3.1 Local preservation

In English, the phonological patterns of complex words sometimes differ from those of simple words and this difference can be attributed to the preservation of the phonological properties of a morphosyntactically related word. This can be called ‘paradigmatic dependency’, which Bermúdez-Otero (Reference Bermúdez-Otero2016a) defines as in (1).

1. (1) Paradigmatic dependency in morphophonology (Bermúdez-Otero Reference Bermúdez-Otero2016a: section 7)

2. The form of a linguistic expression a is predictable from the surface representation of one or more morphosyntactically related expressions {b, c, . . .}.

Cases where a and b stand in a relationship of containment, i.e. b is contained within a, have traditionally been analysed using cyclicity which, according to Scheer (Reference Scheer2011: 85), has been one of the defining properties of generative phonology from its beginnings to more recent theories such as Phase Theory.Footnote ⁶ The cycle can be defined as follows: ‘the computation of the phonological properties of the parts precedes and feeds the computation of the phonological properties of the whole’ (Bermúdez-Otero Reference Bermúdez-Otero and Trommer2012).

Cyclicity was introduced to account for the difference in pairs such as còndensátion ~ còmpensátion, in which the former can have [e] in its second syllable but not the latter (Chomsky & Halle Reference Chomsky and Halle1968: 39). This is attributed to the difference between the bases of these words: condénse has an accent (and therefore a full vowel which can be transmitted to its derivative) on its second syllable, but cómpensate does not. The phonological form of the derivatives is therefore assumed to be dependent on the phonological form of their bases and, in cyclic phonology, the phonological computation of the base is assumed to precede that of the derivative. This particular configuration (cyclic preservation in inter-tonic position) has been argued to provide unconvincing evidence for the cycle because underived words can have an unreduced vowel in that position (e.g. òst[e]ntátion) and certain derivatives have a systematically reduced second vowel even though the corresponding vowel is accented in the base (e.g. ìnf[ə]mátion, despite inf[:]m) (Halle & Kenstowicz Reference Halle and Kenstowicz1991).Footnote ⁷

Let us now consider the case of English derivatives with three pretonic syllables. Monomorphemic words normally have an initial accent,Footnote ⁸ whereas derivatives preserve the position of the accent found in their base: àbracadábra, èlecampáne Footnote ⁹ vs orìginálity (cf. oríginal), famìliárity (cf. famíliar) (Hammond Reference Hammond1989; Kiparsky Reference Kiparsky1979; Halle & Kenstowicz Reference Halle and Kenstowicz1991; Collie Reference Collie2007, Reference Collie2008). Derivatives preserving the accent found on the second syllable of their bases are evidence that the phonological shape of the derivatives depends on that of their bases because the former cannot be predicted by the grammar of simple words (which predicts an initial accent) but the latter can. Because the base is contained within the derivative, this can be analysed through cyclicity as in (2).

1. (2)

   

A single cycle would incorrectly generate *òriginálity and *fàmiliárity (cf. àbracadábra). The contrast between abracadabra and originality can be formalised as in (3) using Optimality Theory (henceforth OT).

1. (3) Accent placement in OT

2. Ident-Accent: An accented syllable in the input should be accented in the output.

3. Accent-Left:Footnote ¹⁰ The leftmost syllable of the word should be accented.

   1. (a)

      

   2. (b)

      

The ranking Ident-Accent >> Accent-L predicts that if an accent is present in the input, the preservation of that accent will systematically override the assignment of secondary accent to the leftmost syllable. Because originality has the accent of original in its input,Footnote ¹¹ it gets a second-syllable accent (3a) whereas abracadabra has no accent in its input and receives an initial accent (3b).

The accentual contour considered in this article, /021(–)/, can be analysed in a comparable manner. In English, monomorphemic words or words containing a bound root with two pretonic syllables all have an initial secondary accent (e.g. àlabáster, guàrantée, màthemátics, sòlidárity).Footnote ¹² Derivatives with a base accented on its second syllable normally do not preserve that accent and also have an initial accent (e.g. aróma → àromátic, gazétte → gàzettéer, specífic → spècifícity). If we add a new constraint to represent the general restriction against adjacent accents which outranks Ident-Accent, *Clash, this can be analysed as in (4).

1. (4) Non-preservation in the second syllable in derivatives with two pretonic syllables

2. *Clash: Avoid adjacent accents.

   • 

In this configuration, *Clash prevents the preservation of the accent on the second syllable of the base and rules out candidate a. Then, candidate b is ruled out because it violates Accent-L, which leaves us the correct form, candidate c.

Exceptions to that analysis have been reported in the literature (Kager Reference Kager1989: 171; Collie Reference Collie2007: 79; Hammond Reference Hammond1999: 329; Pater Reference Pater2000: section 2.4). Consider the examples in (5), which are all taken from Collie (Reference Collie2007: 79) who collected them from Jones (Reference Jones2003).

1. (5)

   

The examples in (5) are exceptional because they violate *Clash, apparently favouring accent preservation to clash avoidance. All the authors who mention cases such as these (see references above) also point out that this contour is only found in derivatives, which in turn strongly suggests that cyclic preservation is the force overriding *Clash. Overall, this is supported by a search in Wells (Reference Wells2008), as the only monomorphemic words or words containing a bound root found are those listed in (6).Footnote ¹³

1. (6)

   

Even though the first three words in (6) are not derived through suffixation, the presence of an accent on their second syllable can be attributed to a morphosyntactically related form: the neoclassical root electro-,Footnote ¹⁴ refráct and relúctant (blended with cònductívity), respectively. Only the last two words in (6) appear to be free from morphological influences,Footnote ¹⁵ but the /021(-)/ pronunciation for these words can only be found in American English. Therefore, in British English, the /021(-)/ contour does seem to be only found in derivatives.Footnote ¹⁶ We can also note the instability of the accentual patterns in (6) which is an additional clue to the exceptional character of the /021(-)/ contour. This variability can also be found in most of the words in (5) and, as will be seen below, in most of the words which can be accented /021(–)/. In the rest of this article, we will refer to the occurrence of this contour as exceptional accent preservation (EAP).

Kager (Reference Kager1989: 171) argues that stress preservation may occur in dissyllabic pretonic sequences if the second syllable of that sequence is heavy, as in the case of the words in (5). This raises two questions: is a heavy second syllable a requirement to allow EAP? And, if that parameter is not sufficient to account for EAP, what other parameters can account for the occurrence of that contour?

In the next two sections, two parameters which may affect accent preservation will be presented: the existence of a more deeply embedded word and the relative frequency of the base and its derivative.

3.2 Preservation from a remote base

When a word is formed through successive affixations (e.g. person → personify → personification), it is generally the immediately embedded constituent (henceforth the ‘local base’) which can transmit its properties to that word, rather than more deeply embedded constituents (henceforth ‘remote bases’). For example, accent is generally inherited from the local base, but not from the remote base, as shown by the examples in (7), which are taken from Guierre (Reference Guierre1979 : 323). The terminology is borrowed from Stanton & Steriade (2014) but is used in a slightly different way for remote bases: the authors adopt an approach in which paradigmatic dependencies may hold between forms which do not stand in a relationship of containment. They analyse morphosyntactically and semantically related forms which are more frequent than a derivative as its remote bases (e.g. atomicity has four remote bases: atom, atomician, atomize, atomization). However, in this article, only forms which are contained within the local base are treated as remote bases.

1. (7)

   

However, there are reported cases of what Collie (Reference Collie2007: 288) calls ‘leap-frogging’ preservation, i.e. cases in which phonological properties appear to be transmitted directly from the remote base to the derivative. For example, Bermúdez-Otero (Reference Bermúdez-Otero2007) reports the paradigm in (8c) in the speech of a former colleague at the University of Manchester:

1. (8)

   

Both (8a) and (8b) correspond to what classic cyclic approaches predict. In (8a), the diphthong of cycle is transmitted to cyclic and then on to cyclicity.Footnote ¹⁷ In (8b), the vowel undergoes shortening in cyclic and that vowel is then transmitted to cyclicity. However, in (8c), cyclicity appears to inherit its vowel from the remote base cycle rather than from the local base cyclic. Additionally, the phonology does not predict a diphthong in this position, which shows it is indeed the preservation of the diphthong of cycle.

However, convincing evidence for leap-frogging preservation is hard to come by. Collie (Reference Collie2007: 289) lists potential examples such as tótal → totálity → totàlitárian ~ tòtalitárian. It could be argued that the second variant of totalitarian preserves the initial accent in total, and especially so if we take into consideration the fact that total is more frequent than totality. But it is very difficult to demonstrate that the initial accent in tòtalitárian comes from total because preservation failure of an accent on the second syllable of the local base results in an initial accent even in derivatives which do not have a remote base (e.g. antícipate → antìcipátion ~ ànticipátion). Here preservation failure (and therefore the default initial accent) cannot be distinguished from preservation from the remote base.

There are two ways one could demonstrate an influence of the remote base. The first would be to show that derivatives with a remote base behave differently from derivatives without a remote base. The second is as follows: as accent preservation failure never results in the accentuation of the second syllable of the derivative if a word only has one base, accent preservation from a remote base could be proposed in a configuration in which the remote base is accented on its second syllable and the local base is accented on its first syllable (but not on the second). In that configuration, if a derivative does not preserve the initial accent of its local base and has an accent on its second syllable, it could be argued to be evidence for leap-frogging preservation. Such cases are listed in (9):Footnote ¹⁸

1. (9) acádemy → àcadémic → acàdemícian ~ àcademícian

2. aróma → àromátic → àromatícity ~ aròmatícity

As pointed out by Ricardo Bermúdez-Otero (personal communication), the derivatives in these examples both have an onsetless first syllable and it has been argued that this could favour the accentuation of the second syllable rather than that of the first syllable (Collie Reference Collie2007: 103; Halle & Kenstowicz Reference Halle and Kenstowicz1991).Footnote ¹⁹ Therefore, the examples in (9) do not constitute incontrovertible examples of leap-frogging preservation any more than cases such as totalitarian.

To sum up, let us formulate the conditions in which one can claim that the phonological shape of a derivative can be said to be influenced by that of its remote base. The derivative must have a phonological characteristic which is:

• not predicted by the grammar of monomorphemic words or words containing a bound root;

• found in the remote base and

  • ○ absent from the local base. In that case only is there evidence for leap-frogging preservation.

  • or

  • ○ present in the local base, but there should be a significant difference between derivatives with remote bases and those with only a local base. This is not evidence for leap-frogging preservation but simply evidence for an influence of the remote base on the derivative.

Moreover, Collie (Reference Collie2007: 289) argues that a remote base is more likely to transmit some of its properties to its derivative if that remote base is more frequent than the local base. Therefore, frequencies have to be taken into consideration in the study of the interaction between bases and their derivatives.

Some of the examples of words which can be accented /021(–)/ cited in (5) have remote bases which have primary accent on the same syllable as the local base (e.g. connéct → connéctive → cònnectívity ~ connèctívity). Therefore, the role of remote bases will have to be evaluated in this study of EAP.

Finally, it is worth noting that traditional approaches to cyclicity cannot account for the type of leapfrogging phenomena discussed in this section. As Collie (Reference Collie2007: 288) points out, ‘while strict locality is assumed in cyclic analyses, it does not automatically follow from theories of lexical access’. Under fake cyclicity (see section 3.4), if the remote base is more frequent than the local base, it may influence the phonological shape of the derivative more than the local base does.

3.3 Relative frequency and Hay's dual-route race model of lexical access

Previous studies have shown that stress preservation (Collie Reference Collie2007, Reference Collie2008) and vowel preservation can be described with reference to the relative frequency of a base and its derivative (Hammond Reference Hammond2003; Kraska-Szlenk Reference Krazka-Szlenk2007), i.e. that these preservation phenomena are more likely to occur if the base is more frequent than its derivative. This can be exemplified by the examples in (10), which are taken from Bermúdez-Otero (Reference Bermúdez-Otero and Trommer2012: section 3.3.3), after Kraska-Szlenk (Reference Krazka-Szlenk2007: section 8.1.2).

1. (10)

   

Collie (Reference Collie2007, Reference Collie2008) claimsthat this supports Hay's (Reference Hay2001, Reference Hay2003) proposal on relative frequency according to which lexical access in complex words can be achieved through two routes: a direct route and a decomposed route. Hay argues that the more frequent a word is, the higher its resting activation level, and the easier and faster that word can be accessed in long-term memory. Therefore, if a base is more frequent than its derivative, the decomposed route should be the fastest, which means that the base is more likely to be perceived inside the derivative. In that case, the base is more likely to transmit its properties to its derivative. Conversely, if a derivative is more frequent than its base, the direct route should be the fastest, and we could expect preservation to fail. This dual-route model of lexical access can be represented as in figure 1.

Figure 1. Schematized dual-route model from Hay (Reference Hay2001). The solid line represents the decomposed route and the dashed line represents the direct route. Resting activation levels are represented by the thickness of the circles (BNC frequencies: sane (289), insane (360)).

In Hay (Reference Hay2001), the proposed parsing line (i.e. the line above which items are more likely to be accessed through the decomposed route) is the arbitrary x = y line. Hay & Baayen (Reference Hay, Baayen, Booij and van Marle2002) refine this proposal with an empirically motivatedFootnote ²⁰ parsing line above which words should mainly be accessed through the decomposed route and below which words should predominantly be accessed via the direct route. The line represents the relative frequencies for which both routes are equally likely. They note that if this line is above x = y, it may be because the direct route is likely to have an advantage due to the added effort of retrieving the different parts of a word. Both parsing lines are represented in figure 2.

Figure 2. The x = y line is represented by the dashed line. Hay & Baayen's (Reference Hay, Baayen, Booij and van Marle2002) parsing line is represented by the solid line.

Collie's (Reference Collie2007, Reference Collie2008) work on relative prominence preservation in -ion derivatives finds that relative frequency is a significant predictor of preservation failure: if a derivative is more frequent than its base, it is more likely to fail to preserve the position of the accent found in its base, and so more likely to be receive an initial accent, like monomorphemic words or words containing a bound root (e.g. antícipate → antìcipátion ~ ànticipátion).Footnote ²¹ Within the framework of Stratal OT (Bermúdez-Otero Reference Bermúdez-Otero and Trommer2012, Reference Bermúdez-Otero2016b; Bermúdez-Otero & McMahon Reference Bermúdez-Otero, McMahon and Trommer2006), Collie uses the concept of ‘fake cyclicity’ to capture the data.

3.4 Fake cyclicity

Like Lexical Phonology and Morphology (LPM; Kiparsky Reference Kiparsky, Van der Hulst and Smith1982, Reference Kiparsky1985; Mohanan Reference Mohanan1982; Kaisse & Shaw Reference Kaisse and Shaw1985), Stratal OT assumes the hypothesis that phonological computation is achieved through the application of three distinct phonological grammars: the stem-level phonology, the word-level phonology and the phrase-level phonology. In classical LPM, the highest stratum, the stem-level, is internally cyclic. This means that a complex form can undergo several passes through the stem-level phonology at every concatenation of an affix defining a stem-level domain. This means that the computation of a word such as Elìzabéthan requires two stem-level cycles: one for Elízabeth and one for Elìzabéthan.

In Stratal OT, all strata are non-cyclic. The effects of the stratum-internal cycle of classical LPM are captured by positing that the outputs of the stem-level phonology are stored non-analytically, i.e. they are stored in a morphologically unanalysed form.Footnote ²² Therefore, the computation of a complex form like Elìzabéthan does not require two online cycles. Elízabeth is stored in long-term memory with the output of the stem-level phonology, including its antepenultimate accent. In the computation of Elìzabéthan, this accent is present in the input as Elízabeth is retrieved from the lexicon. A faithfulness constraint (such as Ident-Accent) then ensures that the accent on the second syllable of Elízabeth is preserved in Elìzabéthan. The computation of the accentual contour of Elìzabéthan is therefore comparable to that of orìginálity shown in (3a). Once performed, the output of that computation is stored as well and the computation becomes a ‘lexical redundancy rule’ (Jackendoff Reference Jackendoff1975).

Crucially, the retrieval of the base in the lexicon can fail, as predicted by Hay's model of lexical access. If it fails, then the computation of the complex form has no accent in the input to preserve and it is therefore performed independently of the accentual contour of the base. For example, miscegenation is more frequent than its base, miscegenate and, as a consequence, is more likely to be accessed through the direct route. Therefore, the accent on the second syllable of miscégenate will not be present in its input. If so, the computation of miscegenation will be as in (11).

1. (11)

   

In that configuration, there is no accent to preserve and so the word receives the ‘default’ initial accent, just like monomorphemic words or words containing a bound root (e.g. àbracadábra, èlecampáne, ròdomontáde). To sum up, in that analysis, the failure of accent preservation in a derivative is attributed to a direct lexical access, which is caused by the high frequency of that derivative relative to the frequency of its base.

3.5 Interim summary: what could explain the /021(–)/ contour?

In the preceding sections, several parameters which could potentially determine the occurrence of EAP have been mentioned. Let us briefly summarise these parameters here.

The first parameter is syllable structure. As Kager (Reference Kager1989: 171) reports the /021(–)/ contour only for derivatives with a heavy second syllable and a light first syllable, we could expect the weight of the first two syllables to be a determining factor. However, this could be an effect of absolute weight (e.g. the weight of the second syllable) or of relative weight (e.g. the weight of the second syllable relative to that of the first syllable). Both will have to be tested.

Besides, consonants and vowels have been shown to affect stress in different ways. Let us consider two examples. Firstly, in English, final long vowels have been claimed to attract final stress regardless of the category of the word,Footnote ²³ whereas final consonant clusters only attract final stress in verbs (Chomsky & Halle Reference Chomsky and Halle1968; Hammond Reference Hammond1999; Hayes Reference Hayes1980). Secondly, consider the examples in (12), which are taken from Burzio (Reference Burzio1994: 54–5).

1. (12)

   1. (a) assíst → assístant

   2. (b) rev[íə]re → rév[ə]rent

In (12a), stress is maintained on the second syllable, whereas in (12b), it moves back one syllable, although both final syllables are heavy in the bases, which predicts that both derivatives should be stressed identically. Therefore, vowels and consonants will be treated separately in the analysis.

In order to evaluate the relative weight of the first two syllables, the mora counts in Hammond (Reference Hammond1999: 145) which are listed in (13) were used. They are based on distributional regularities in English and on the assumption that syllables should contain at least two morae (except if they contain schwa) and three morae at the most.

1. (13)

   

Footnote ²⁴Footnote ²⁵

The second parameter is word frequency. It has been shown that a high frequency of the base relative to that of its derivative can be expected to favour preservation. However, since Fidelholtz (Reference Fidelholtz1975), high-frequency words have been shown to be more likely to undergo lenition (see Myers & Li (Reference Myers and Li2009) for a review). As a consequence, absolute frequency has to be controlled for. Finally, we saw that a high-frequency remote base could be expected to favour preservation. Therefore, the study of frequency will have to take into consideration the frequencies of both local and remote bases.

The last parameter which might be expected to influence accent preservation is suffix-specific idiosyncrasies. It is possible that certain suffixes reject accent clash more than others. Some suffixes may also be morphologically more decomposable than others (Hay Reference Hay2003; Hay & Baayen Reference Hay and Baayen2003), the morphological decomposability of a given suffix being linked to the frequency of derivatives containing that suffix relative to that of their bases. Ideally, each suffix should be studied individually. If the numbers per suffix are too low for a separate analysis, differences will still have to be evaluated and (if possible) accounted for.