Syllable structure and prosodic words in Early Old French

This paper presents a comprehensive analysis of the phonotactics of syllable rhymes based on all unique tokens in two Early Old French texts. Based on the data from this single, conservative variety, I develop Jacobs’ (1994) proposal that the Old French stress rule is underlyingly trochaic and that wordinal stress is caused by the presence of an empty-headed inal syllable. I argue that this analysis can only be valid while words with inal stress systematically end in a consonant that can, and often must, be parsed as the onset to an empty-headed syllable. Although this is not the case in most later varieties of Old French, the prediction is borne out by our data. I conclude by examining the implications of this analysis for the accentuation and phonotactics of monosyllables and for the study of prosodic change in Old French.


Introduction
Lexical stress in Old French (9 -13 century) shows a simple surface pattern: it is inal (1a), except where the inal vowel is a schwa, in which case it is penultimate (1b): (1) a. petit [pə.ˈtit] 'small. ' b. ensemble [eñ.ˈsem.blə] 'together' In one sense, this surface pattern persists until the apocope of inal schwa at the end of the 16 century, when stress becomes regularly word inal. In another sense, it undergoes important changes during the Old and Middle French periods, as lexical stress is partially replaced by 'group stress', i.e. stress assigned not at the level of the prosodic word, but at the right edge of a higher prosodic constituent (cf. Rainsford 2011, Marchello-Nizia 2015. This change is re lected in John Palsgrave's (1530) detailed description of French pronunciation, where he notes that all monosyllables are unstressed except before a 'poynt', i.e. at the end of a clause: Generally there is no worde of one syllable in the frenche tonge that hath any accent, except the comyng next vnto a poynt be the cause of it (Palsgrave 1530: 112).
While lacking the theoretical tools to formalize the observation, philologists and historical linguists have long recognized the importance of the development of group stress for French phonology. Pope (1952Pope ( [1934), for example, identi ies a turning point between Early and Later Old French, dividing the historical development of the language into 'Period I' and 'Period II': The dominant factors in the evolution of pronunciation in Later Old and Middle French are the gradual lessening of the heavy tonic stress that characterised Period I and a new tendency to link closely together words closely connected in thought (Pope 1952: §170).
The proposed link is between a change in nature of stress and the development of new 'linking' phenomena, most notably liaison, a process which resyllabi ies all inal consonants into the onset of a following vowel-initial word; before a consonant, they are deleted (Pope 1952: §611- §624). While I reject Pope's views on the nature of stress -'heavy tonic stress' implies that stress systems can vary according to how vigorously the speaker exhales during the tonic -the proposed link between stress and a linking phenomenon which fundamentally affects syllable structure remains a valid one. Seǵeŕal and Scheer (in press: ch. 22) provide a 21 -century formulation of the same fundamental observation: there was an enlargement of the computational domain beyond the prosodic word, affecting a range of (morpho-)phonological processes, including stress assignment, the syllabi ication of inal consonants and even the cliticization of subject pronouns.¹ ¹ There is, however, no evidence from the principal 16 -century descriptions of French prosody (Palsgrave 1530, Meigret 1550) that the position and realization of primary stress in words of more than one syllable is affected by this shift in the computational domain. Indeed, as in the quote above, all of Palsgrave's observations which can be cited in support of a group-stress analysis of 16 -century French are explicitly restricted to "words of one syllable". This suggests that we must consider the possibility that the prosodic word retains a role in the prosodic system despite the shift in the computational do-The goal of this paper is modest in comparison with such fundamental questions. I will show that there is a clear association between the stress rule and phonotactic constraints on syllable structure at a stage in Early Old French when the prosodic word was still the relevant phonological domain for stress and syllabi ication, i.e. the end of Pope's Period I. I will not attempt to analyse 'Old French' in general, since this encompasses a plethora of diatopic and diachronic varieties which were certainly at different stages of prosodic change. Instead, I will provide an comprehensive account of a single conservative variety: early literary Anglo-Norman as recorded in two of the oldest extant Old French manuscripts. Although the analysis I propose draws heavily on previous work by Jacobs (1994), by accounting for all -and only -the forms found in these two texts, I will show that this analysis makes clear and accurate predictions about what is and what is not possible when the prosodic word is the relevant domain for stress and syllabi ication, and suggest therefore that these predictions can be used as diagnostics for the advance of prosodic change.

Method
The data and analysis are based on an exhaustive study of every graphical form in two Early Old French texts: the Life of Saint Alexis (ed. Rainsford and Marchello-Nizia 2018, henceforth AlexisRaM) and the Song of Roland (ed. Moignet 1972, henceforth RolMoign). In each case, the editions are based on a single, early 12 -century manuscript copied in England (1121 for AlexisRaM, 2ⁿ quarter of the 12 century for RolMoign). Not only are the manuscripts contemporary and from the same region, they are also among the earliest manuscripts containing Old French texts of suf icient length to allow a comprehensive study of phonotactic constraints.² The texts themselves were composed in the mid-to late 11 century and in France, possibly Normandy. It is therefore valid to assume that the variety of French used in both texts is very similar. To avoid ambiguity, I will refer to this particular variety of early 12 -century literary Anglo-Norman simply as the ' AlRol' variety of Old French.
To gather the data, I irst extracted all unique graphical forms with accompanying lemmatization and part-of-speech annotation from the Base main. Although I will address this question directly in forthcoming work, it is tangential to the goal of the present paper and will not be developed further. ² Earlier texts, such as the 9 -century Strasbourg Oaths and Sequence of Saint Eulalia are too short, while the 10 -century Passion of Clermont and Life of Saint Leger were copied in a mixed Franco-Occitan scripta, making them unsuitable for phonotactic analysis. . Distinctions in the vowel system were especially problematic, since phonemic differences are not systematically transcribed. Here, the reconstruction was informed by historical grammars (Pope 1952, Fouché1952-1961 [ə] in pre-and post-tonic position. Reconstructed vowel phonemes and the corresponding graphemes are presented in Table 1. More details about the transcription of speci ic segments are provided in the associated material at the end of the paper. Finally, transcriptions were semi-automatically syllabi ied based on the sonority hierarchy.³ Primary stress was assigned to the inal nonreduced vowel of a polysyllable. Each transcription was then exported as a data table listing properties of the onset, nucleus and coda of each syllable by its position in the word (pre-tonic, tonic and post-tonic) (cf. Rainsford 2020).
.3 ') and consequently may be represented phonemically as a nasal archiphoneme /N/ underspeci ied for place features.⁴ The phonemic opposition between the lateral phonemes /l/ and /ʎ/ is neutralized in word-internal codas and re lexes of both phonemes are realized [ɫ] (later vocalized to [w]), which I will represent phonemically as /L/. Finally, regressive voicing assimilation neutralizes the opposition ³ The Python scripts used remain under development, but the code is freely available at https://sourceforge.net/projects/syllabic-verse-analysis/. ⁴ A small number of forms suggest that nasal assimilation was not systematic, e.g. conpta ?[ˈkũm.tə] 'count.
. ' < Lat. ̆ ̆ . The historical development of this form shows late syncope of the Vulgar Latin penult, see Morin (2003) for a detailed discussion.   Dell 1995), they are exceptional in AlRol. Given the small number of cases, I will assume that Latin forms were not (yet) well enough integrated to affect French phonotactics.

Type B: Clusters impossible in word-inal position
Included in type B are all attested consonant clusters which are phonotactically excluded from appearing in word-inal position. These clusters are broadly identical to those which blocked the apocope of inal non-low vowels historically (cf. Pope 1952: §258). Type B1 consists of tautosyllabic obstruent + liquid clusters, which behave as branching onsets and require a support vowel. For historical reasons, B1 clusters are frequently found post-consonantally, i.e. in ˈVC__ə# contexts. Where a sonorant, a lateral, /s/ or /z/ were brought into contact with /r/ or /l/ through syncope of an intervening vowel, an epenthetic consonant developed in all cases except [ 3) are also of type B and are grouped together as type B2.⁵ These are heterosyllabic and cannot occur in ˈVC__ə contexts. Type B3 contains only the affricate /ʤ/, which has an exceptional distribution: it is the only obstruent which cannot appear, devoiced, in word-inal position.⁶ Table 3 are all word-inal consonant clusters except those ending in a lexional /s/ or /θ/, which we will consider in section 3.4. Voiced obstruent phonemes /b d g v ʣ ð z/ are given in parentheses as they are underlyingly present but subject to inal obstruent devoicing. .3 '. I make the conservative assumption that inal obstruent devoicing remains an active phonological role in AlRol and therefore that the surface [f] in [rə.ˈʦejf] is underlyingly an instance of inal /v/. However, before the second wave of inal vowel apocope in the 16 century, the inal devoicing rule becomes inactive and the feature [-voice] is lexicalized, giving rise to modern French morphological alternations such as neuf /noef/ 'new.

Included in
' vs. neuve /noev/ 'new. '. If this were already the case in AlRol, type B3 could be extended to include all voiced obstruents. .3 '. All type C clusters, with both voiced and voiceless obstruents, are also attested in ˈV(C)__ə# contexts.
The groups [-rn] and [-rm] are unusual in that they are the only clusters of two sonorants which are of type C rather than type B. However, variants jurn/jur 'day' and carn/car ' lesh' show a change in progress eliminating these forms by deletion of the inal nasal.

Final /s/ and /θ/
Final /s/ and /θ/ (→ [t] / C__#) are descended from Latin inal consonants and are most frequently found as lexional morphemes.⁷ Phonotactically, /s/ and /θ/ can be appended to any word, regardless of its type. In forms with a inal post-tonic vowel, this results in surface forms ending either [əs]  .3 '. Flexional consonants are the only consonants found following a post-tonic reduced vowel (but see also §6.2). Without a inal post-tonic vowel, adding /s/ or /θ/ causes the cluster to be simpli ied, as shown in  Table 4 show that adding inal /s/ and /θ/ does not add to the inventory of type C inal clusters, except in the case of [st] (see §3.6.1). Previously word-inal sonorants show assimilation typical of word-internal codas.

The curious case of ˈV#
Word-inal tonic vowels in AlRol are curious because they are extremely rare. In polysyllables, most cases are due to the unconditioned loss of /ð/,⁸ a change which I consider on the basis of the orthographic variation to be still in progress at this time. For example, [mɛr.ˈʦi(θ)] 'mercy' is written merci or mercit 'mercy'; the 3 future suf ix [a(θ)] can be written -at/-ad or a.
Setting aside this sound change, the only remaining polysyllabic exceptions in AlRol involve seven forms with inal tonic [i]: ami 'friend. ', ⁷ In nominals, /s/ marks plural on feminine nouns and adjectives and case and number ( . or . ) on masculine nominals; in the verbal domain, it marks 2 and it is also an adverbial marker. /θ/ marks 3 on verbs.

Phoneme Result
Example Latin proto-fr. OFr.  Table 5, the development of words derived from Latin -̄ and -̄ passes through an [ij] stage. Moreover, in -̄ and -̄ suf ixes, this stage is still attested in AlRol.
This leads to an important claim: (2) *ˈV# The tonic vowel is never word-inal.
This claim holds true for AlRol but not for varieties of Old French after /ð/ has been lost. This does not make the observation any less important, as it shows on the contrary that the loss of /ð/ triggers a key phonotactic change.

[st]
Historically, the cluster [st] must be analysed as heterosyllabic in order to account for patterns of sound change.

The -ent suf ix
The third-person plural suf ix is consistently written -ent and is always syllabic. Depending on the pronunciation -and the only pronunciation that can be reconstructed with any certainty is [ət]⁹ -this constitutes either the only example of a word-inal cluster ([nt]) or of [t] after a posttonic vowel.

Proparoxytones?
AlRol contains a small number of semi-Latinate words which may have had proparoxytonic stress. These are aneme 'soul', angele 'angel', ydeles 'idols', humeles 'meek', imagene 'image' and virgene 'virgin'. It is certain both from the verse form and subsequent phonetic developments that these forms were stressed on the antepenult. It is also certain that the two orthographic post-tonic syllables only occupied a single metrical position.¹⁰ RolMoign also contains the variants anme for aneme and angle for angele, showing that paroxytone variants were also available. In Later Old French, the forms become paroxytonic, either by syncope of the penult (anme, humble) or by apocope of the whole inal syllable (ange, image, virge).
None of this, however, explains why a proparoxtone orthography differing from the Latin spelling would be used if the pronunciation were not in some way atypical in AlRol. With regard to their phonotactics, it can be observed that syncope of the penult would give rise to an inadmissible consonant clusters, and this combined with the in luence of church Latin pronunciation could explain the retention of proparoxytonic stress (cf. Pope 1952: §642- §645). It is also the case that the inal consonant is always a sonorant and thus potentially a valid word-internal coda. I will propose a possible explanation for these forms in §6.2. ⁹ Cf. Pope (1952: §437). There are no comparable developments found in other morphemes. ¹⁰ This is often taken as 'evidence' for a paroxytone pronuniciation (e.g. Fouché1952-1961, but this assumes a one-to-one relationship between metrical positions and phonetic syllables. It is possibly signi icant that proparoxytone variants in RolMoign only occur at the end of the line, a position in which post-tonic syllables do not 'count' for the metre.

Syllable rhymes
This section brie ly examines possible syllable rhymes in AlRol with particular attention to syllable weight. The vowel system used in the reconstruction was outlined above in Table 1. In pretonic position, only monomoraic vowels are identi ied, while in tonic position, bimoraic primary diphthongs also occur as a result of Vulgar Latin stressed open syllable lengthening (cf. Loporcaro 2015). Although its precise phonetic value is uncertain, the vowel [eː] -descended from [a] in a stressed open syllable and written simply as e -reconstructed as a phonemically long monophthong.¹¹

Pretonic syllables
Pretonic syllables contain only simple vowels (cf. §437), most evidence comes from descriptions of word-inal (and therefore post-tonic) nasal consonants. On the basis of very few types whose phonetic value is not certain, it is unclear that our analysis should allow for three-element rhymes in pretonic syllables.

Tonic syllables
Tonic syllables may have V (3a), VC (3b), VV (3c), or VVC (3d) rhymes, illustrated here by words with a inal post-tonic syllable: ( Unlike pretonics, tonic syllables may contain a bimoraic nucleus. Moreover, the range of three-element rhymes at the tonic is far wider than the few cases attested in pretonics, including not only V ȷñ and VjS groups but also diphthongs and the long vowel /eː/ followed by a coda.
Without a inal post-tonic vowel, the same typology is found, provided that the inal consonant is considered to be 'outside' the syllable: As noted in section 3.5, word-inal tonic vowels are always followed by a consonant or a glide. In short, the maximal rhyme of a pretonic syllable is VC; a tonic may additionally show VV or even VVC syllable rhymes.

Final consonants in Type C clusters
As in many languages (cf. Côté2011), type C inal consonants in AlRol are not normal codas: clusters and obstruents other than /S/ are allowed, and place distinctions are retained in nasals and laterals. Seǵeŕal and Scheer (in press: ch. 22) argue that single inal consonants in Old French are intervocalic, while inal consonants in C__# contexts develop like word-internal onsets in the same context. Similarly, Jacobs (1994) analyses inal consonants as onsets to a catalectic inal syllable, highlighting that Gallo-Romance inal vowel apocope provides a good historical justi ication for this analysis (cf. Pope 1952: §256).¹² In summary, inal consonants in type C words are outside the inal syllable of the surface form and are most similar to word-internal onsets, not codas.

Jacobs' (1994) stress rule
The point of departure for our analysis of stress is Jacobs' (1994) proposal of an Old French stress rule based on syllabic trochees parsing right-to- ¹² Dell (1995) proposes a similar analysis to account for inal consonants in Modern French which result in part from a second wave of post-tonic vowel apocope dating from the 16 century. He argues that that inal consonants in words such as table are onsets to a 'degenerate syllable '. left. The framework used is that of Hayes (1995) but with a ban on monosyllabic 'degenerate' feet. The prosodic word receives stress by End Rule Right. This gives a correct prediction for all words with a post-tonic vowel, whether of type B or type C: For words of type C without a inal vowel, Jacobs posits a catalectic inal syllable, which ills the weak position of the trochaic foot: The position of primary stress for all data presented in section 3 can be correctly parsed by this rule, suggesting that Jacobs' analysis is correct for AlRol.

Modi ications to Jacobs' analysis
However, two issues require further clari ication. First, Jacobs (1994) accounts for the fact that clusters of type B do not appear in word-inal position by positing that under certain conditions, the nucleus of the 'catalectic' inal syllable must be spelled out. Discussing the derivation of colp 'blow' and jalne 'yellow. ', Jacobs begins by positing an underlying representation with a catalectic inal syllable, i.e. /kɔL.p∅/ and /ʤaL.n∅/, to which the trochaic stress rule is then applied as in (7) above. Subsequently, the inal consonant of colp is subject to the following resyllabiication rule, associating it to the inal syllable: (Jacobs 1994: 56) In (8) I adopt Jacobs' formalism and resyllabi ication is indicated by the dashed line. In the case of jalne, none of the proposed resyllabi ication rules apply and consequently "the catalectic syllable is phonetically spelled out as schwa" (Jacobs 1994: 56 but never *per. Second, and perhaps more importantly, Jacobs maintains that the catalectic syllable can exist even when it contains no phonetic material at all, i.e. when the tonic vowel is inal in the word, as in di 'day', ni 'nest', pré ' ield' and bru 'heath'. Not only is this circular -an empty syllable is posited because stress is underlyingly trochaic, but stress is only underlyingly trochaic if we posit empty syllables -but it fails to account for the impossibility of ˈV# in AlRol. Of the four examples cited by Jacobs, ni, pré and bru all show loss of inal /ð/ and di is [i]-inal. In short, I consider that a trochaic stress rule is correct for AlRol, but will begin to break down as soon as independent surface phonotactic evidence for the inal catalectic foot has been lost. Pigott's (1999) mechanism of 'remote-licensing' formalizes constraints regulating the appearance of empty-headed inal syllables. In this approach, all prosodic segments must be licensed by a higher prosodic constituent. Most segments are licensed by the prosodic constituent which immediately contains them, i.e. consonants are licensed by syllables, syllables by feet, etc. However, Pigott posits a mechanism of 'remote-licensing' of prosodic constituents, de ining the conditions in which a prosodic constituent can be licensed by a constituent even higher in the prosodic hierarchy: -An element β may be R-licensed by α iff it is leftmost/rightmost in α and is immediately dominated by a prosodic category that is leftmost/rightmost in α (Pigott 1999: 165). More speci ically, the parameter -/ stipulates that inal consonants are R-licensed by the prosodic word (cf. Pigott 1999: 167). When active, inal consonants must be immediately dominated by a syllable at the right-edge of the prosodic word; where they cannot be parsed as codas due to phonotactic constraints, they must consequently be parsed as onsets to an empty-headed syllable. Unlike other syllables, Pigott stipulates that empty-headed syllables cannot themselves license any segments, i.e. they may only contain consonants R-licensed by the prosodic word (1999: 168).
In short, combining Jacobs' stress rule with the R-licensing of inal consonants, we predict that surface oxytonic stress can only be generated if there is (i) a inal consonant R-licensed at the right edge of the prosodic word which (ii) may be parsed as the onset of a inal empty-headed syllable which, in turn, (iii) ills the weak position of a trochaic foot. Where there is no inal consonant, there is no empty-headed syllable, and thus stress will be paroxytonic.
The analysis allows for only two possible prosodic structures for the inal surface syllable in a word or for a surface monosyllable: In (9a), the surface syllable has a inal R-licensed consonant parsed as the onset to an empty-headed syllable (i.e. it is of type C); it must form a prosodic word in order for the consonant to be licensed and it therefore bears primary stress; the stressed syllable may have a bimoraic nucleus or be super-heavy (cf. §4). In (9b), the surface syllable is underlyingly a single syllable with maximally a single consonant parsed in the coda (i.e. it is of type A); it cannot bear stress and cannot have a bimoraic nucleus.

Preventing over-generation: the status of inal /ə/
However, the analysis proposed above over-generates in two speci ic cases.¹³ First, when there is a single word-inal sonorant or /s/ in a word of more than one syllable, it is unclear why this is always treated as an onset to an empty-headed syllable and never as a coda of the inal syllable, i.e. why is only (10a) below grammatical when (10b)  no consistent morphological alternation with any full vowel in this position.¹⁵ Moreover, since it is the only vowel which can occur in inal unstressed syllables, it is by de inition in complementary distribution with all other vowel phonemes and exists only in opposition to an empty nucleus. These indications suggest an underlying vowel phoneme which, although present in the lexicon, is (at best) underspeci ied for place features and is distinct from all other vowel phonemes.
Returning to examples (10) and (11) above, I consider that the ungrammaticality of (10a) is due to the fact that the inal nucleus in a prosodic word cannot contain a segment with place features: it must either be empty or contain underspeci ied /ə/. The ungrammaticality of (11), on the other hand, is due to a converse restriction that other nuclei -and in particular the head of a PWd -require vowels to be speci ied for place features. A further formalization of these restrictions goes beyond the scope of this paper.

Consequences
The analysis proposes a clear link between the nature of the segments at the right-edge of a word and the position of primary stress. In this section, I will brie ly examine the consequences of this analysis for monosyllables ( §6.1) and for the status of inal lexional consonants ( §6.2).

Monosyllables
The core prediction of the analysis for many surface monosyllables can be summarized as follows: monosyllables with no inal consonant must have a Type A structure (cf. 9b) and are not PWds; monosyllables with a inal obstruent ( Perhaps more interesting is the prediction that some VC monosyllables are compatible with either analysis. In some cases, there is evidence to suggest that they were consistently parsed with a Type A structure: for example, the inal nasal of the clitic en [eñ] shows place assimilation to the initial consonant of the following word, e.g. ne l'em puet hom blasmer 'no-one could blame him for it', showing that the inal /N/ was underlyingly a coda. This contrasts with the retention of nasal place features in the underlying inal onset consonants of hum 'man' and num 'name' in Al-Rol (cf. §3). Others were likely to have been consistently parsed as type C, such as gros [grɔs] 'fat. ', which occurs in an in lectional paradigm with forms that are unambiguously of type C, in this case, grosse [ˈgrɔ.sə] 'fat. '. In some cases, however, it is likely that a single surface form was parsed in some functions as type A and in other functions as type C. A typical case is that of the pronouns nus and vus [nus]/[vus] 'we/us', 'you. ' < Lat. ̄ / ̄ . These occur in three separate paradigms of personal pronouns: object clitics (13a), subject pronouns (13b) and disjunctive pronouns (13c): (13) a. me, te, se, le, la, li, les, lur, nus, vus b. jo/je, tu, il, ele, nus, vus c. mei, tei, sei, lui, ele, els, nus, vus Of these, the object clitics and the subject pronouns are all compatible with a type A analysis,¹⁶ while the presence of bimoraic vowels and the inal consonant cluster in els clearly show that most disjunctive pronouns were of type C. The most obvious conclusion here is that nus and vus, forms which were compatible with either structure (9a) or (9b), could be independent prosodic words, but could also be parsed as a single syllable and integrated into a following prosodic word. Perhaps signi icantly, even in dialects in which diphthongization of [o] occurs, forms such as ¹⁶ The claim that subject pronouns in AlRol are not prosodic words is controversial, and indeed there are a small number of other vowel-inal monosyllables in our data that philologists would hesitate to categorize as 'unstressed', such as the deictic adverbs ça and la. Yet these cases are not as straightforward as they seem. For example, if jo could be a prosodic word, it is unclear why some texts show doublets with a bimoraic diphthong (gie or jou); moreover, jo subsequently develops a reduced vowel in pretonic position (cf. Marchello-Nizia 2015). Ça and la occur also in other texts with a inal glide (çai and lai). In short, while they were not clitics, it is not clear that they were prosodic words either. Note that this is not the same as saying that they were completely 'unstressed': they may still have headed a foot and borne a secondary stress. *neus and *veus are never attested, a fact that neogrammarian approaches considered 'not easy to explain' (cf. Pope 1952: §827). However, such a development is compatible with the hypothesis that doublets of this kind can be eliminated -assuming that they develop at all -if a form can be parsed either with or without a inal empty nucleus.¹⁷

Codas in post-tonic syllables
Flexional consonants [s], [θ] and [t] in the suf ix -ent can occur after inal [ə] (cf. §3.4). Two possible analyses could be envisaged here: either these consonants are underlyingly onsets (like other inal consonants), in which case we must explain why the empty-headed syllable containing them is invisible to the trochaic stress rule, or they are word-inal codas, in which case we must account for why, of the three consonants attested, only one ([s]) regularly occurs in word-internal coda position.
The existence of word-inal codas is predicted by the analysis outlined in 5, as every nucleus licenses an optional coda consisting maximally of a single consonant. However, if we admit the existence of word-inal codas after inal [ə], we might expect clearer evidence that sonorant codas are also possible. While they are extremely rare for historical and morphological reasons, there is nevertheless some evidence to support this view. With regard to [r], the proper noun Sizer 'Cize' assonates in RolMoign with dire [ˈdi.rə] 'to say', mise [ˈmi.zə] 'put.
. ', etc. which indicates a pronunciation [ˈsi.ʦər]. With regard to [ɫ] and [n], an interesting possibility is that the proparoxytone spellings (cf. §3.6.3) actually denote a paroxytone with a inal sonorant, e.g. angele [ˈaẽn.ʤəɫ] 'angel', virgene [ˈvir.ʤən] 'virgin'. In this view, inal orthographic e is not syllabic and appears partly due to in luence from Latin spelling and partly to highlight the atypical use of a voiced segment after the post-tonic vowel. It is of course encouraging that the reconstruction corresponds exactly to the form borrowed into English. A further example showing the addition of a non-syllabic e is found in AlexisRaM. The unique form el for the object clitic le in the hemistich en terre el metent 'they place him in the ground' (AlexisRaM, l. 588) may indicate enclisis of the pronoun le, hinting at a pronunication ¹⁷ I note in passing that the present analysis has important implications for sandhi phenomena, in particular the many monosyllabic host + enclitic combinations (cf. , Rainsford 2014, all of which are compatible with either a type A or a type C analysis. In fact, in AlexisRaM, nasal place assimilation is found in nen for nem (ne + me). Further research is needed on the interaction of stress, syllable structure and sandhi in Old French.
[ˈtɛr.rəɫ] for terre el. In summary, although there is perhaps no conclusive proof that non-lexional word-inal coda consonants were possible in AlRol, assuming their existence would allow us to account for a number of otherwise exceptional surface forms in a uni ied way.
Final post-tonic [t], which never occurs as a word-internal coda, is exceptional, however it is analysed. As it only occurs in AlRol as a part of one morpheme (the -ent [ət] '3 ' suf ix), it is dif icult to ascertain the reason for its exceptionality with any con idence. It is perhaps relevant to note that the related -no '3 ' suf ix in Standard Italian is invisible for the purposes of stress assignment. Primary stress in Italian is lexically free but falls on one of the inal three syllables in the word, with the sole exception of the 3 form of verbs which are proparoxytone in the 3 , where the additional the inal -no cause stress to fall on the fourth syllable from the end of the word, e.g. telefono [te.ˈleː.fo.na] 'phone.3 ' but telefonano [te.ˈleː.fo.na.no] 'phone.3 ' (cf. Nespor 1993: 175, note 29). Were a similar analysis to be applied to -ent in AlRol, it would imply that inal [t] is in fact a inal onset consonant but that the empty-headed syllable is exceptionally invisible to the trochaic stress rule.

Conclusion
In this paper, I have argued that there was a link between syllable structure and prosodic words in a single variety of Early Old French: namely, that oxytonic word stress is only possible where there is a inal consonant which can be parsed as the onset to an empty-headed syllable. On this basis, the conservative variety of Old French in the two texts that I have described shows clear evidence that the prosodic word was still the relevant domain for syllabi ication and stress assignment in Early Old French.
The analysis relies on a number of archaic features, most signi icantly the retention of /ð/ in inal position, but also the analysis of vowel + i digraphs as Vj clusters, and the analysis of 'proparoxytone' orthography as representing sonorant codas after the post-tonic vowel. Evidence that these features are being lost is present even within the two early texts studied and they were rapidly eliminated over the course of the 12 century. The analysis is deliberately backward-looking: it sketches a prosodic system on the brink of collapse and makes clear predictions as to why a shift towards generalized surface CV syllable structure later in Old French would prove terminal for the existence of a ixed trochaic wordstress rule.

Comments invited
PiHPh relies on post-publication review of the papers that it publishes. If you have any comments on this piece, please add them to its comments site. You are encouraged to consult this site after reading the paper, as there may be comments from other readers there, and replies from the author. This paper's site is here: https://doi.org/10.2218/pihph.5.2020.4433