PhiN 29/2004: 40

Richard Waltereit (Tübingen)

Fossils of a former phonological rule:
Irregular raddoppiamento fonosintattico^*

Raddoppiamento fonosintattico (RS) is a sandhi phenomenon of Italian with the effect of geminating initial consonants after certain words (the RS-triggers). There is an open and regular class of RS-triggers, which consists of all Italian oxytones, and a small, closed class of irregular RFS-triggers. Irregular RFS is triggered mainly by many (but not all) unstressed monosyllables. According to a widespread view, irregular RS is the direct residue of Latin consonant assimilation. Then there should be no irregular RS-triggers whose Latin etymon ended in a vowel. There are some, however. Even more puzzling is the fact that in Old Italian texts, whose spelling often reflects RS, there are even more irregular RS-triggers with etyma ending in a vowel. An alternative solution is to claim that in Old Tuscan RS was triggered not only by word stress but by phrasal stress too. This was the case especially in sequences of phonological clitics, where every first clitic could be stressed in order to comply with foot structure. Loss of rhythm-driven RS in the history of Italian resulted in a fossilization of RS with some of these phonological clitics, which incidentally are today's irregular RS-triggers. That rhythm-driven RS was fossilized with some words instead of simply disappearing is related to the fact that these words belong to the most frequent ones of the language, and obsolescent external sandhi may remain active in words with high token frequency (cf. Bybee 2001).

1 The problem of raddoppiamento fonosintattico (RS)

Raddoppiamento fonosintattico (henceforth RS) is a popular subject of study in Italian phonology. Essentially, RS is a phenomenon of external sandhi with the effect of geminating initial consonants after certain words (the RS-triggers). With respect to the choice of these words, there exists a regular, open and productive class of RS-triggers and a small, closed class of irregular RS-triggers. In this paper, I want to reassess the relationship between the two classes and to explain the composition of the irregular class.

1.1 Regular RS: Following all oxytones

The class of regular RS-triggers includes all Italian oxytones. All oxytones trigger gemination of the following word's initial consonant, whereas words that do not have main stress on the ultima do not:

Puzzlingly, however, RS exists in another (and much less studied) class of cases that seem to resist generalization, which I will henceforth refer to as irregular RS.

1.2. Irregular RS: after some unstressed monosyllables and some paroxytones

Irregular RS is triggered mainly by many (but not all) unstressed monosyllables:

Other word-forms that belong to this class are:

Unstressed monosyllables that do not trigger RS include the preposition di ‚of' and the clitic pronouns:

Furthermore, some (very few) disyllabic paroxytones trigger RS:

These are the essential data for Standard Italian (cf. Camilli 1965). The Italian dialects, however, show a slightly different picture. First, RS is regionally restricted to the central and southern dialects. Second, regular, i.e. stress-driven RS is restricted to Tuscan dialects. Third, and most importantly for the purposes of this paper, the set of irregular RS-triggers varies greatly from dialect to dialect, to the extent that, according to Weinrich (1958: 52), every city has its own list of RS-triggers. For example, da 'from' is an RS-trigger in the standard language, but not in some dialects (cf. Rohlfs 1949: 290–3). Ample information on dialectal variation is provided by Fanciullo 1986 and Canepari 1991. Varied as they may be, the set of irregular RS-triggers obeys consistently the following constraint across dialects: they are short, mostly monosyllabic, words.

RS is not reflected in the orthography of Modern Italian, except in fixed compounds:

Regular RS has been receiving much attention in the literature of the last decades and many of its aspects are now quite well understood. Less clear, however, are the following problems that will be dealt with in this article:

What exactly is the relationship between regular and irregular RS?

Why is it that irregular RS-triggers are mostly unstressed monosyllables?

I will provide a historical answer to both questions. In particular, I will try to give evidence for the claim that irregular RS is the fossilized residue of a no longer productive Old Italian stress rule. In doing so, I want to show that Loporcaro's (1997) claim that irregular RS is a direct successor of Latin consonant assimilation needs some revision.

2 When is a monosyllable a regular or an irregular RS-trigger?

A remark is in order concerning the size of the set of unstressed monosyllables (5). Previous accounts of RS vary as to their definition of what an unstressed monosyllable is. For example, depending on whether ha 's/he has' is regarded as unstressed or stressed, it counts as regular or irregular RS. It is indeed not always easy to decide whether a monosyllabic word is stressed or not. If the monosyllable is a content word, like re 'king', it can reasonably be considered as bearing stress. If the monosyllable is a function word, like di 'of', it is arguably unstressed (cf. Hayes 1995: 24), because that function word will obligatorily have a (stress-bearing) content word on its right. However, there are cases that seem to be intermediate, like ha 's/he has', which as a full verb is a content word and as a past tense auxiliary is a function word. Similarly, it is not immediately obvious whether a preposition like fra 'between' is a content word or a function word.

For Loporcaro (1997), only the following monosyllables are unstressed and therefore irregular RS-triggers:

Other accounts however tend to eliminate the category of irregular RS altogether by assuming that all RS-triggers are by definition stressed (Camilli 1965, Korzen 1980, Basbøll 1989, see the review of the literature in Loporcaro 1997: 2–9). But this position, as criticized by Agostiniani (1992) and Loporcaro (1997), is plainly circular. It leads to the undesirable consequence that, for example, the grammatical preposition a must be considered as stressed, because it triggers RS, and that the equally grammatical preposition di must be considered as unstressed, because it does not trigger RS.

Not only is it not clear whether a monosyllable is underlyingly stressed or not; it is also, as noted by Agostiniani (1992), often impossible to tell the accentual difference between homonymous "stressed" and "unstressed" monosyllables:

The form da can be a preposition with the meaning 'of, from' or the 3SG present indicative of the verb dare 'to give'. As a preposition, it counts as unstressed and belongs to the set of irregular RS-triggers. As a verb form, it counts as stressed on the last (and only) syllable and therefore as a regular RS-trigger. However, the alleged difference in stress is not reflected in the surface form of these sentences. But note, that da always triggers RS, irrespective of whether it actually receives stress in the sentence or not. Agostiniani (1992) reports that in the dialect of Lucca (Tuscany) some stressed monosyllables do not trigger RS. On the basis of this evidence he suggests that RS triggered by monosyllables cannot count as regular RS. Rather, if monosyllables trigger RS, they are lexically marked as RS-triggers. This in turn means, and I would like to endorse this conclusion, that all monosyllabic RS-triggers are irregular RS-triggers.

3 Stress and weight: On the phonology of regular RS

In this section, I will discuss the phonology of regular RS in a somewhat simplified fashion. This will serve as a backdrop for the subsequent discussion of irregular RS. For a more detailed discussion, I refer the reader to Absalom & Hajek (1997). Regular RS is a phonological process that changes the syllable structure of the involved words. The first of the two geminate consonants is the coda of the last syllable of the preceding word (the RS-trigger, ₁), and the second of the geminates is the onset of the first syllable of the host word (₂). Concomitantly, in the sequence città pulita 'clean city' the last, stressed, syllable of città, changes its structure from CV to CVC, thereby becoming heavy (bimoraic):

Hence, resyllabification along the lines of (14) yields a result which is desirable in terms of Natural Phonology. Stressed syllables should be heavy, bimoraic (µµ) rather than light, monomoraic (µ). This preference is referred to as "stress to weight" constraint in Optimality Theory (cf. Absalom / Hajek 1997, Borrelli 2002). A different strategy to comply with this constraint, and apparently more common cross-linguistically (including Northern Italian dialects, cf. Borrelli 2002), is vowel lengthening:

The peculiarity of RS with respect to syllable weight is therefore that it offers the possibility to increase syllable weight without vowel lengthening.

The fact that RS is phonologically motivated makes irregular RS all the more puzzling. With irregular RS, too, a light syllable becomes heavy. But the increase in syllable weight is not motivated by universal preferences, because the last syllable of (₁ is not stressed and therefore does not require two moras – recall that irregular RS-triggers are either monosyllables or paroxytones:

Irregular RS is therefore typically explained, when dealt with at all, as a historical residue of some other phenomenon.

4 Loporcaro (1997): Irregular RS continues Latin consonant assimilation

The diachronic origin of regular as well as irregular RS is customarily located in yet another sandhi phenomenon, a phenomenon of the spoken Latin language. In Vulgar Latin, word-final consonants assimilated to the initial consonant of the following word (Schuchardt 1874, Hall 1964, Loporcaro 1997):

Note that this assimilation rule was completely independent of stress. The only requirement for the assimilation to occur was that the word ended in a consonant. Already Hugo Schuchardt (1874) pointed out that Italian RS is probably the material successor of Vulgar Latin consonant assimilation. Schuchardt referred to the change between Latin and Italian as analogy. However, as convincingly criticized by Loporcaro (1997), this change should better be termed reanalysis. The change may be illustrated as in (17):

The Latin phonological representation as in underwent the Latin phonological rule of consonant assimilation (), thereby yielding the post-lexical string . From some point in time onwards, speakers (or, better: hearers) did not relate this to the phonological representation any more, probably because word-final consonants disappeared anyway. Rather, the geminates in the string were related to final stress, because a high percentage of words were oxytones anyway and gemination of consonants satisfied the natural preference of stressed syllables for heaviness (cf. section 2). Hence, reasoning by abduction related the same surface string to a different underlying phonological representation (), thereby creating a new phonological rule, i.e., raddoppiamento fonosintattico. As is typical for reanalysis as a type of language change, a secondary, previously not rule-governed property of some surface string, i.e., final stress in our case, is integrated into the rule, thereby remotivating the structure of the surface string. The Vulgar Latin assimilation rule of regressive assimilation

had been reanalyzed as stress-driven gemination, i.e. RS:

This reinterpretation of Schuchardt's analysis allows Loporcaro (1997) to account elegantly for the rise of stress-driven (regular) RS. But what about irregular RS? Recall that irregular RS-triggers do not have stressed ultimae and therefore do not fall under the scope of the reanalysis outlined above. Given that the reanalysis in (17) provides for oxytones to trigger RS, non-oxytonic words triggering RS, i.e., the irregular RS-triggers, cannot be explained as reanalysis of Latin consonant assimilation. This state of things leads Lopocaro (1997) to assume that today's irregular RS is a direct residue of Latin consonant assimilation. That is, according to Loporcaro, irregular RS, unlike regular RS, is not the result of the reanalysis, but it is a residue of the former diachronic stage, which has not undergone reanalysis. In other words, irregular RS is phonologically not the same as regular RS, but it is, as it were, still an instance of consonant assimilation and only superficially identical to RS. As evidence for this assumption, Loporcaro cites the fact that most of today's irregular RS-triggers had Latin etyma ending in a consonant. Bullock (2001) makes the same point for an analysis of RS in Neapolitan. This is indeed a necessary prerequisite for their analyses. If irregular RS is claimed to be a direct residue of Latin consonant assimilation, then the irregular RS-triggers must of course have had Latin etyma with final consonants. This is in effect the case with most of them:

Some, however, had etyma ending in a vowel:

The fact that there are irregular RS-triggers whose etyma ended in a vowel is a problem for Loporcaro's and Bullock's analyses. It makes the statistical argument quite weak. The size of the set (21) is indeed not completely negligible, given the size of the set (20). In Neapolitan however, the dialect analyzed by Bullock, the numerical distribution is better. Only two of the RS-triggers in that dialect had an etymon ending in a vowel. Bullock comments on this in the following way:

[T]he etymological evidence of final consonants in the RS [raddoppiamento sintattico, RW] triggers in Neapolitan is overwhelming. With the exception of two forms, ogni < omnem and quacche < qual(e) che (sia), where no etymological consonant can be reconstructed in final position, all other forms are attested with final consonants. [...] It is clearly not the case that I am claiming that speakers can actively reconstruct etymological consonants [...]. What I am proposing is that a diachronic residue of the language's phonological structure has become lexicalized in the synchronic grammar of the language. (Bullock 2001: 51)

One might defend Bullock's and Loporcaro's point of view by saying that words of the type in (21) were added later to the set of irregular RS-triggers by analogy. But this latter solution, too, suggested by quite a few authors (Weinrich 1958: 53, Canepari 1991: 100, Absalom / Hajek 1997: 176, Loporcaro 1997: 24n), faces a problem:

Traditionally, analogy is conceived of as involving a morphological relation between the forms in question. For example, analogy may reduce stem allomorphy when one stem type extends to all exponents of a paradigm. This conception of analogy however does not fit irregular RS. The list of irregular RS-triggers does not constitute a paradigm in the morphological sense of the word. Its items are morphologically unrelated.

However, not all theorists of analogy confine analogy to morphologically related words. For Kiparsky (1988), analogy is "grammar simplification". This means that analogy can be invoked also for phonological changes in morphologically unrelated sets of words. It might therefore apply to irregular RS. But even so, the extension of the set of irregular RS-triggers cannot count as analogy, because, as explained above, irregular RS is phonologically marked. It would have been a simplification of the grammar if the set of irregular RS-triggers had been eliminated over time!

What is more, the purely etymological explanation for irregular RS provides absolutely no answer to the question as to why the irregular RS-triggers are consistently very short, most often monosyllabic, words. If etymology was the sole reason for the rise (or maintenance) of irregular RS, one would not expect the set of RS-triggers be limited to monosyllables. Rather, there should be no such restriction as to the membership in the set of RS-triggers and there should be irregular RS-triggers with various word lengths.

A different solution is therefore in order.

5 Raddoppiamento fonosintattico in Old Italian

5.1 Phonology and orthography

Old Italian texts, which did not yet have standardized spelling, sometimes have words beginning with double consonant letters. I would like to analyze this as graphical notation of raddoppiamento fonosintattico. Of course it would be naive to take double consonant letters in medieval texts straightforwardly as the expression of phonological gemination. But several reasons enable us to assume that in the older language RS was at least to some degree reflected in the spelling.

First, double consonant letters show up only very rarely at the left word edge in Northern Italian texts. This coincides with the absence of RS in Northern Italian dialects, thereby suggesting that the occurrence of double consonant writing in Central and Southern dialects indeed reflects a phonological phenomenon.

Second, some quite peculiar spelling customs suggest that scribes had a metalinguistic awareness of RS (cf. Michel 1997: 297). For example, some Old Neapolitan texts connect the two words linked by RS with a hyphen:

And, one of very first Italian texts, the inscription of the Roman catacomb of Commodilla (9^th century), has one of the initial double consonants superscripted:

This looks very much like an attempt to reflect RS in the writing without altering too much the overall gestalt of the word boce 'voice'.

We can therefore assume that double consonant letters reflect indeed a phonological phenomenon.

5.2. RS in Old Italian texts

For methodological reasons, I will restrict myself mainly to Old Tuscan texts. Only these can be taken as legitimate historical evidence, because Modern Standard Italian is the historic successor of that dialect. On inspecting Old Tuscan texts in chrestomathies and corpora like OVI, one makes two important observations:

1.) Most of today's irregular RS-triggers could trigger RS already in the Middle Ages. That is, one finds occurrences for them with the next word beginning with double consonant letters.

2.) Medieval texts show RS with even more non-oxytonic words whose etymon ended in a vowel than in contemporary Italian. In other words: We find words, like the preposition di and the clitic pronoun se, whose etymon ended in a consonant and which do not trigger RS in modern Italian but apparently did so in Old Italian:

In order to assess the quantitative relevance of these examples, I conducted a study on the OVI corpus, a 20,000,000 word online corpus of medieval Italian. OVI contains a very large percentage of the Italian literature written before 1375. Among its remarkable features is the fact that it enables the user to restrict searches regionally. That is, it is possible to take only texts from a certain region or a certain variety of regions into consideration. In this study, only Tuscan texts have been considered. My goal was to find out how often today's irregular RS-trigger provoke RS in the medieval language and to what extent this behavior reflects properties of the Latin etymon. This study relates therefore three periods of Italian: Today's irregular RS-triggers are divided into two groups, those with a Latin etymon ending in a vowel and those with a Latin etymon ending in a consonant. For each of these groups it is assessed how often their respective members trigger RS in the medieval language. Due to technical restrictions (number of expressions matching the query) the RS-effect has been assessed only on words with two letters. The results are the following.

The tables (25) to (27) read as follows. For example, the preposition a 'at' (table 25) occurred 39951 times (n) before a two-letter word in the corpus. Of these occurrences, 950 were followed by a geminate (_#CCV#), which amounts to a percentage of 2.38%.

It turns out that, on average, words whose etymon ended in a vowel (table 26) triggered RS even more often than those whose etymon ended in a consonant (table 25). If one assumes that etymology is the decisive factor for irregular RS, as Loporcaro (1997) and Bullock (2001) do, this is a surprising and unexpected result. One would expect that the average percentage of (25) were much lower, if not downright zero, than the average percentage of the words in (26).

Table(27) is even more surprising, since it has the highest average percentage of all three tables. On the etymological account, we would expect it to have a percentage close to zero. It has to be said, however, that the relatively high rate of 3,22 % in table (27) is largely due to the clitic pronoun ce, which triggers RS in 32,65 % of its occurrences. And the near-entirety of occurrences of RS triggered by ce are in fact in one single text. Most of the other monosyllables which do not trigger RS today did not do so either in the medieval language. This finding confirms the initial assumption that double consonant letters in the medieval texts reflect phonological gemination.

What we do find, then, is that there is indeed a correlation between double consonant letters in the texts and phonological gemination, as initially assumed. However, comparing tables (25) and (26), there does not seem to be a relevant overall correlation between the etymology of a monosyllable and its status as an actual RS-trigger. Many, though not all, words with an etymon ending in a vowel (e.g. fra, tra) trigger RS much more often than words with an etymon ending in a consonant (e.g. qui, qua). Even though there are large (and as of now unexplained) differences between the individual words as to their respective rate of RS-triggering, the figures confirm the suspicion that the presence vs. absence of etymologically motivated final consonants cannot be solely responsible for the triggering of irregular RS. There must be another reason. I would like to put forward the claim that today's irregular RS is largely a residue of a phrasal stress rule of Old Tuscan.

6 Irregular RS is the fossilization of phrasal stress in Old Tuscan

It is a crucial prerequisite for my claim that there are different types of stress in language, as is customarily assumed in stress theory (cf. Hayes 1995, Nespor 1999 for an overview). In particular, it is necessary to distinguish between word stress vs. phrasal stress, on one hand, and between primary vs. secondary stress on the other hand.

6.1 Word stress vs. phrasal stress; primary vs. secondary stress

Word stress means that a word-form receives stress on a certain syllable at the lexical level. Word stress assignment may be phonologically predictable (fixed) or lexically listed (free).¹ I symbolize it with the asterisk (*). In Italian nouns, as in most Romance languages, word stress is assigned within a three-syllable window at the right edge of the word (cf. Roca 1999, Ghini 2001).

Italian has even minimal pairs, that is, pairs of words that differ only with respect to stress placement: príncipi ‚princes' and princípi ‚principles', for example. Given the discussion in Section 2, word stress applies only to polysyllabic words.

There is another type of stress, namely phrasal stress, symbolized with a + sign. Phrasal stress is assigned post-lexically in order to comply with rhythm, syntax or emphasis requirements. This entails that phrasal stress is not always assigned to the same syllable of a word. Rather, its placement varies, according to the position of the word in the sentence. This can be seen in its maybe most obvious form in nursery tunes (cf. Hayes 1995: 26–7):

In these cases, stress is not a predictable property of the words, but is governed by the rhythm requirements of the tune.

There is yet another distinction important for my argumentation, that between primary and secondary stress. Words, especially long ones, do not have only one prominent syllable. Rather, there may be a secondary prominence, i.e., secondary stress (marked with a gravis (`), in contrast to primary stress, marked with an acute (´)):

elettricità may have, besides its main (primary) stress on the ultima (lá), secondary stress on è and trì. In the same way, caratterizzabile has main stress on the antepenult (zá) and secondary stress on cà and tè.

It has been long noted (Vogel and Scalise 1982, Lepschy 1992, Peperkamp 1998) that secondary stress enjoys some variability. That is, the stress patterns in (30) are not the only ones. The following ones are available, too:

Whereas primary stress always falls on the same syllable, secondary stress is to a certain degree variable, especially in long words like those in (30) and (31). However, secondary stress assignment is not completely arbitrary; it tends to obey the following constraints (Vogel and Scalise 1982, Peperkamp 1998):

The choice between the variants in (30) and (31) is itself not random. It depends to a certain degree on the position of the word in the sentence. Compare (33a/b):

The stress patterning and the position of the word elettricità in these sentences are not identical. In (33a), elettricità follows the pattern (30) and it is the first word in the sentence; in (33b), it follows the stress pattern (31) and it is placed after a monosyllabic verb. That is, in both examples secondary stress assignment complies with the principles listed in (32). In (35a), principle (32a) is observed by stressing è because the first syllable of the word elettricità is, together with the elided article, the first syllable of the sentence anyway. The next secondary stress on trì is governed by principle (32b), requiring an alternation of stressed and unstressed syllables. In (33b), principle (32a) is observed, too, because the word elettricità immediately follows a stressed monosyllable. The principle of binary rhythmical spacing (32b) requires therefore the first syllable of elettricità to be unstressed, yielding pattern (33b). Note that the syllable count underlying secondary stress assignment is not confined to the word, but may cross word boundaries, as in (33b).

The upshot of this discussion is that secondary stress is not an additional type of word stress, but an instance of phrasal stress (cf. Lepschy 1992). It is not a lexical property of the word-forms, lending itself to phonological contrasts; rather, it follows eurhythmic principles of even rhythmical spacing in the sentence (Bertinetto 1981), thereby giving rise to different stress patterns in long words, depending on the contexts these words are inserted in.² That is, the contrast between primary and secondary stress in Italian is twofold: not only is it a contrast between a primary and a secondary prominence in the word, but also is it a contrast between word stress and phrasal stress.

6.2 RS in Old Italian was triggered also by phrasal stress

The basic idea of my approach, now, is that the same principles of phrasal stress which are active in the assignment of secondary stress in the modern language were also at work in the medieval language, with the crucial difference that in Old Tuscan phrasal stress thus assigned was able to trigger RS.

Let us consider one instructive case. Example (34), taken from one text, has two occurrences of se ‚him/herself'.

Se is a clitic, i.e. a word without word stress. In the first occurrence se is followed by another clitic (la), which has raddoppiamento. In the second occurrence se is followed by a verb form with word stress (mise ‚he/she put'), without raddoppiamento. If we apply the principles listed in (32) to these sequences and if we assume that phrasal stress assigned to the ultima according to these principles was able to trigger RS, too, then the difference between the two sequences in (34) follows straightforwardly. The principle of binary (trochaic) rhythmical spacing (32b) enforces phrasal stress on clitic se in the first sequence, but not in the second sequence. Assuming that phrasal stress on the ultima triggered RS, too, it becomes clear that we get gemination on la, but not on mise. Given that se was stressed or unstressed according to its context, it became allomorphous. It had two variants: an unstressed one, which did not trigger RS and one with phrasal stress, which did trigger RS.

Other instances suggesting that phrasal stress on the ultima could trigger RS are the following:

(Breve di Montieri, 1219, C 47)

Principle (32a) favors phrasal stress on the first element in clitic clusters, i.e. in sequences of word-forms that do not have lexical stress, whereas principle (32b) disprefers phrasal stress on the second element. In clitic clusters, we therefore find RS on every other clitic.

What all this amounts to is the following hypothesis: In contemporary Italian, regular RS is triggered by every oxytone, final stress being assigned at the word level. In Old Tuscan however, RS was apparently triggered by every oxytone AND by words that happened to be stressed on the ultima due to post-lexical processes. Phrasally enforced RS created allophonic variation, because the word is stressed or not depending on its phrasal context and accordingly it will or will not trigger RS.

An empirical prediction of this claim is that we will find many more clitic clusters with RS on the second clitic than clitic clusters with RS on the first clitic, given that the principles of phrasal secondary stress (32) provide for every other syllable to be stressed, beginning with the first syllable in a sequence. This prediction is borne out by the facts. A search on the Old Tuscan texts in OVI reveals the following proportion:

This means that the number of clitic clusters with RS on the second clitic is about seven times higher than the number of clitic clusters with RS on the first clitic. This fully complies with the assumption that principle (32a) favors, but not enforces, phrasal stress on the first element of a clitic cluster, whereas principle (32b) disprefers, but not precludes, phrasal stress on the second element in a clitic cluster.

The difference between Modern Italian and Old Tuscan, then, is that phrasal stress on the ultima was able to trigger RS in the latter but not in the former. Given that one clitic could surface with a simple or a geminated initial consonant, depending on whether the previous element triggered RS or not, there were allophonic variants. With the loss of phrasally triggered RS, the regular relationship between the allophonic variants (stressed or unstressed) was lost. As a consequence of this, the allophonic variation between stressed and unstressed variants was reanalyzed as allomorphic variation. Only one of the allomorphs, either the stressed or the unstressed one, could survive. If the stressed one was the lucky one, then it turned into a lexically specified irregular RS-trigger. This state of things can be illustrated as in (45).

Which one of the allomorphs survived as the new unit is perhaps a random choice. With the clitic pronoun se, the variant without RS has been selected; with the (equally clitic) preposition a, the variant with RS has been chosen. But it proberbly could as well have been the other way round.

I do not want to claim that etymology plays absolutely no role in the determination of whether a word becomes an irregular RS-trigger or not. For instance, it is quite striking that the preposition di (< DE) 'of', which today is not an RS-trigger, had a RS-triggering rate of only 0,14 % already in the medieval language, whereas the semantically very similar preposition a (< AD) 'at', which today is an RS-trigger, had a rate of 2,38 % at that time. Probably the large differences with respect to RS-triggering documented in tables (25) to (27) cannot be attributed to a fossilization of phrasal stress alone. But if phrasal stress can be shown to be a source of irregular RS, then it is no longer necessary to assume that irregular RS is a direct residue of Latin consonant assimilation.

6.3 Why unstressed monosyllables?

Why is it that irregular RS-triggers are overwhelmingly unstressed monosyllables? On the standard view, there is no explanation for this fact. Analyzing irregular RS as the immediate residue of Latin consonant assimilation does not offer an avenue of explanation, given that Latin consonant assimilation was not at all sensitive to word length. But under the view advocated here, two reasons, independent from each other, can be advanced in order to explain the distribution of irregular RS-triggers.

1) Only words that do not have word stress qualify for phrasally enforced RS. That is, only clitics can have phrasal stress on the ultima (i.e. phrasal stress without having word stress), and only short words, preferably monosyllables, can be clitics. There is, hence, a strong tendency for monosyllables to be recruited as irregular RS-triggers.

2) The irregular RS-triggers are among the most frequent words of the Italian language. Nearly all of them have a very high rank in the frequency count of LIP (see appendix). It is, furthermore, immediately obvious that the words in the list (5) belong to the absolutely basic vocabulary of the language. Now, it is well-known that words with high token-frequency are preferably short (Zipf's law), which again favours their inclusion in the set of irregular RS-triggers. And, importantly, it has been shown that sandhi phenomena that have turned unproductive over time tend to remain active longer in words with high token-frequency (Bybee 2001). This favours the preservation of the capability to trigger RS even after the loss of phrasal stress.

Two factors, lack of lexical stress and high token frequency, constrain the set of possible irregular RS-triggers. They favour short words, ideally monosyllables. Within this set, there is a range for variation. Proof of this is that still today, the Italian dialects differ greatly with respect to their set of effective irregular RS-triggers.

7 Conclusion

Summarizing, I hope to have shown the following:

1.) Irregular RS is not an immediate residue of Latin consonant assimilation. Rather, today's irregular RS is a relic of a phrasal stress rule of Old Italian. In Old Italian, RS was triggered also by phrasal stress, if assigned to the last syllable of a word, thereby creating allophonic variation in clitics. With the loss of this rule, allophonic variation was reanalyzed as allomorphic variation.

2.) Today's irregular RS-triggers are fossilizations of one of those allomorphous variants.

3.) Some of the lexically unstressed words have fixed the variant with RS, others did not. Even today, the degree of variation found in Italian dialects as to the selection of RS-triggers testifies the variability of the selection process.

8 Bibliography

8.1 References

Absalom, Matthew / John Hajek (1997): "Raddoppiamento sintattico: What happens when the theory is on too tight?", in: Pier Marco Bertinetto et al. (eds.), Certamen Phonologicum III. Papers from the Third Cortona Phonology Meeting, April 1996, Torino, 159–179.

Agostiniani, Luciano (1992): Su alcuni aspetti del 'rafforzamento sintattico' in Toscana e sulla loro importanza per la qualificazione del fenomeno in generale, Quaderni del dipartimento di linguistica 3, 1–28.

Basbøll, Hans (1989): "Phonological weight and Italian raddoppiamento fonosintattico", in: Rivista di linguistica 1, 5–31.

Bertinetto, Pier Marco (1981): Strutture prosodiche dell'italiano, Florence.

Borrelli, Doris (2002): Raddoppiamento sintattico in Italian, New York.

Bullock, Barbara E. (2001): "Consonant gemination in Neapolitan", in: Repetti (ed.), 45–58.

Burzio, Luigi (1994): Principles of English stress, Cambridge/UK.

Bybee, Joan (2001): Phonology and language use, Cambridge/UK.

C = Castellani, Arrigo (1982): La prosa italiana delle origini. Vol. 1: Testi toscani di carattere pratico, Trascrizioni, Bologna.

Camilli, Amerindo (³1965): Pronuncia e grafia dell'italiano, Florence.

Canepari, L. (1991): "Rafforzamento sintattico": Teoria, terminologia e geolinguistica, in: G. Borghello / M. Cortelazzo / G. Padoan (eds.): Saggi di linguistica e di letteratura in memoria di Paolo Zolli, Padua, 99–116.

Fanciullo, F. (1983/86): "Syntactic reduplication and the Italian dialects of the Centre-South", in: Journal of Italian Linguistics 8, 67–104.

Ghini, Mirco (2001): Asymmetries in the phonology of Miogliola, Berlin.

Hall, Robert (1964): "Initial consonants and syntactic doubling in West Romance", in: Language 40, 551–556.

Hayes, Bruce (1995): Metrical stress theory. Principles and case studies, Chicago.

Kiparsky, Paul (1988): "Phonological change", in: F. Newmeyer (ed.): Linguistics: The Cambridge Survey, Cambridge/UK. I, 363–415.

Korzen, Iørn (1980): "Il raddoppiamento sintattico e la geminata nella variante toscana dell'italiano-standard. Risultati di un'indagine sperimentale", in: Studi italiani di linguistica teorica e applicata 9, 333–366.

Lepschy, Giulio (1992): "Proposte per l'accento secondario", in: The Italianist 12, 117–128.

LIP = Tullio de Mauro et al. (1993): Lessico di frequenza dell'italiano parlato, Rom.

Loporcaro, Michele (1997): L'origine del raddoppiamento fonosintattico. Saggio di fonologia diacronica romanza, Tübingen.

M = Michel, Andreas (1997): Einführung in das Altitalienische, Tübingen.

Nespor, Marina (1999): "Stress domains", in: H. van der Hulst (ed.), 117–159.

OVI = Opera del Vocabolario Italiano. [http://www.csovi.fi.cnr.it/italnet/OVI]

Peperkamp, Sharon (1996) [1998]: "A representational analysis of secondary stress in Italian", in: Rivista di linguistica 9, 189–215.

Repetti, Lori (ed.) (2001): Phonological theory and the dialects of Italy, Amsterdam.

Roca, Iggy (1999): "Stress in the Romance languages", in: Harry van der Hulst (ed.), 659–811.

Rohlfs, Gerhard (1949): Historische Grammatik der italienischen Sprache und ihrer Mundarten. Bd. 1: Lautlehre, Bern.

Schuchardt, Hugo (1874): "Phonétique comparée", in: Romania 3, 1–30.

Van der Hulst, Harry (ed.) (1999): Word prosodic systems in the languages of Europe, Berlin: Mouton de Gruyter.

Vogel, Irene and Sergio Scalise (1982): "Secondary stress in Italian", in: Lingua 58, 138–167.

W = Wartburg, Walther von (1961): Raccolta di testi antichi italiani, Bern.

Weinrich, Harald (1958): Phonologische Studien zur romanischen Sprachgeschichte, Münster.

8.2 Further Literature

Beckman, Mary G. (1986): Stress and non-stress accent, Dordrecht.

BLI = Biblioteca della letteratura italiana. [http://www.letteraturaitaliana.net]

Iannucci, James E. (1948): "Gemination of initial consonants and its semantic function in Neapolitan", in: Romance Philology 2, 237–239.

Maiden, Martin (1995): A linguistic history of Italian, London.

Mayerthaler, Willi (1982): "Markiertheit in der Phonologie", in: Theo Vennemann (ed.): Silben, Segmente, Akzente, Tübingen, 205–246.

Nespor, Marina (1993): Fonologia, Bologna.

Nespor, Marina and Irene Vogel (1986): Prosodic phonology, Dordrecht.

Repetti, Lori (1991): "A moraic analysis of raddoppiamento fonosintattico", in: Rivista di linguistica 3, 307–330.

Vincent, Nigel (1988): "Non-linear phonology in diachronic perspective: stress and word-structure in Latin and Italian", in: P. M. Bertinetto, M. Loporcaro (eds.): Certamen phonologicum. Papers from the 1987 Cortona phonology meeting, Turin, 421–32.

Zwicky, Arnold M. (1977): On clitics, Bloomington.

Tekavcic, Pavao (1972–74): Grammatica storica dell'italiano, Bologna.

9 Appendix: Irregular RS-triggers in LIP

The following table is extracted from LIP (Lessico dell'italiano parlato). It lists the token frequency of Italian lexemes in the spoken language in descending order, until the occurrence of fra (item no. 186). All of the irregular RS-triggers listed in (5), except the onomatopoetic fru, are highlighted in boldface and are contained in the table. That is, the irregular RS-triggers are among the 186 most frequent words of the language. The table shows furthermore that irregular RS-triggers occur massively in the highest frequency ranks and become rarer and rarer as frequency descends.

^* This paper has been presented at the University of Tuebingen (Dec. 2002), the 11^th Manchester Phonology Meeting (May 2003) and the University of Konstanz (Dec. 2003). I would like to acknowledge comments from the audiences. I am indebted to Matthew Absalom, Ulrich Detges, John Hajek, Peter Koch, Astrid Kraehenmann, Judith Meinschaefer as well as two anonymous reviewers for excellent comments on earlier versions. I would like to thank especially one anonymous reviewer who made very detailed and inspiring comments and concrete proposals. Furthermore, the help from Mark Olsen from ARTFL is gratefully appreciated.

¹ Today, word stress in Italian is generally considered as largely predictable (cf. Burzio 1994, Roca 1999, to name just a few references). But the predictability of word stress is not crucial for the distinction between word stress and phrasal stress.

² Bertinetto 1981 shows that secondary stress in Italian is not distinctive. That is, there are no minimal pairs of words differentiated exclusively by placement of secondary stress.