have lemmatized corpora for either medieval or contemporary spoken Galician; thus, it is very difficult to obtain reliable statistics. Bugarín López (2007)9 is a frequencies dictionary that can be used as an approximation. Some figures are presented in Table 2.

The frequency of /s.l/, /ɾ.l/, and /n.l/ where /l/ belonged to the definite article caused these sequences to be adapted to the syllabic restrictions described by Vennemann. This could explain why even though we find /ɾ.l/ in the same prosodic domain in comer lo (pan) and merlo, we have different outcomes. The high ranking of Domain-Contiguity in the word internal sequences might have derived from low frequency of use; the high ranking of constraints based on syllabic structure and strength in the case of the definite article might have been the outcome of high frequency of use.

It is known that frequency of use is a determining factor in the organization and reorganization of grammar, as Bybee (2001, 2010) shows. Grammar and grammatical structures emerge from use. Frequency and adjacency are powerful explanatory tools to account for the historical development of the morphology of the Galician definite article. Units that tend to appear together may end up forming new, more complex units (this is the origin of the contractions in (4)). Strictly linguistic notions such as domains, contexts and strength may not suffice to explain the structure of the language, or at least to account for the phonological changes we have witnessed.

In the usage based model, it is assumed that every time a linguistic item is used it leaves a trace in the speaker’s memory (Bybee, 2001; Langacker, 2000; Pierrehumbert, 2002): the more frequent the use, the stronger the trace. If the constructions [/s##l/-Def-Art], [/ɾ##l/-Def-Art], and [/n##l/-Def-Art] were frequent enough, they might have been memorized and automatized as chunks, in such a way that they may have been treated as units and adapted to the pressures of the strength scales.

On the other hand, according to Venneman (1988) the assimilations found in /s.l/ and /n.l/ should have favoured the weaker segment, so that both sequences ought to have merged into /lː/; this is true of /s.l/ (comes lo > comelo), but not of /n.l/, where the stronger segment assimilated the weaker, giving /nː/ (comen lo > comen no > comen-o). However, there exists a clear association between the syntactic label Verb, the morphosyntactic property 3pl and a nasal segment at the end of the word form of 3pl verbs; this could have constituted a schema that strongly links all these features (Langacker, 2000; Taylor, 2002). Besides, it should be noted that indices of frequency of use of 3pl forms are high; the common use of 3pl, expressed with [+nasal], reinforces the link between the properties and the nasal feature, through the lexical connections that comen lo establishes with comen. This could be another expression of faithfulness constraints, which may be the combined results of both frequency and the analogical pressures of strong schemata like this one. That may be the reason why the less optimal option was selected regarding the direction of assimilation in /n.l/: it seemed preferable to maintain the expression with a nasal segment of the 3pl. Furthermore, if /s.l/ and /n.l/ had both developed to /lː/, the word forms of the 2sg and 3pl of verbs would have become homophones.

In this section I have been using concepts such as constructions and schemata, taken from a cognitive, functionalist, non-derivative model of grammar. This model is not syntacticocentric in the sense of Jackendoff (2002), and does not allow two processes of lexical insertion in a syntactic structure previously designed by the syntax, or an opposition between surface and underlying forms; hence neither rules nor constraints are needed to link the two levels. In this model, schemata and constructions freely combine information from different components (phonology, syntax and semantics), so different levels of phonology are not necessary: some schemata contain only phonological information, while others contain information from other components of the grammar. The model naturally incorporates pure phonological processes and morphosyntactically conditioned phonological processes, without needing to distinguish between lexical, postlexical 1 and postlexical 2 levels. In fact, the model was conceived to incorporate information about syntax, semantics and phonology in a single schema. The contexts that block 2FA in present-day Galician (described meticulously by Fernández Rei, 2002, pp. 223–232) show how difficult it is to deal with these exceptions without referencing syntactic information.

In this article I have discussed the linguistic conditions which allowed the phonological changes that created morphological alternations in the Galician definite article. I have assumed that these linguistic conditions reflected the grammar at the time when the changes took place. The definite article has three allomorphs whose selection depends on the final segment of the preceding word. The current Galician definite article resembles a simple clitic in that it may attach to different word classes and its position is syntactically determined; however, like an affix, it has no stressed allomorphs and its phonological form may be affected by the host. It seems to be a ditropic clitic, phonologically attached to the preceding word but maintaining direct syntactic relations with the following words. I have set out to describe the linguistic state in which those allomorphs developed in Old Galician, between the ninth and tenth centuries.

The three allomorphs are the historical outcomes of phonetic assimilations; these assimilations were word-specific, since they only affected the definite article and the 3.acc clitic. In order to explain them, we are forced to contemplate interaction between purely linguistic notions such as syllabic structure, segmental changes, scales of segmental strength, word classes, schemata, constructions, etc., and the frequency of use of the linguistic items involved in the assimilations. After presenting some of Vennemann’s laws regarding syllabic structure, I propose that an explanation using an exemplar model should be possible, since this model is designed to combine in a single scheme or construction precise phonetic, morphosyntatic and semantic information and to permit interaction between different units stored in memory, which in generative models have to be created “on the fly”.