The Role of Prosodic Templates in Diachrony
Abstract and Keywords
Drawing evidence from Mixtec, Germanic, and Semitic, we demonstrate the role of prosodic templates in accounting for phonological and morphological change and structure. The chapter illustrates the resilience of such templates across time. Dialect data is also examined to underscore the need for a templatic approach.
It has been widely observed among the world’s languages that prosodic templates can shape lexical classes by determining canonical forms of words (e.g. McCarthy & Prince 1986, 1995, Holsinger 2000, Wiese 1996, 2001, Booij 1998). While templatic well-formedness shapes small sets of items in some languages, it defines most vocabulary in others. We argue that when viewed diachronically, prosodic templates provide a grounded explanation for otherwise divergent sound patterns across languages. We draw evidence from Mixtec, Semitic, and Germanic and show how each language or language family provides evidence for distinct roles for prosodic templates in change.
We provide a brief background in section 17.2 on templates in sound change and turn in section 17.3 to patterns best accounted for using prosodic templates. In section 17.4 we explore why forms may fail to conform to prescribed templates, highlighting the tension between prosody and other phonological and morphological forces. Lastly, dialect data in section 17.5 illustrate the resilience of templates across time in languages, where sometimes templatic effects have been expanded, while in other cases they have been lost or obscured. This section brings our discussion full circle before we close in section 17.6.
17.2 Background to Prosodic Templates and Sound Change
Until the early 1980s, theories of sound change focused overwhelmingly on the segmental or featural levels with little attention paid to structures above the segment. (p. 263) With Murray & Vennemann’s (1983) seminal article on the Preference Laws for syllable structure and Vennemann’s (1988) book elaborating on those laws, researchers began searching for motivations for sound change above the segment at the syllabic level (see Mailhammer, Restle, & Vennemann, this volume, for a contemporary consideration of such work). Drawing on Vennemann’s earlier synchronic work on Syllabic Phonology (e.g. Vennemann 1972b, 1974b, 1978), Murray & Vennemann’s new Preference Laws demonstrated that sound change can result from preferences for syllables to conform to the ideal CV shape. Seeking explanations for sound change up the prosodic hierarchy emerged on the heels of work in synchronic phonology (e.g. Vennemann 1972b, 1978), providing researchers with a means of connecting seemingly divergent sound patterns across unrelated language families. Many historical phonologists adopted these developments from synchronic phonology as has been done with other theoretical advances before and since, e.g. feature geometry.
Another key development likewise offers historical linguists a further set of tools to explain sound change and patterns across the lexicon, namely prosodic templates. Templates generally require reference to specific prosodic structures, moving our focus up yet another level in the prosodic hierarchy from the syllable to the foot. Many phonologists, starting in the late 1970s, focused on this area, with one of the best known phonological approaches to use prosodic templates being the programme initiated by McCarthy & Prince (1986, et seq.). To account for a range of synchronic patterns, including reduplication, truncation, language games, and other nonconcatenative processes, McCarthy & Prince proposed the Prosodic Morphology Hypothesis (PMH), which simply states that all such processes manipulate and refer only to ‘authentic units of prosody’. These units are defined as the hierarchical components of the prosodic hierarchy (e.g. Selkirk 1980a,b) illustrated in (1):
So what are these templates in practical terms and how do we view them here for the purposes of this chapter? We use the term ‘template’ in a broad sense, following Macken & Salmons (1997) who state that ‘a prosodic template is a conventionalized unit—a single unit—that imposes constraints on the surface form of words and, in so doing, encodes a particular relationship between words thus related’ (p. 37). In this broader sense, templates can account for divergent phenomena ranging from the required surface structure of Semitic verbs or nouns, to the choice of plural suffix in Dutch or German so that the plural form matches a specified prosodic shape, to even (p. 264) the licensing of features or segments that surface or not based on their position relative to the template (see below). By taking a more general approach to the term template, we see that predetermined prosodic patterns play a role in determining what segments or features can and will surface in the phonology and morphology of a language. Such an approach to templates has been taken elsewhere in phonology, including by adherents of Government Phonology who have made substantial use of templates in their work (e.g. Caha & Scheer 2008). Moreover, Harris (2004), Wiese (2001), and others have argued for the important role of the foot in shaping the phonology and morphology of various languages, an assumption we likewise make in our discussions of prosodic templates in this chapter. We nevertheless remain somewhat agnostic with regard to specific theoretical frameworks, to help introduce a broader audience to the benefits of appealing to prosodic templates in explaining language change.
Recent work proposes in fact that while many patterns robustly support the Prosodic Morphology Hypothesis, others show that it may be better viewed as a strong tendency rather than an absolute. Blevins (2010a) states that ‘phonological form classes… may best be modeled as statistical positive generalizations over the lexicon’. For instance, in Proto-Austronesian over 90 percent of lexical bases were bisyllabic.1 While this prevalence for bisyllabicity is certainly reflected in modern-day Austronesian languages, it is not an exception-free generalization.
Another important theme currently being explored by various researchers (Bybee 2001, Pierrehumbert 2001, Blevins 2004a, 2010a, Chitoran & Hualde 2007, Wedel 2007), and one we apply here, is that strong statistical generalizations could exert an effect on language change without being exceptionless. That is, it should not be surprising to uncover patterns that while robust are nonetheless not exhaustively followed within a particular language. Given the impressive number of cues and features language learners can attend to (e.g. Newport & Aslin 2004), we expect to find noisiness in the data, in other words exceptions to the expected patterns. Thanks to research most recently by Blevins (2004 et seq., this volume), historical linguistics is seen by some as playing a larger role in our understanding of synchronic patterns. Blevins argues that common sound patterns evidenced both synchronically and diachronically stem from the fact that languages at any point in time are simply the result of earlier language change. In other words, synchronic patterns are the result of diachronic developments. Consequently, understanding the role of prosodic templates in the history of a language helps us better account for the patterns we see synchronically. Moreover, these templates could be viewed as an additional means of automation of repeated behaviours, namely matching words to specified shapes and patterns that take hold in the language, which automation Bybee (this volume) notes plays a key role in language change and they serve as patterns that speakers use to perceive and produce language (cf. Fertig, this volume).
(p. 265) We begin here with an overview of foot types found cross-linguistically. One of the most comprehensive surveys of foot typology is found in Hayes (1995), which contains not only a wealth of data on prosodic and metrical structures in language but also examines attested patterns, discovering an unexpected asymmetry (see also Lahiri, this volume, for a discussion of foot typology, and of the role of feet in phonological change). Of four logical possibilities, just three foot types appear to be prevalent cross-linguistically: the moraic trochee, the syllabic trochee, and the moraic iamb. Interestingly, the syllabic iamb is missing. Again, while robust patterns exist, further study frequently reveals that ‘universals’ are less universal than originally thought. We advocate a similar view here for templates in general: templatic effects can predominate in a language even if they are not absolute. Moreover, while templatically-driven changes can in principle apply throughout an entire language, they tend to apply within a given set of lexical classes (Smith 2004, 2007b, 2009, Ussishkin 2005, Blevins 2010a).
Of relevance here are the syllabic and moraic trochee. The syllabic trochee consists of a stressed-unstressed syllable sequence as in (2a). Foot boundaries are indicated by ‘[ ]’:
The moraic trochee, on the other hand, is based on the amount of segmental material found within a syllable’s rhyme, material referred to as morae (μ). When rhymes consist of VC, e.g. [gas], or a single long vowel, V̄ , e.g. [thrā], they are considered minimally bimoraic and form a heavy syllable (H). These heavy syllables are capable of forming their own foot (cf. 2bi). A monomoraic syllable containing just a short vowel is considered light, unable to form a foot on its own, e.g. the final syllable ti in gasti above in (2bi). However, a sequence of two light syllables (LL), e.g. ste+di with their two morae combined can form a foot equal to that of a single heavy syllable (cf. 2bii) (cf. Hayes 1995).2 Foot structure provides one means speakers use to perceive and segment sounds and words, starting already in the first year of life (Jusczyk et al. 1999).
(p. 266) Previous research has demonstrated repeatedly that phonetic considerations, combined with potential misperceptions (Ohala 1981, 1989, Blevins 2004a, this volume, Blevins & Wedel 2009; see also Yu, this volume) may play a major role in language change. One of our goals is to complement this work with arguments that language change can also be shaped by prosodic considerations that can be captured with a template.
17.3 Role of Prosodic Templates in (Morpho-)Phonological Change
Drawing on evidence from three unrelated language families, we illustrate the range of language changes and structures for which prosodic templates can account. While not all language families or languages exhibit prosodically-driven changes, some languages do; the evidence below thus shows the breadth of languages and structures that are templatically-shaped, underscoring the fact that these templates are neither accidental nor coincidental. We start with morphological and phonological evidence from Semitic, before turning to lenition, onset licensing and harmony in Highlands Mixtec. We then conclude with historical and dialect examples from West Germanic.
17.3.1 Well-known Templatic Properties: Semitic
We begin with the Semitic sub-group of the Afro-Asiatic language family. Semitic languages are well-known for what has become known in the literature as ‘root-and-pattern morphology’, a sub-type of nonconcatenative morphology, a more general label for morphological processes requiring an interface between morphology and phonology. This interface is manifested via the persistent and pervasive template patterns found across the Semitic languages. Much evidence has accumulated in the synchronic literature concerning templatic effects in Semitic (e.g. McCarthy 1979, 1981, Ussishkin 2005, Bat-El 2011). Perhaps the best-known examples from Semitic concern verbal classes (known in the Semitic literature as binyanim), which are defined by templates holding consistently throughout various categories. In general, the template tends to correlate with syntactic properties, though not always (Laks 2010, 2011). Well-known verbal templatic structures are found in a diverse array of languages, including the Semitic stratum of verbs in Maltese, all verbs in Arabic and Hebrew, and verbs in Ethio-Semitic languages (which are templatic despite failing to conform to the binyan system in Arabic and Hebrew). Other morphological categories are likewise marked by templates; nouns also exhibit numerous templatic properties, though across Semitic nouns seem to not be limited to as strict a set of templates as verbs.
One question in diachronic approaches concerns how these templates arose. Furthermore, it is important to examine what tendencies exist across Semitic languages (p. 267) and dialects. Recent evidence from dialectal study reveals some robust patterns. Simpson (2009) presents a vast array of dialectal data from numerous Semitic dialects showing that a single type of change (vowel deletion and reduction) results in modification of templates from previous stages of the language but reinforces different templates, affecting verbal systems in particular. The result is that where the change has taken place, the inflected verbs that are affected lose a syllable of the stem template, resulting in CCVC-stems. To begin, consider the Classical Arabic inflected perfective verbal paradigm in (3):
Note that in Classical Arabic, the verbal stem (katab) conforms consistently to a bisyllabic template. However, across numerous dialects of Arabic descended from Classical Arabic, vowel reduction is a common occurrence. According to Simpson (2009), such reduction usually targets either the final stem vowel under vowel-initial suffixation (e.g. Classical Arabic kátab-at > Jewish Tunisian kə́tb-ət ‘to write fem.sg.perf.’) or the first stem vowel under specific prosodic conditions (when the final vowel attracts stress; see data in (4)):
In (4), the stem consistently surfaces as monosyllabic, so the typical bisyllabic Classical Arabic template does not hold for this dialect. Rather than a trochee consisting of [CV.CV] (the final C in CVCVC is extrametrical; see McCarthy & Prince 1995), in this dialect the stem becomes a monosyllabic moraic trochee of the shape CVCC. Similar (p. 268) changes are found in numerous related forms of Arabic described in detail by Simpson (2009), including, the Al-Mahabšeh dialect of Yemen, Cairene, Upper Egyptian, and Khartoum Arabic, the Levantine dialects of Palmyra and Baksinta, Meccan Arabic, Yemeni southern plateau Arabic, Baghdad Christian and Jewish Arabic, Eastern Libyan Arabic, and Kuwaiti Arabic. In addition, Simpson documents such changes in Babylonian and Palestinian Jewish Aramaic, other Aramaic dialects, Biblical Hebrew, Ge’ez, and Tigre.
These phonologically induced changes result in a system that maintains consonantal melodies consistently across paradigms of related words as well as a consistent prosodic pattern. As Simpson (2009) argues, these outcomes result from reinterpretation over time of segmentally-conditioned alternations and prosodically-conditioned alternations. Templatic behaviour thus emerges as the result of these alternations over time.
It is well-known that Arabic makes use of the foot in its templatic structure. McCarthy & Prince (1990) and Hayes (1995), among others, have observed before that moraic trochees explain metrical and stress facts in Classical Arabic and in many spoken varieties. This observation forms the core of Ussishkin’s (2000) prosodic account of Classical Arabic verb structure, where the productive verbal classes are seen to involve a single moraic trochee in uninflected stems.
17.3.2 Prosodic Templates in Highlands Mixtec
In one of the first diachronic analyses to incorporate a prosodic template, Macken & Salmons (1997) demonstrated that a series of sound changes and patterns found in Highlands Mixtec could not be explained solely based on syllables. Instead, these changes could be accounted for using a bisyllabic template. An Otomanguean language spoken in southern Mexico, Highland Mixtec displays templatic effects in the behaviour of glides, sC clusters, lenition of medial consonants, vowel harmony and tonal melodies. As they argue, ‘the distinction between the foot template and other syllable sequences in the grammar is crucial for all [these] phonological processes [… ], because non-template syllables fail to participate in those processes which should otherwise be available to monosyllabic morphemes’ (p. 42). Their template is provided in (5).
We highlight a few trends that can be related to this template.
(p. 269) 22.214.171.124 Loss of Template-medial Consonants
In both the San Miguel and Chalcatongo dialects, medial consonants in ivi sequences are lost but to differing degrees. In the examples in (6), Chalcatongo is more innovative, eliminating medial /b/, i.e. [v], from sequences that surface as Cɨvɨ in San Miguel (Macken & Salmons 1997: 456):
According to Macken & Salmons, Chalcatongo Mixtec lost labial obstruents in template-medial position. As the examples in (6a) show, this loss would have resulted in dispreferred syllable structure CVCV → CVV à la Murray & Vennemann (1983) with the loss of the consonant in syllable onset position. This runs counter to Vennemann’s (1988: 13–21) Head Law which states that the syllable onset is a position of strengthening since the greater the consonantal strength of an onset, the more preferred it will be (cf. Mailhammer et al., this volume). Although the labial glides have indeed strengthened to /b/ in word-initial syllable onsets, they have conversely undergone deletion in syllable onsets word-medially. In sum, syllable structure cannot account for these data. If the syllable alone were responsible, then we would expect strengthening and retention of /b/ in syllable onsets regardless of whether these onsets were in word-initial or medial-position. When viewed from a templatic perspective, however, the explanation is more straightforward. Simply put, [v] is retained in template-initial position, but lost in template-medial position. Macken & Salmons (1997: 51) explain this as follows: ‘Synchronically, the tendency is for the first syllable of the template to carry substantially more phonological material than the second […] just as the tendency toward CVV roots reflects the second-position weakening tendencies’.
Syllable structure also fails to account for the distribution of sC-clusters. While many have attributed (s)(C)V sequences as part of the syllable canon, this approach misses a critical generalization, namely that sC-clusters are not found in all syllable onsets. These (p. 270) clusters are restricted to template-initial position.3 Arising from sV sequences prefixed onto roots and still evidenced in the alternation between s- and sa-causatives, the loss of the prefix vowel worsens syllable structure, e.g. sVCVCV > sCVCV. This change, however, prosodically reshapes the words such that the preferred disyllabic structure at the very heart of the template in Mixtec is reestablished. Indeed adherence to the foot template trumps syllabic preferences. But perhaps most notably, vowel loss creates these new clusters in the grammar in precisely the strongest prosodic position where the most contrasts and phonological material are licensed, namely in template-initial position, the same position where glides were strengthened to [b], rather than lost.
126.96.36.199 Vowel Harmony
A number of other developments and patterns can also be explained using the same template already proposed above for Chalcatongo. One such pattern among innovative speakers in Chalcatongo is vowel harmony that typically eliminates the contrast between /i/ and /ɨ/ found in San Miguel in favour of /ɨ/:
While /ɨ/ can co-occur with the so-called ‘outer vowels’ /i, a, u/ in San Miguel as illustrated by (7a) and (b), evidence from Chalcatongo Mixtec demonstrates the ‘increasingly restrictive limits on the amount and type of phonological material the prosodic template can carry’ (p. 52). An autosegmental account of vowel harmony can be applied across the template, where ‘/i/ assimilates to the backness of the /ɨ/’ (p. 55) regardless of which syllable the /i/ is initially found in. In other words, within the template, just one single value is licensed for the feature [back].
Despite superficial differences between these seemingly divergent phonological changes, the Mixtec analysis illustrates how one common structure, namely the prosodic template, can shape a language. While syllabic analyses encounter difficulties accounting for the individual changes, peering up the prosodic hierarchy to the foot level not only helps account for individual problems such as sC-clusters and glide strengthening (p. 271) and loss, but it helps unify a broader set of sound patterns and developments, including vowel harmony.
17.3.3 Prosodic Templates in Germanic
Germanic has been the focus of many analyses employing the prosodic template in recent years both synchronically, e.g. German and Dutch plurals (Booij 1995, 1998, Wiese 1996, 2001), and diachronically, e.g. lenition, vowel loss and reduction (Holsinger 2000, 2009, Smith 2004, 2007a,b, 2009, etc.). Perhaps most striking is the resilience of the foot-based template to shape entire classes of words or sequences of sounds across the history of the West Germanic languages despite the fact that the trochee at the heart of the template has changed from moraic in the earliest Germanic languages to the syllabic trochee of the modern languages (on our analysis; compare Lahiri, this volume). Data below illustrate this resilience with a few examples.
188.8.131.52 Vowel Loss and Reduction: Old Saxon i-Stem Nouns
For more than a thousand years, West Germanic vowels in unstressed syllables have been subject to reduction and loss, particularly word-finally, as described in the Auslautgesetze or Laws of Finals (cf. Prokosch 1939, Boutkan 1995) resulting in the loss of inflectional endings and ultimately the collapse of entire verb and noun classes. While all vowels were subject to the Auslautgesetze, i and u were initially more resilient to loss and reduction than others. Prokosch (1939: 134) noted that these vowels were first lost ‘after a long syllable, or after two syllables (which phonetically, or metrically, amounts to the same thing)’ before they disappeared after short syllables. Prosody’s critical role in loss and retention is perhaps best illustrated by Old Saxon (OS) i-stem nouns. Germanic nouns were historically formed from three parts, namely a root+thematic vowel+inflectional ending (number and case), e.g. OS ferd+i+o ‘journey, pl. gen.’ With the loss of many inflectional endings, the thematic vowels, especially in the nominative and accusative singular forms, became exposed word-finally and subject to additional loss.4 As illustrated in (8), the short final i was lost after heavy stems (or so-called long syllables), i.e. those ending in either VCC (e.g. fard ‘journey’) or V̄ C (quān ‘woman’), but retained after light stems, i.e. stressed short syllables, i.e. stems ending in VC, e.g. uuini ‘friend’ and stedi ‘city’.5
(p. 272) (8)
By appealing to a template based on the moraic foot rather than solely on the shape of the preceding stem syllable, a simple generalisation emerges regardless of whether the word is monosyllabic, e.g. +wurmi6 >wurm ‘worm’, or polysyllabic, e.g. friundskepi ‘friendship’.
In the case of the heavy stem nouns in (9), the heavy syllable of the root formed its own foot leaving the light syllable containing the thematic vowel i unfooted. This unfooted i was consequently lost. Conversely, for the light stem nouns, the syllable containing the i was footed with the preceding light syllable thereby forming a LL foot via resolution.7 By eliminating the unfooted vowels, the noun stems fit the trochaic foot template shaping this lexical class:
Indeed the preservation and loss of i was driven by the need to fit the words of this noun class to the trochaic template. Words were not all shortened alike; only (p. 273) the [H]L > [H] change was necessary to eliminate prosodically extraneous material. Since the [LL] sequence already satisfied the template, the footed thematic vowel was preserved in these words for a longer time. The template was thus the means of reshaping words: vowels which were not mapped to the template were unlicenced and thus lost.8 This analysis situates this sound change within the broader context, providing a more unified approach to understanding prosodically-driven changes in Germanic.
184.108.40.206 Consonants: Lenition of Medial Stops in German Dialects
Holsinger (2000, 2009) has demonstrated that lenition processes in West Germanic as divergent as elision, flapping, and spirantization as well as fortition processes can be unified in terms of a prosodic template based on the trochaic foot. Consider the lenition data in (11):
In these examples, [d] is retained word-finally in nasal+obstruent clusters, e.g. blind, but lost medially although the stop is in syllable-onset position, typically a position of strengthening (Vennemann 1988). Holsinger (2000, 2009) argues that the full range of phonemic contrasts, e.g. /t/ versus /d/, in templatically-driven dialects are exhibited primarily in template-initial position. Template-medially, on the other hand, these contrasts are neutralized or the obstruent is lost even if in syllable-onset position. Just as for Mixtec above, medial onsets are considered the weak branch of the template and are consequently the target site for lenition processes (cf. (12)). Holsinger refers to these types of patterns as ‘weak position constraints’ where phonological features like [spread glottis], are fully expressed in some prosodic positions, such as foot-initially, but weakly expressed if not lost entirely in others, e.g. foot-medially as seen in (12).
(p. 274) (12)
As with Mixtec, analyses based on syllable structure alone cannot account for loss of obstruents in syllable onsets in word-medial position. Within the prosodic template, however, the onset of the strong branch is the sole position in which the full range of features and contrasts is retained. Consequently, a templatic analysis provides the most straightforward account of these lenition processes akin to those found in Mixtec above.
17.4 Template Blocking
Although templates can appear to be imposed across all forms, documented cases exist in which a template fails to apply as expected. In each instance clear morphological or phonological factors prevent the template’s application. In this section, we outline examples from both the Semitic and West Germanic languages.
17.4.1 Template Blocking in Semitic
In Hebrew and Arabic, data which fail to conform to the prescribed template fall into two categories. The first category includes morphologically-based template blocking: in Hebrew, verbal stems are overwhelmingly bisyllabic, conforming to the optimal prosodic word (Ussishkin 2000, 2005, Bat-El 2003). However, of the seven verbal categories of binyanim found in the language, one of them has a trisyllabic stem: the hitpa’el binyan. All verbs in this class have three syllables, rather than two. As Ussishkin (2005) explains, the reason for this over-large word is morphologically-based. Namely, the affixal material for the hitpa’el class contains three vowels, each of which becomes a syllable nucleus, and the presence of this material overrides the bisyllabic template observed in the remaining verbal categories.
The second category of template blocking is phonologically-based. Laks (2010) documents cases of unattested verbs in the verbal systems of Hebrew and Palestinian Arabic. (p. 275) These gaps are unexpected, and have no sensible morphosyntactic basis; rather, Laks (2010) argues that these gaps are the result of what would otherwise be phonologically ill-formed words. The following examples from Laks (2010: 6) illustrate a case of otherwise productive decausativization of a pi’el verb to a hitpa’el verb being blocked; compare well-formed hitlaxlex with questionable *hittanef.
In fact, such blocking occurs consistently when the verbal stem is t- or d-initial, and when such forms are morphologically needed, other verbal categories must be called upon to fill the gap. For example, the verb diber ‘speak’ has a derived reciprocal form in the nif’al binyan, since a hitpa’el form would violate phonotactic restrictions against two adjacent coronal stops, resulting instead in the less common nif’al form nidbar ‘talk to one another’. As Laks (2010) observes, a pi’el-nif’al pair is highly unusual and in fact is always the result of a blocked form in another binyan.
Turning to the examples in (14) from the Maltese dialect of Marsaxlokk, on the southeast coast of the island of Malta, inflected forms conform to a bisyllabic template, even when this template is achieved by two medial adjacent consonants.9
However, a clearly non-etymological vowel is usually epenthesized to break up triconsonantal clusters arising as a reflex of affixation. This results in some trisyllabic forms, as seen in the dialect data documented by Schabert (1976):
(p. 276) (14b)
Similar phonotactically-based blocking occurs in Palestinian Arabic, as also documented by Laks (2010). Palestinian Arabic disallows nn and nm clusters, and when such clusters might otherwise result from the application of a productive derivational process, the derived verb occurs in a different binyan. This manifests in the seventh binyan, usually used for passive verbs. To illustrate, consider the active form nafa ‘deny’; as an n-initial form, the otherwise expected passive form *innafa is disallowed. As in Hebrew, occasionally some forms find a way to surface, but always in a different binyan, one otherwise not used for expressing passives. In this way, nisi ‘forget’ has a related passive form intasa ‘be forgotten’, rather than the otherwise expected *innasa.
17.4.2 Old High German jan-Verbs
Data from Old High German (OHG) jan-verbs likewise reveal that the template can be overridden by phonotactics and morphological transparency. In Germanic the weak verbs formed their preterite by adding a dental preterite ending, e.g. -ta or -da,10 to the verb root + connecting vowel. For the jan-verbs the connecting vowel was i, e.g. den+i+ta ‘lengthened’, hoor+i+ta ‘heard’. In many if not most West Germanic languages this connecting vowel was lost when unfooted after a heavy stem, but retained when footed after a light stem as in (15):
(p. 277) In short, the dental preterite ending was affixed to the right edge of the foot-based template. Any connecting vowel left unfooted, i.e. unmapped to the template would be lost permitting the preterite ending to affix directly to the foot edge. While a number of exceptions can be found, these exceptions systematically prevent phonotactic violations and difficult sound combinations or they help to maintain morphological transparency.11 Consider the following examples:
The expected preterite forms would have been subject to i-loss following the heavy verb stems. However, as the asterisk indicates, these forms are unattested.13 Simply, if syncope occurred, it would lead to strongly dispreferred clusters, e.g. -hnt- and -snt- violating West Germanic phonotactics. Speakers had two processes available to avoid these clusters. First, i-syncope could be blocked altogether (16a) or an anaptyctic vowel, typically a, could be inserted to break up the cluster. Interestingly, the infinitive (p. 278) irtruosanen already had an a before the stem final n akin to the output of epenthesis for pauhhanta. With the expected loss of i following this stem, the a already found in the infinitive form could have served to avoid any dispreferred cluster from arising in the first place. Yet when this vowel did not appear in the preterite, the alternative process, namely the blocking of i-loss, had the similar result of avoiding the dispreferred cluster -snt-.
In another set of apparent exceptions, i is retained following heavy or polysyllabic stems ending in d or t. These exceptions, however, are more restricted in their occurrence and typically limited to a specific set of manuscripts:
The failure of i to undergo syncope may be the result of the Obligatory Contour Principle (OCP) which ‘prohibit[s] adjacent identical elements on the gestural tier’ (Broselow 1995: 179; cf. also McCarthy 1986). Assuming voicing assimilation of -dt- or -td- to either -dd- or -tt-, the resulting gestures would be identical. To avoid the adjacency of two such identical segments, syncope of i could be blocked as the examples above illustrate. The result would be stems which violate the prosodic template since the unsyncopated i would be in an unfooted syllable:
In these dialects the avoidance of OCP violations has a clear priority over strict adherence to the prosodic template. The failure to syncopate i helps maintain morphological transparency between the stems and preterite endings, since the t or d of the stem is kept distinct from the dental of the preterite suffix.
The examples from Semitic and Germanic thus illustrate the conflict between different levels of the prosodic hierarchy and the ability of lower level phonotactic requirements or morphological transparency to override the prosodic template.
(p. 279) 17.5 Resilience, Expansion, Loss of Templates: The Importance of Dialect Data
One crucial area of study that helps elucidate how templates behave when viewed from a diachronic perspective is the realm of non-standard or dialectal language varieties. These varieties provide scholars with a rich array of data that provide key insights into the nature of structural properties in language. With respect to prosodic templates, data of this type can help us figure out whether templatic restrictions are in a stable state at a given time within a language. In this section, we examine templates from this perspective for Semitic and Germanic.
17.5.1 Template Resilience and Productivity in Semitic
Examples from the Semitic family illustrate just how pervasive templatic structure is in these languages. The clearest cases come from the Semitic verbal systems, in which all verbs must conform to a template.
In Hebrew, templatic requirements on verbs are stronger than on nouns. All Hebrew verbs that trace their source to a borrowed word are actually denominal, having been formed on the basis of a borrowed noun in the language. Such nouns are borrowed with no templatic modification, nor any prosodic modification for that matter, such that even the stress tends to correlate with the stress in the source language. Bat-El (1994) provides numerous data illustrating this phenomenon, including those given in (19):
When such Hebrew nouns become verbs, they are required by the grammar to conform to the rigid templatic system which includes seven binyans or verbal categories, resulting in rigid application of prosodic structure:
Template resilience can also be seen in Maltese, where heavy lexical borrowing has occurred. In Maltese, Italian verbs that were borrowed relatively long ago conform to the binyan system of Maltese, unlike recent borrowings which undergo circumfixation of i- -ja. For instance, consider the following pairs of forms:
Most telling is the third example, which shows the noun-verb pair serp ‘snake’/serrep ‘he zigzagged’. Mifsud (1995) notes that only verbs borrowed relatively long ago behave this way; the more typical pattern for borrowed verbs in Maltese is to ignore the Semitic-based templatic requirements and simply append a concatenative circumfix, as in the form ittowja ‘he thawed’.
Additional data can be found in Palestinian Arabic, which as documented by Laks (2008, forthcoming) shows verb innovations that always conform to the existing verbal templates. Laks has documented such cases as attested in Elihay (2005), and as the data below illustrate, such forms tend to occur productively in the second verbal binyan of the language, with the template CVCCVC:
(p. 281) Other examples can be found in verbal binyan 5, with the template tCVCCVC, the result of not only templatic resilience but also morpho-syntactic considerations (this binyan tends to host verbs derived by thematic operations such as decausativization, reflexivization, and reciprocalization):
17.5.2 Expansion and Loss of the Prosodic Template: Dutch Diminutives
As the examples from Old Saxon and Old High German illustrate, the trochee played a key role in shaping both nouns and verbs historically. Although the type of trochee changed from moraic to syllabic, and the lexical classes impacted 500‒1000 years ago no (p. 282) longer exist, e.g. i-stems in Germanic are long gone and no longer exist in their previous form, the trochee nevertheless continues to play a role shaping lexical classes in the modern daughter languages. Notably data from the dialects provide clearer insights into the role of the prosodic templates in the actual dialects.
In Dutch, the diminutive ending -(T)je14 is attached to the end of nouns. However, nouns ending in sonorants have one additional stipulation: an additional schwa appears after light noun stems, e.g. ball-e-tje ‘little ball’, i.e. those ending in VC, but no schwa occurs after heavy noun stems, e.g. laan-tje ‘little lane’, i.e. those ending in V–C or VCC, or disyllables, e.g. bakker-tje ‘little baker’.
Bisyllabic (σσ) and heavy (H) stems pattern together in taking the -Tje diminutive ending, while light stems have both schwa and -Tje:
Recall from above that [σ́ σ] (where σ1 is typically L, σ2=L or H) was equivalent to [H] in early Germanic, forming a trochee via resolution, although in that case, [H] referred to a heavy syllable rather than a heavy stem comprised of superheavy syllable as argued for here for Dutch.17 L. C. Smith (2009) thus argues that -Tje is simply affixed to the right edge of the prosodic template requiring the noun to end either in a heavy stem, i.e. superheavy syllable (laan+tje) or a bisyllabic stem (bezemp+je). However, when the noun stem is L, e.g. bal, unable to form a foot on its own, schwa expands the stem so that it fits the prosodic template allowing the diminutive suffix to attach to σ́ σ, i.e. balle-. Although a L stem ending in VC could arguably form its own foot, it undergoes epenthesis to conform to the specific shape of the prosodic templates [H], i.e. VCC or V:C, or (p. 283) [σ́ σ] required for suffixation. In sum, schwa epenthesis lengthens light stems to fit the prosodic template necessary for diminutivization.
This template is akin to that used for OS i-stem nouns over a thousand years ago, where the prosodic equivalence between H and σ́ σ motivated the modification of noun stems. This is notable since the foot of Modern Dutch is generally considered to be syllabic. Yet these heavy stems, in this case resulting from superheavy syllables, continue to play a role in Dutch prosody today. While the role of these superheavy syllables is known in the accent system, they also play a role in the dialects in diminutive formation. Here, light noun stems are expanded not by the use of schwa, but rather by lengthening the stem vowel to create a [H] stem:18
Indeed, in some dialects two competing forms may co-exist as the example for bal ‘ball’ from the Drentish dialect of Hoogeveen illustrates (p.c. Geert Booij):
In these dialects speakers can either lengthen the stem vowel or add schwa to ensure the noun stem fits the prescribed template for diminutive formation.
In other dialects, the role of the template itself is changing. It has either been extended beyond those stems ending in sonorants, e.g. Huizen, or it is falling out of use altogether, e.g. in Formerum. Compare light stem nouns from these two dialects in (28). Forms differing from Standard Dutch diminutives are highlighted.
(p. 284) (28)
As the Huizen examples illustrate, schwa epenthesis occurs not only as expected with stems ending in sonorants, but it has also been extended to apply to other light nouns ending in obstruents. Conversely, the Formerum data reveals that other than for karretje [kɑrəkə], schwa epenthesis does not regularly occur where it would be expected after light stems ending in sonorants. Whereas the use of the diminutive template has been extended in Huizen, just 100km away in Formerum, the template has all but fallen out of use in the latter location. Moreover, despite the shift to a syllabic trochee from the moraic trochee of West Germanic elsewhere in Dutch, we still see the use of a quantity sensitive stem at work in the language shaping diminutive formation.
We have examined the role played by prosodic templates (and thus by the foot) in language change. Data from numerous languages in three unrelated language families show pervasive effects of foot-based prosodic templates in shaping the outcome of language change over time. One of the most important results is to demonstrate that no matter how different the genesis of the respective templates investigated in the individual languages and language families, the templates that ultimately emerge are a crucial factor in shaping a variety of morphological classes and licensing features and segments in various template positions. These templates provide a unified approach to divergent language patterns not only within languages but cross-linguistically. Indeed what has (p. 285) been well known to guide our understanding of the Semitic languages can provide insights that help us comprehend patterns in Germanic and elsewhere.
In the end, we stand firmly with Blevins (2010a) in calling for additional research on prosodic templates, and importantly, we have justified why such research should draw on data centering on language change, as well as on data from numerous varieties of spoken language, including dialectal data. Like much current work in linguistic theory, our research here validates approaches that value data from previously understudied language varieties. (p. 286)
(1) We acknowledge that reconstructions are not attested, and that comparative reconstruction often introduces biases, including shaping a proto-lexicon into something more uniform in some ways than the actual language might have been (see Fox, this volume for a discussion of issues related to reconstruction).
(3) Although Macken & Salmons (1997) provide a detailed explanation of sC-clusters, there are unfortunately no explicit examples provided in their article. It is for that reason that no examples are provided in this section.
(4) The consonants z and n forming the singular nominative and accusative inflectional endings respectively were prone to loss (cf. Bammesberger 1990), e.g. +stadiz > OS stedi ‘place, town’.
(5) Long i, e.g. i formed by coalescence of i with a following j, was not subject to loss like its short counterpart. It could be argued that the long vowel was able to form a foot on its own thus protecting it longer. Additionally, long vowels typically underwent reduction to a short vowel over time while short vowels often underwent loss. Moreover, short vowels followed by additional endings, e.g. OS enstim ‘honour, dat. pl.’ were protected within the trochaic ending.
(6) A raised “+” signifies a reconstructed form.
(7) See Lahiri (this volume), for a discussion of foot structure and resolution in Germanic. Although Lahiri’s Germanic foot differs somewhat from that used here, the principle of resolution at work in the West Germanic languages is nevertheless highlighted.
(9) Many thanks to Andrew Simpson for providing the Maltese data cited here.
(11) While Smith & Pulsipher (2008) have shown that geminates found in the present tense did under certain circumstances lead light stem verbs to be treated as heavy, this is beyond the scope of the chapter.
(13) The superscript “+” indicates a reconstructed form, while “*” suggests an impossible unattested form that should not be reconstructed.
(14) T represents a voiceless stop that matches the place of articulation of the preceding consonant, or appears as t after a vowel.
(16) In this environment, the dental-initial suffix -tje is affixed without additional assimilation as suggested by -Tje.
(17) The reader should note that light and heavy refer to stem types not syllable types. Light stems end in either light or heavy syllables, whereas heavy stems end in superheavy syllables. Superheavy syllables play an important role in Dutch prosody in that they can attract stress which would typically be assigned via the syllabic trochee. In this way, these superheavy syllables do exist on par elsewhere in the language with the syllabic trochee (cf. Booij 1995 and van Oostendorp 2002 for details regarding Dutch stress placement and superheavy syllables).
(18) Dutch dialect data come from the Goeman-Taeldeman-Van Reenen project supported by the Meertens Institute and published as Morfologische atlas van de Nederlandse dialecten (MAND). This project recorded speakers from 613 dialects in the Netherlands, Belgium, and Friesland from 1980 to 1995.
(19) h in this example represents a lengthened vowel.