15 Change in Word Prosody: Stress and Quantity Purchased

Diachronic data has become consistently more relevant to models of linguistic structure. However, segmental developments dominate phonological discussions compared to word prominence and stress, and stress systems are generally less meticulously described and discussed. Textbooks rarely index changes in stress or metrical systems, although related phenomena like quantity and weight may be discussed at length. In this chapter, we address change in stress systems under the broader perspective of prosody, meaning all aspects of phonological grammar that relate to phonological quantity, syllable structure, foot structure and stress.¹ While prosodic changes are principally phonological, they can be affected by morphological structure as well. Phonological properties of affixes and phonological domains that result from root-cum-affixes affect prosodic organization. For instance, an affix may provide a domain which would trigger shortening, deletion or gemination and have serious prosodic consequences (cf. English suffix {-ity} triggers Trisyllabic Shortening as in sane~sanity, or OE causative suffix /j/ which causes gemination tem-j-an > temman ‘to tame’). Alternatively, a difference in stress organization may express differences in morphological categories, as in tormént_verb  ^~  tórment_noun.

Analyses of synchronic phonological systems make a case for assuming that assignment of stress is part of a complex system of metrical organization rather than the realization of emphasis on a particular vowel. Metrical organization includes foot construction, assembling syllables into feet, and edge prominence. An alteration in any one part may lead to a change in the entire system. A major difficulty in understanding prosodic change is that even if the outward realization of stress has not changed, the underlying prosodic organization may have changed. A trisyllabic word, with a sequence of three light syllables, bearing main stress on the middle syllable, could be analysed as a trochaic or an iambic pattern, with an initial or final unfooted syllable. The ambiguity is due to the fact that this prominence could be obtained depending on whether the foot is aligned to the right or the left edge of the word, as in (1). Only a larger data set can determine the real organization.

Motives for sound change are diverse, as explored in many chapters of this volume, especially articulatory, acoustic and/or auditory ones (see especially Donegan & Nathan, Bybee, and Blevins, all this volume) and phonological ones (see especially Dresher, Purnell, & Raimy, and Kiparsky, all this volume). For word prominence, however, uncertainties of articulatory and acoustic correlates of stress make it more difficult to directly relate stress change to phonetic causes alone. It is equally difficult to reliably compare the phonetic correlates of stress differences between dialects. Nevertheless, native listeners are aware of prosodic differences across dialects, and orthoepists, then and now, have recorded certain fixed patterns of what (they think) they heard.

This chapter highlights the consequences of normal language transmission through generations, interacting with language contact and the nativization of loans leading to prosodic reorganization. For any change of stress pattern to become part of the phonological grammar, the underlying prosodic system must reorganize, or else the shift will be unstable. Lahiri (2002) and Dresher & Lahiri (2005) drew attention to a characteristic of grammars called pertinacity. A rule or pattern may persist over time, though its realization may change. This type of pertinacity can be summed up as same pattern, different output realization. A converse kind of pertinacity exists where output forms persist despite changes in the grammar, which involve a reanalysis of the output form, triggered by changes elsewhere in the system. This type of pertinacity can be summed up as different pattern, same output realization. Since learners acquire their grammars guided by the output forms they are exposed to, we do not expect these forms—especially those that make up the ‘core’ or ‘primary’ data—to change in the course of acquisition.² Reanalysis of grammar that does not involve an immediate change in output forms is thus a significant type of language change. Changes in stress systems are prime examples of both types of pertinacity.

Word stress is not merely a matter of locating the syllable which carries maximum prominence. Word prosody implies metrical organization composed of independent, but interrelated parameters. A description of the parameters that provide underlying metrical organization enables us to provide a typology of possible types of word-prosodic change. In section 15.4, we draw data specifically from Old English (OE) to Middle English (ME), drawing attention to the overarching principle of pertinacity which constrains language learners from making abrupt changes.

In languages where word stress is significant, every word has one main stress, typically aligned with one edge of a word. Stress contrasts predictably require more than one syllable to be discernible. The term ‘phonemic’ stress usually implies that minimal pairs exist which contrast only in stress, e.g. pérmit_noun  ~ permít_verb or German áugust ~ Augúst ‘proper name ~ month’. Languages with ‘fixed’ stress allow a specific syllable at one edge of a word to bear main stress; for example, in Bengali and Finnish main stress falls always on the initial syllable, while word-final stress is typical for French or Turkish. However, even in languages where stress can occur more ‘freely’, main prominence is restricted within a three-syllable window at a word edge. Main stress in the middle of a word would be fortuitous in the sense that this may happen only when the centre of a word converges with an edge. In a five-syllable word like univérsity, stress falls on the middle syllable, the third-syllable from either edge (cf. Hayes 1995). As we shall see, the preference in English happens to be the right edge of a word.

Although stress prominence appears on a syllable, the organization of stress patterns involves a larger constituent, the foot. Metrical feet group syllables together based on organized principles and are the best means of characterizing stress patterns. Complex stress patterns are difficult to account for without metrical organization, particularly when words carry secondary stresses. For instance, còriánder and aspáragus have four syllables each, but main stress falls on different parts of the word and only the former has secondary stress. The pattern can be accounted for by assuming that (còri)(án)der has two feet, with main stress on the right foot while a(spára)gus has only one.

Thus, we need to distinguish factors such as syllable quantity and weight, a typology of feet, and principles for grouping syllables into feet. The correlation between a syllable’s internal structure and its weight is language dependent, but in languages where syllable weight plays a significant role, syllables with long vowels and closed syllables usually count as heavy.³ Headedness (which edge of the foot is strong) allows for two types of feet, left-headed trochees and right-headed iambs (2a).⁴ Two further considerations are necessary in grouping syllables into feet, direction of parsing (2b) and the word-edge which is most prominent (2c). An additional factor necessary in some languages is the potential inertia of a final prosodic constituent (2d).⁵ These are summarized in (2).

In what follows, we show that a change in any of these parameters need not necessarily produce a difference in surface prominence. Only a subset of words in a language provides the critical evidence for setting the parameters. For instance, on the basis of a word such as húrricàne one cannot determine whether the foot type in English is an iamb or a trochee, since there are two feet each of which is a heavy syllable. After reviewing proposed foot types, we show how differences in parameter settings may or may not reveal surface differences and how identical surface stress patterns may be analysed by different parameters. We argue that these are precisely the type of situations which may lead to a change in metrical organization.

In quantity-sensitive systems, ideally feet are asymmetric such that the head of a foot is always aligned to a heavy syllable. However, if only light syllables are available, then these would be grouped into feet as well.

Other than the basic foot types in (3), evidence supports some variation in the above claims.⁶ A moraic trochee as suggested in Hayes (1995) is not asymmetric; i.e. if the head of the foot is bimoraic, then the foot itself must consist only of one syllable.

The difference between an asymmetric trochee and a moraic trochee can be seen by grouping sequences of light (L) and heavy (H) syllables into feet as in (5). The first three words from the left are parsed identically, but the last two show marked differences. (d) has two feet under the moraic trochee analysis but one foot in the asymmetric grouping. As for (e), both allow two feet with identical prominence patterns, but in the moraic trochee analysis the medial light syllable remains unfooted, which may have different consequences if such syllables are reduced or deleted.

With identical feet, the direction of parsing can lead to crucial differences in grouping. (6) illustrates a series of parsing possibilities with the aid of moraic trochees and iambs. A word with a [HLL] sequence shows identical grouping for both moraic trochee and iamb irrespective of the direction of parsing. However, for the other words, we find major differences.

Consequently, identical sequences of heavy and light syllable words may reveal different foot structures following a change in the direction of parsing, but not always. Sequences like [LLLHL] or [HLL] are insufficient to determine accurately the direction of parsing for an iamb. These are just the sorts of circumstances where the language learner will have different options for parsing a subset of the data. If the crucial piece of data is not available, the learner can opt for a different analysis than the previous generation, leading to change. For instance, when faced with two stresses on a word with a sequence of [HLL], the learner could deduce an iambic foot where the direction of parsing could be from either direction. If the learner opts for left-to-right (6iv) even if the correct analysis was right-to-left (6iii), and continues to generalize the pattern on a [HLLL] word, she would incorrectly fail to assign a stressed foot on the final syllable.

The third parameter is more straightforward in the sense that it is limited to assigning main stress, either on the right or left word-edge. If we consider the parsing data with an iamb as in (6iv), we can see how the choice of edges will contribute to different main stress.

If a consonant, syllable or even foot is not included in the parsing, this invokes extrametricality. In Latin, it is essential to assume an extrametrical final syllable, since stress assignment entirely ignores the structure or weight of this syllable, as in (8). We also add the basic pattern for English nouns where again an extrametrical syllable is apparent.

The stress parameters not only provide the means of describing metrical patterns in languages, but allow us to determine expected types of changes. As we have seen, a single parameter is insufficient to provide the range of patterns that may exist in a language. The interaction of several parameters allows us to determine the actual patterns. Even then, only a subset of words point to the correct set of parameters for a given language. A pattern such as [σ̆σ̆́σ̆] is not enough to tell us whether the foot is a trochee or an iamb. Thus, a change in a single parameter will not affect all words, only a subset. The next section lays out the types of opacity of stress patterns, which may lead to a change in the metrical organization of a language. Since stress changes are closely related to weight, we examine possible modifications in syllable quantity and weight. We focus on two conflicting ways systems may or may not change: different metrical patterns, same output realization versus different output realizations but with similar metrical patterns.

Opacity can occur in stress systems in two ways; many words can have ambiguous metrical interpretations, or the same underlying pattern persists, but due to other interacting processes, surface variation arises. We discuss each in turn.

a typical example where numerous words maintain the same prominence pattern for generations, which is eventually reinterpreted as a different rule, occurred in Latin.⁷ Pre-Classical Latin is assumed to have had stress on the left edge, while Classical Latin as we have seen above, clearly prefers stress on the right edge. The opposite happened in Bengali, where Old Bengali preferred stress on the right edge, while Modern Bengali stress is reminiscent of Proto-Germanic with stress strictly assigned to the left edge. With the available parameters one might argue that these are clear instances of a change in where the main stress falls. If we consider the Latin examples in (8) and only change Main Stress Left to Main Stress Right keeping all other parameters constant, what would be the outcome?

Comparing (8) and (9), only in one example is the main stress different when the Main Stress parameter changed; main stress in incūdis shifted from the leftmost foot (9iv) to the right in Classical Latin, but the other words are stressed exactly as before since they only have one foot. This makes sense in a scenario with two feet and two stresses, possibly adjacent. The native speaker has difficulty in deciding which is the main prominence, particularly if stress clash occurs within phrases triggering variation. For example, main stress on thirtéen can shift to the first syllable in the phrase thírtèen mén due to a clash in stress. Such alternations may lead to a shift in the main-stress alignment, which could have happened in Latin.

The same surface patterns can be produced by two different underlying metrical systems. The surface evidence of the Main Stress parameter is restricted to certain types of words, namely those that have more than one foot, and when the change occurred, it must have resulted from ambiguity in words with two feet. In fact, the change is quite minor involving a decision as to which foot to give more prominence to when a word has more than one.

An alteration in any of these parameters may lead to a stress shift in only a few words and not others; Latin would be an example where many words maintain exactly the same surface forms although the underlying pattern has changed. Consequently, to deduce the stress parameters correctly, the language learner requires sets of unambiguous words. This is difficult since many words in our lexicons could be accounted for by various stress parameters. Consider the words in (10). A learner faced with these stress patterns could easily opt for any of the four sets of parameters given below.

All four options give the correct stress patterns in these words. The disambiguating example we need is a word like América, with the pattern [L'LLL]. The only option that covers the full range of data is (10iv). Thus, the possibilities of change are many. If three groups of learners came up with a separate option based on the first set of words, and if they resisted changing the pattern despite the newer data, or if they never come across the new data, then their dialects will differ from their parents, giving the same outputs with different underlying grammars.

a different problem arises when the same underlying system persists, but different surface realizations can occur. This occurs particularly when other word prosodic changes (such as changes in syllable quantity) intervene. Metrical structure governing stress is sensitive to the number and weight of syllables. The latter is sensitive to vowel and consonant length and the internal structure of syllables. All may change independently of stress, which in turn influences a change in foot structure.

More languages contrast vowel than consonant length, but the older Germanic languages show both vowel and consonant quantity contrast. For example, OE nominal stems were of the following types: CVC, CV̄ C, CVC_aC_a, CVC_aC_b.

Interestingly, the contexts in which vowels and consonants are lengthened or shortened appear to differ. The most frequent context for vowel lengthening or shortening is prosodic: closed syllable shortening, open syllable lengthening, and variations therein. In contrast, consonantal lengthening or gemination is often due to some process of assimilation.

A frequent type of gemination occurs when a consonant is followed by a sonorant, leading to complete regressive assimilation, as in the Latin data in (12). In West Germanic, the suffix or stem extension /j/ triggered gemination if the preceding syllable was light, as in the West Germanic data in (12).

Our interest is in the effect of gemination on foot structure. After gemination, the medial long consonant invariably closes the preceding syllable, and if a closed syllable counts as heavy, the foot structure changes among other effects. Compare the nouns with original H H stems in OE, which belong to two different stem classes, a-stem and ja-stem respectively; only wēsten is subject to gemination because it has a stem extension /j/.⁸ As a result of gemination, the second syllable in wēstenne becomes heavy, ending with a long consonant, with immediate consequences for foot structure (Dresher & Lahiri 1991).⁹

In OE, secondary stress is blocked on non-branching feet (13c,d). The critical result of gemination in (13iv) leads to an HHL sequence, while (13iii) maintains an HLL sequence. Demoting the final non-branching foot has equal effect on the nominative singular forms (13c,d), but the dative singular forms suddenly differ. The word wḗstènne (13b) keeps its secondary stress on the second syllable while (13a) carries only main stress. Consequently, gemination had a direct consequence on the surface stress pattern distinguishing two classes of nouns, which were exactly parallel before gemination. Thus, the underlying foot structure remains, but surface realizations have changed.

The other side of the coin is degemination, which converts a geminate into a singleton. An obvious context for degemination is when another consonant follows, as in OE: fyll-an ‘to fill-inf’, fyll-d-e > fylde ‘to fill-past-indicative.1p,3p’; cyss-an ‘to kiss-inf’, cyss-d-e > cyste ‘to kiss-past-indicative.1p,3p’ (Lahiri 2009). Degemination can, however, occur without any obvious syllabic context; most modern Germanic languages have lost original geminates: cyss is still written with two fricatives but pronounced as [kɪs]. Although degemination led to the loss of quantity, the metrical structure was not affected. A German example shows this: the word Küsse ‘kiss-pl’ now consists of two light syllables (LL) rather than an HL sequence when gemination was present, but main stress has remained on the initial for centuries.

Shortening and lengthening a stressed vowel can have similar consequences. Perhaps the best example of shortening and the effect on the overall stress pattern comes from Trisyllabic shortening (TSS). In (14), the OE words have the structures HHL and HLL, where the first word has two feet while the second has only one. TSS has the effect of shortening the first stressed vowel (Lahiri & Fikkert 1999).

Although OE words already tended to be short, they could have more than one foot and carry two stresses as in ṓþèrne. After shortenings such as TSS, native English words with two feet tended to be no longer than a single foot, like láverke and cícene. Again the metrical structure itself has not changed, but surface realizations have.

The opposite of TSS was open syllable lengthening (OSL) which applied to most West Germanic languages (Lahiri & Dresher 1999). However, interacting with TSS, OSL had no substantial effect on trisyllabic words: láverke > OSL lā́verke > TSS láverke. But disyllabic words did show a lasting effect as in the long initial vowel in words like naked, beadle, beacon, such that an original LH word would become HL after OSL and final consonant extrametricality. Again, lengthening did not have an immediate effect on stress, but moved words towards a much preferred trochaic foot.

Consequently, any change in consonant and vowel quantity directly influences syllable structure, which in turn may affect the number of feet in a word.¹⁰ Nevertheless, as in these instances where surface realizations of words have changed, the underlying metrical pattern can remain consistent despite the increase or decrease in the number of feet. Thus, many forces are at play within a given language without any outside interference, which may trigger changes only in a part of the system while the rest of the grammar remains constant. There are, however, outside influences as well which may play a role in initiating change and we turn to these next.

As we have seen, individual stress patterns of words do not always lead to an unambiguous answer as to which parameters are crucial for determining the metrical system of a language. The surface stress patterns of the words in (10) could be accounted for with different parameters, and only an addition of a fifth word with a particular sequence of syllables, allowed us to choose between the options. The surface opacity makes it quite clear that the data relevant for setting parameters from generation to generation is vulnerable to change. Yet rarely do native speakers change their prosodic patterns and we are very sensitive to non-native stress patterns.

There is, however, the additional issue of language contact.¹¹ A significant change in most Germanic languages is the shift of main stress to the right edge of a word from earlier left-edge prominence, and the general claim is that contact with Romance languages is to blame. How does this happen? If adult native speakers are so sensitive to non-native patterns, how do foreign patterns not only become absorbed, but actually trigger substantial changes? Recall that the Germanic shift is the reverse of the change from pre-Classical Latin to Latin where left-edge prominence moved to the right edge (cf. (9)). As noted, monopedal words remained ambiguous to the setting of the edge parameter; main stress on the rightmost or leftmost foot gave the same results. The shift must have occurred in words with two feet, and some of which could be adjacent.

Consider two scenarios that may lead to prosodic change—reanalysis of the system due to opacity vs influence from a different metrical system. Recall that metrical change involves a change in one or more parameters and occurs when native speakers encounter data which are ambiguous in terms of certain parameters (i.e. same surface pattern, different metrical systems). But speakers are conservative and pertinacious in maintaining underlying systems. Loans or internal prosodic changes in quantity can lead to changes in surface realizations, but without changing the system. Thus there is always a conflict between surface change, but same grammar versus surface same, but different grammar. To examine how these conflicting scenarios actually interact, we take English as a case study to trace how and why the shifts occurred, in section 15.4. We propose the sequence of events in (15) (see footnote 9 for the Resolved Moraic Trochee).

Three aspects of modern English stress are rather different from early English patterns and require explanation: (i) the shift of stress from word-initial position in some loans while maintaining the stress pattern of native words; (ii) absorbing stress-affecting affixes, and (iii) the introduction of noun-verb doublets alternating in stress. As we will see, Romance loans did not come in carrying the stress of the donor language. We will argue that the shift occurred in stages; neither French nor Latin patterns were absorbed without alteration.

While there is complete consensus that Proto-Germanic stress fell on the left edge, with some modifications, most synchronic accounts of modern Germanic (Icelandic being the exception) would agree that word prominence tends to be on the right edge. This represents a radical change in some words, but not others as shown in (16).

The diachronic scenario given above was described by Lahiri, Riad, & Jacobs (1999) as a change from the Resolved Moraic Trochee, Left-to-Right, Main Stress Left to the modern languages with Moraic Trochee, Right-to-Left, Main Stress Right. Scholars differ on details, but the overall sequence remains the same. In this section, we sketch the story for English, relying on Dresher & Lahiri (2005) and Fikkert, Dresher, & Lahiri (2006). Although undoubtedly Latin and Old French loans did have an effect on English stress, the change in the parameters did not happen all at once. Parameter changes went stepwise, the foot direction changing first, followed by main stress prominence shifting to the right.

The OE foot, or Germanic Foot, is an expanded (resolved) moraic trochee (cf. footnote 9) where the head must contain at least two moras and the dependent at most one. The two moras of the head need not come from the same syllable. As we have seen that not all surface forms provide evidence for metrical structure, examples are given in (17), varying in syllable structure and weight (Dresher & Lahiri 1991).

In (17e), the initial heavy syllable has two moras and occupies the head of the foot; the second syllable is light (one mora), and occupies the dependent branch. In (17d), the initial syllable is light, and so the second light syllable joins it (a process called resolution) to make up the head position of the foot. The third syllable occupies the dependent position. (17c) is similar, except resolution is with a heavy syllable. Two light syllables may form a foot, but it is non-branching as in (17a).

Evidence for the resolved foot also comes from syncope processes like High Vowel Deletion, which deletes high vowels in the weak branch of the foot. In the examples above, only the double underlined [u]‌ are deleted. Thus, since the word-final [u] in nī́tenu is not in the dependent branch of the foot, it survives, while the [u]s in hḗafudes and wordu are deleted, yielding hḗafdes and word. The final high vowel in the non-branching foot of lofu is not deleted.

What concerns us here are the stress patterns. In Old English, main stress falls on the leftmost foot. However, not all subsequent feet bear secondary stress; only branching feet can bear secondary stress. This is indicated by an underlined H-syllable, which marks non-branching non-initial feet. Thus, hḗafdes and ṓþer have only one stress, while ṓþèrne bears secondary stress. The lack of secondary stress on a word-final non-branching foot, is in effect, defooting and has the consequence of making the final closed syllable light. This is a significant change because, in effect, the final consonant is now being interpreted as extrametrical, as shown in (18).

We have seen before that another process which affected quantity, during the transition from Old to Middle English, is Trisyllabic Shortening (TSS), already applicable in OE. As in (14), the interaction of these rules led to a metrical shortening such that OE words that had more than one foot were reduced to a single foot in ME: OE lā́ vèrke, ME láverke.

The changes sketched above had no effect on the position of main stress, and the stress system in early ME remained essentially as in OE. However, TSS did metrically ‘shorten’ words and with the addition of final consonant extrametricality, many words that had more than one foot in OE were reduced to a single foot in ME (Lahiri & Dresher 1999: 709). Although OE words already tended to be short, they could have more than one foot and carry two stresses as noted already for ṓþèrne. Following defooting and further metrical shortenings, native English words tended to be no longer than a single foot, like láverke and cícenes.

OE suffixes were, as their descendants still are today, ‘stress neutral’, suggesting that they did not participate in the stress domain and main stress always fell on the first syllable, indicative of compound stress: frḗondscipe. Nothing until now suggests that there was data to prompt a shift in directionality or position of main stress. In the eleventh to twelfth century, the phonological system faced an onslaught of French borrowing after the Norman Conquest. French loans poured in, and one could imagine that the ground was fertile for a change in stress, as claimed in early generative accounts like Halle & Keyser (1971).¹²

According to Halle & Keyser (1971), in addition to a Germanic stress rule, a Romance stress rule was added to English around Chaucer’s period.¹³ The new rule incorporated two different patterns in Chaucer:

Their storyline is that this rule originally competed against the dominant OE stress rule, and was gradually extended over subsequent centuries. The relatively few early borrowings from Latin were secured by a much larger number of French words with the French stress pattern. This is possible under such an analysis, since the French and Latin stress rules are united. The later (c.1650) onslaught of Latin borrowings provided further evidence for a pattern that had already gained a foothold in English.

This appears to be flawed in several respects. First, under any analysis, Latin and French stress rules are dissimilar enough to yield quite different patterns. Second, the French pattern has no lasting effect on English prosody. Third, the Latin stress pattern, as distinct from the French pattern, is hardly attested in Chaucer. We consider each in turn and argue that Romance loans had little effect during Chaucer’s period.

The stress rules of Latin and French are indeed rather different and should yield different stress patterns.

One major difference is the treatment of final syllables. Latin does not permit word-final stress unless the word ends with a tense vowel while French prefers final stress unless there is a word-final schwa. Consequently, for the words in (23) French and Latin stress rules would give different patterns. The words in (23a) have to be marked as exceptions to the unified Romance stress rule.

Another inconsistency occurs in words with more than two syllables, where the final vowel is schwa and the penultimate syllable is light. Here, the French rule would give penultimate stress, while Latin has antepenultimate stress. Halle & Keyser claim that Old French words all had heavy penults, thus avoiding a conflict in words borrowed from that source, but many words borrowed into English from Latin had light penults and followed the Latin, not the French rule, undermining a unified approach to Old French and Latin stress patterns.

A second difficulty is that the French stress pattern has had no lasting effect on English prosody. Words with French stress in Chaucer could have been stressed according to the native English pattern, as required by the meter.

A word like service could receive word-initial or penultimate stress. The final e must have been pronounced, else the lines would lack the requisite number of syllables. In (24a), it must have three syllables and bear initial stress. No other stress assignment is possible in this word since an odd position can be occupied by a stressed syllable only when it is not adjacent to unstressed syllables on both sides. In contrast, in (b) stress must fall on the penultimate syllable. Hence we find doublets in Chaucer such as citée ~ cítee, comfórt ~ cómfort, divérs ~ díverse, Plató ~ Pláto, presént ~ présent, etc. With few exceptions, the present day English reflexes of Romance words with French stress in Chaucer have initial stress consistent with Germanic stressing, such as ábbot, bárren, cíty, cómfort, Égypt, fórtune, gíant, góvern, hónour, Pláto, sérvant, sólemn, témpest, tórment (noun), vírtue, sérvice, sécular, séction.

Furthermore, in many modern English words stress has been retracted from the final syllable as in Chaucer: ascéndant (cf. ascénd, Chaucer ascendént), cánon (Chaucer chanóun), and purvéyance (cf. purvéy, Chaucer purveyáunce; the more usual ME form was púrvey-). The few exceptions to the generalization are degrée, divíne, and rewárd with final stress (see below).

Thus, in general, Romance loans borrowed before the fifteenth century have initial stress in Modern English irrespective of the weight of the initial syllable (Svensson & Hering 2003), as shown in (25).

Today, many early and late trisyllabic loans bear initial stress. These words may once have been longer. Examples include báttery (c 1500), cápital (1290), ópera (1648), énemy (c 1340), crócodile (c 1300), érudite (c 1425), etc.¹⁴ In contrast, disyllabic Romance loans with final stress in English, which went entirely against any OE stress pattern, tend to have been borrowed much later: canal (1449), bourgeois (1564), gazelle (1582/1700), moustache (1585), gazette (1605), hotel (1644), champagne (1664), salon (1715), bouquet (1716), brochure (1765), beret (1850), taboo (1777), mirage (1800). There are, however, dialectal differences and some of these words are pronounced with initial stress in British English: bourgeois, salon, brochure, beret, mirage. Indeed, certain words which have final stress nowadays had initial stress earlier as cement (1300) where ME síment had initial stress until the nineteenth century, or antique which was identical in pronunciation to ántic until the eighteenth century. For the month July, final stress can be traced back to Johnson’s period.

A third problem is that the Latin stress pattern, as distinct from the Old French one, is hardly attested in Chaucer. If, as argued, the Latin stress pattern is distinct from that of Old French, evidence for the introduction of a Latin-type stress rule into English in Chaucer’s time must rest on words that exemplify this pattern. However, such words are rather rare in Chaucer, and tend to be Latin names. Halle & Keyser cite the following instances (p. 99).

We conclude that there is no clear evidence that either the Old French or the Latin stress rule gained a foothold in English at the time of Chaucer (see also Minkova 1997, Redford 2003, and compare Ratliff, this volume, on the care that needs to be taken when invoking contact as a cause for change in a prosodic system). Our hypothesis is that early loans were originally borrowed as morphologically simplex (we follow Lahiri & Fikkert 1999, also Minkova & Stockwell 1996). Thus, reverence was not initially derived from revere, nor austerity from austere. This accounts for the stress patterns of these words, and provides evidence that direction of parsing and placement of main stress had not changed before 1530. Further evidence comes from words that exhibit so-called ‘Medial Laxing’ where the stressed vowel in the verb is unstressed and short in the noun.¹⁵ In these medial-laxing pairs, the ‘derived’ forms have stress consistent with the OE pattern, firmly secure in the initial syllable, and moreover are often borrowed earlier than the base form.

If these pairs were not really morphologically related when they were borrowed, why do the ‘underived’ forms (typically verbs) bear stress on the second syllable? Recall that this pattern is consistent with the native model of unstressed prefixed verbs. The only non-initial stress in OE was found in prefixed verbs: forbḗodan ‘to forbid’, becúman ‘to become’, fullfýllan ‘to fulfill’. Furthermore, as Halle & Keyser point out, pairs like bī́gang_noun ~ begángen_verb ‘practice ~ to practice’ are very much part of the OE system (also see footnote 13). In all probability, loans such as revére could be treated as if they had unstressed prefixes.

Thus, most ‘derived’ words that came into the language during Chaucer’s period and earlier were borrowed as indivisible wholes. Only later was the derivational relationship established. For the fifteenth-century native speaker the underlying stressed vowels in reverence or president were short, and the first syllable of the ‘derived’ forms (typically nouns) in (27) were part of the stem and not treated as affixes. Thus, despite loans and variable surface realizations, the stress parameters in early ME remained largely the same.

Old English words tended to be short and suffixes were, as their descendants still are today, stress neutral. With metrical shortening, native words tended to be no longer than a single foot. Therefore, evidence for setting the parameters of directionality and main stress was in short supply. The loans in the preceding section show that it was not sufficient to borrow Latin words to provoke a change in directionality. What led to this change?

Following Danielsson (1948), Poldauf (1981), and others, we associate this change with the introduction of words with Latin suffixes such as -able/-ible,-ation,-ic(al),-ity,-ator, etc., each of which could be a foot on their own and behaved like original compounds where the second element would bear stress: brȳ́dgùma ‘bridegroom’, dǽ gsḕage ‘daisy’. However, the absorption of these words was still not straightforward as we can see from the sixteenth and seventeenth century literature. Peter Levins’s Manipulus Vocabulorum, printed in 1570, is perhaps most important for the study of change in stress patterns in English. It is a reverse (rhyming) dictionary, indicating the location of stress. Although Levins lists many words that appear to have stress on the right edge, a fairly large number do not conform to a straightforward Latin stress rule.

Levins’s remarkable list leaves little doubt that initial stress was very much present even in this period. To account for these patterns, Halle & Keyser evoke a stress retraction rule since their Romance stress rule would normally always provide stress on the right edge; thus, the Romance stress rule would predict canoníse, requiring stress retraction to give the main stress at the left edge, cánonise. As we have seen, little evidence suggests that the earlier loans had any effect on the stress parameters of English and no Romance rule was in evidence in the mid 1400s. A century later, when many more words had come into the language, the stress parameters seem to require some adjustment.

What appears to be happening is that native speakers of English continue their Germanic nominal/verbal distinction, where initial light syllables of verbs are treated as prefixes. We can deduce four notable facts from Levins’s dictionary. First, when disyllabic verbs are marked with final stress, the first syllable continues to function like a prefix: depúte, recórd, contráct, rebél, quarrél, rewárd. Four corresponding nouns have initial stress, suggesting that this analysis is on the right track: députe, rébel, quárrel, récord. Second, suffixes like -ity, -ible/-able, -ate were not following ‘normal’ Romance stress rules. Indeed, there are no words in -ory which do not have word-initial stress. Third, in comparison to verbs, very few disyllabic nouns bear final stress: flagón. Fourth, there has been a substantial increase in longer verbs ending in non-schwa vowels due to the suffixed loans: -ory, -ate, -ity.

Which parameters changed during Levins’s period? During the earlier period, morphologically complex words were always coming in as simplex forms, as is typical for borrowing crosslinguistically. This appears to have continued in the sixteenth century, during the early Shakespearean period. Consequently, suffixed words with -ible/-able, -ity were treated as monomorphemic. Nevertheless, unlike the earlier period where all words like admirable and residence bore initial stress, there are some exceptions and one begins to find surface alternations like húmidity versus antíquity. Without doubt there is still a very strong main stress parameter associated with the left edge. However, during the mid sixteenth century we begin to see three interacting reanalyses taking place.

First, and most important, we perceive the onset of morphologically governed stress alternating doublets in the synchronic grammar: nouns rébel, quárrel; verbs rebél, quarrél. This was a consequence of borrowed verbs being stressed on the root (parallel to forgíve), thereby strengthening the number of unstressed prefixes. The pattern is, however, not novel. Old English did have doublets, where the noun was derived from the verb, but with clear quantity distinction in the related prefixes: bī́ ~ be, ánd ~ on. prefixed disyllabic noun and verb pairs occurred, where the verb was derived from the noun: ándswaru ‘answer’ ~ ándswarian ‘to answer’, but these would always carry stress on the same syllable. What was new in Levins were doublets with stress differences as tórment_noun  ~ tormént_verb. Thus, the existing OE prefix-extrametricality for verbs increased.

The second factor is that although it is highly likely that as before most suffixed nouns came in underived, we do not systematically find word-initial stress as in Chaucer’s réverence. Words with initial stress like précipitate, bárbarity need to be contrasted with fratérnity, infírmity. A careful look at the data suggests that despite a preference for the left edge, the parameter that has changed is the direction of parsing, which now begins from the right edge.

In (32) we can directly compare the parameters around 1400 with those proposed for Levins. Main stress is aligned to the left of a word.

The two sequences which differ in main stress are (32a,d). The pattern in (32d) is largely found in verbs such as recognize, fraternize, or in affixed words beginning with light syllables which have a prefix-like structure such as agreeable, fraternal, etc. To establish the foot direction, however, the pattern in (32a) is critical. A further point is that additional support for the continuing existence of resolution, i.e. LX=H, comes from the lack of secondary stress on final feet in (32c,e), neither of which can bear stress since the feet are not branching.

Final superheavy syllables increase slightly, including ones with final long vowels. English ears would have been unfamiliar with these words and consequently the option was either to use straightforward Germanic stress (word-initial) or lexicalize them with final stress, which happened during Shakespeare’s time (flagón, turmóil, survéy), but they reverted to initial stress as the native system prevailed. A few instances of final long vowels are becoming acceptable.

It was not sufficient to borrow Latin words to provoke a change in main stress although the directionality parameter changed. The noun-verb alternations continued to stabilize, marking the prefix as extrametrical. As Cooper (1768: 113) notes, ‘some nouns by translating the Accent are changed into verbs’, continuing with pairs like réfuse ~ refúse, tórment ~ tormént.

Nevertheless, initial stress remains rather stable. According to Lowe’s Critical Spelling Book (1755), stress was prescribed on initial syllables in words like ácademy, cónfessor, délectable rather than acádemy, cónfessor, deléctable, which shows a strong mid-eighteenth century preference for the left edge. Danielsson (1948) attributes to Walker (1791) the observation that classical words were pronounced in English with alternating secondary stresses (e.g. Latin àcadémia). When ‘Englished’, the tonic and countertonic changed places to conform to English ‘speech habits’ (e.g. ácadèmy). The Countertonic Principle shows that the main stress parameter remained set to left for some time after the change of directionality to right. The addition of words stressed according to the Countertonic Principle would have increased evidence for main stress left. Thus, a word like ácadèmy clearly shows two feet, of which the left has main stress.

What was the next parameter change? Halle & Keyser point to the differences between Levins’s period and that of John Walker (1791), who also wrote a pronunciation dictionary:

Words like ádversary were stressed in the same way while others like refráctory had shifted to the antepenultimate syllable. Levins, as noted above, had only initial stress in words with -ory/-ary. Halle & Keyser argue that the words in the first column are derived by the early stress retraction rule, pulling the stress to the first syllable, but that those in column 3 are lexically stored.

We should consider some other examples before coming to any conclusion about changes in parameters. Consider the words in (34).

Halle & Keyser also drew attention to words like confíscate, stressed on the penultimate in Shakespeare (born six years before Levins’s book was published), but not in Levins. Walker, in contrast to Levins, has a much larger number of words where stress falls on the penultimate in trisyllabics. Furthermore, in four syllable words, we see consistently different patterns between Levins and Walker. This variation between three and four syllable words sets the stage for a reanalysis in the treatment of the final syllable. Unlike Levins, where the left edge was clear except for prefix-like words, it is not clear how to differentiate cónsistory (4σ) and confíscate (3σ) in terms of an unstressed prefix analysis. Nevertheless, main stress left remains strong, given that words like ádversary and cónsistory still bear stress on the left edge.

This was the onset of final syllable extrametricality. First, many of the words in (33) could also be accounted for by assuming the final syllable to be extrametrical, and hence would have had no surface change. Recall that secondary stress did not appear on the third foot of words like ádvèrsary and hence the final syllable played little role in stress assignment. Second, the deviations between Walker and Levins largely depend on the internal structure of disyllablic suffixes. The head of the foot on the suffixes -ory or -ary, bore no stress, and so the initial vowel of the suffixes -o or -a would very easily have been reduced to schwa. If so, syllable extrametricality was almost inevitable for the next generation. Compare the Levins-Walker words in (35).

Walker and Levins also differed as to the main stress in anniversary: Levins had main stress on the initial syllable while Walker stressed the antepenultimate syllable. Note that with the stress parameters that Walker had, he would have obtained two feet, and as he was a proponent of making the words ‘English’, he would have used his countertonic principle and obtained ánnivèrsary.

A further ambiguity now arises for the foot type which was still a resolved trochee during Levins’s period; otherwise réfractory and sátisfactory would not have borne main stress. During the end of the eighteenth century with the onset of extrametricality, the foot type also becomes opaque and a moraic trochee analysis would also work for Walker. Clearly, however, the main stress still appears to be on the left edge, as words like ádversary suggest. In such cases, the same words that provoked a change of directionality to right reinforced the evidence for main stress left.

The turning point must have been right after the onset of syllable extrametricality when the resolved trochee also became opaque. Probably as in the shift from the left edge to the right edge in Classical Latin, words with two feet were ambiguous for main stress. Words like ádvèrsary and sátisfàctory had two feet and therefore two stresses. In running speech, if stress clash occurs, main stress would shift to the right rather than the left edge. Second, some trisyllabic loans lost their final vowel and ended up with final stress, like bazaar from earlier bazaroo. Third, suffixes such as -ade, -ee, -eer, -esque, -ette, -oon, became more established, adding to sets of words with two feet. Though words like càscáde, cànnonér may have entered the language earlier, they may not have systematically had final stress until a later date. In (36) compare Walker’s pattern using a regular moraic trochee with main stress on the right edge where it is invariably the words with two feet that diverge.

Despite hundreds of years of loans, the urge to keep a foot at the leftmost edge remains as we see in many new loans such as Tìconderóga (Hayes 1981). The changes in the parameters were step-wise and it is off the mark to suggest that English moved abruptly from a ‘Germanic’ to a ‘Romance’ stress system.

Although changes in stress systems are not as well described or widely discussed as other aspects of phonology, we can still trace their history. Historical data for metrical organization include verse, as well as records of orthoepists, and for reconstruction we rely on comparative evidence. The lacunae within historical linguistics, where changes in metrical systems remain unreported, are probably related to the fact that stress changes are rarer than segmental modifications. Native speakers are very sensitive to ‘non-native accent’ and incorrect stress placement. Thus, any change in word stress from one generation to another must be subtle and discrete, since speakers are pertinacious in maintaining their grammar. The question remains: how do such systems change?

Word stress is not merely prominence on a single syllable or vowel; rather, it emerges as a combination of several parameters, such as foot type, quantity sensitivity, direction of foot parsing and word edge preference for main stress. Moreover, extrametricality determines whether any particular prosodic constituent at word edge plays an active role in setting the parameters. Change in any one parameter may lead to an alteration of stress patterns but it may affect only a small portion of words. The reverse also holds; the same surface stress pattern may be the result of different parameter settings. Consequently, any change in stress systems is rather complex and, indeed, stress patterns are surprisingly pertinacious, and universal metrical preferences and constraints govern possible and impossible prosodic shifts. Even under external influence, acquisition and learnability appear to be necessary factors.

Again, how does a parameter change? We drew attention to a characteristic of grammars called pertinacity. A rule or pattern may persist over time though its realization changes. An example is the persistence of a particular metrical pattern in a language, though it may apply to new forms and no longer apply to forms that it used to apply to. As a case study we argued that the Germanic Foot continued over a long period. This type of pertinacity can be summed up as same pattern, different output realization. There is a converse kind of pertinacity where output forms persist despite changes in the grammar. Such changes involve a reanalysis of the output form, provoked by changes elsewhere in the system. This type of pertinacity can be summed up as different pattern, same output realization. Since learners acquire their grammars guided by the output forms they are exposed to, we do not expect these forms—especially those that make up the ‘core’ or ‘primary’ data—to change in the course of acquisition. Reanalysis of grammar that does not involve an immediate change in output forms is thus a significant type of language change. Changes in stress systems are prime examples of both types of pertinacity.