Binomials in English and French: ablaut, rhyme and syllable structure

1.1 Cooper and Ross (1975) and subsequent experimental studies

The influential paper by Cooper and Ross (1975) posited constraints on constituent order in binomials that reflect considerations of onset sonority and complexity, vowel quality and length, and coda sonority and complexity, as shown in (1).

^[1]The psycholinguistic studies of Birdsong (1979), Green and Birdsong (2018), and Pinker and Birdsong (1979) offer insights into speakers’ sensitivities to these constraints. The experimental procedures in these studies involved asking participants to indicate the degree to which nonsense sequences exemplifying phonological constraints sounded right to them. Intuitions of this nature were elicited from native and non-native speakers of English and native and non-native speakers of French. For example, in Birdsong (1979) and Pinker and Birdsong (1979), native speakers, intermediate learners, and beginning learners of English and French were sensitive to ablaut-like alternations under the Vowel Quality constraint. All groups strongly favored nonsense sequences whose first element had a higher/more fronted vowel than the second. For example, the English item “The wet cereal was all gligy and glagy” was preferred to “The wet cereal was all glagy and gligy” and the French “J’en ai assez de ton rique-raque” was preferred to “J’en ai assez de ton raque-rique”. Using different elicitation methods, Green and Birdsong (2018) observed similar intuitions for English nonsense items (e.g., clizzip-clazzip/*clazzip-clizzip) and French nonsense items (e.g., calite-calate/*calate-calite) among native and high-proficiency non-native speakers of the two languages.

1.2 Reduplicatives in Optimality Theory

Under OT, phonological output forms are subject to a set of ranked constraints. Constraints are of two major types: faithfulness and markedness. An example of a faithfulness constraint is linearity, which prohibits the reordering of segments, i.e., metathesis (McCarthy and Prince 1995). An example of a markedness constraint is nonfinality, which states that a word-final syllable should not receive stress (Prince and Smolensky 1993).

Crucially, no output form satisfies all constraints. Rather, phonologies “optimize” by selecting forms that incur fewer or less serious violations of constraints. Constraints in classical OT are universal, while rankings (the relative importance of the constraints) are specific to individual languages.

Considerable work on reduplicatives has been conducted within the OT framework, which “remains the most popular umbrella theory for analyzing reduplication data in terms of their formal characteristics” (Schwaiger 2015: 471). For example, in the classical OT account of McCarthy and Prince (1994), the underlying representation of a reduplicative expression consists of a stem and a phonologically empty Affixred. On the output level, the correspondent of the stem is the base (i.e., the first constituent), and the correspondent of the Affixred is the reduplicant (i.e., the second term). Three kinds of correspondence relations are included in this model: Input–Output Faithfulness, Base-Reduplicant Identity (Output–Output matching) and Input-Reduplicant Faithfulness (arguably, the latter is less important). When Input–Output Faithfulness dominates a certain Markedness constraint (for example, nocoda), while this Markedness constraint in its turn dominates Base-Reduplicant Identity, an unmarked structure will emerge in the reduplicant. Under such constraint ranking, the reduplicant will be an “improved” or “optimized” version of the base. In OT accounts this process is referred to as the emergence of the unmarked (tetu). However, other OT accounts assume a modular approach to reduplication, where phonological constraints do not interact with the morphosyntactic structure of the reduplicative expression. Kentner (2017) analyzes German reduplication with an OT approach where phonology does not have access to the morphological status of the underlying morphemes. That is, phonological constraints are blind to what part of the output is the base and what part is the reduplicant since the phonology is assumed to operate independently from morphology.

Both Yip (1999) and Minkova (2002) suggest that the two constituents of reduplicative expressions have equal morphological status, and therefore, terms such as “base” and “reduplicant” may be unnecessary. Yip (1999) asserts that rather than the realization of a phonologically empty affix, the output of reduplication is the production of sequences that rhyme and alliterate. Thus, as shown in (2), Yip captures the phonological properties of reduplicatives (and, as noted above, conjoined sequences as well) in her proposed constraints rhyme and alliterate.

Note that the definitions in (2) can apply to adjacent feet or stressed syllables, to account for alliteration and rhyme in reduplicatives that are polysyllabic (e.g., shilly-shally, higgledy-piggledy).

It is apparent that alliterate, which stipulates the copying of the onsets of the constituents of reduplicatives, does not specify the nature of the stressed vowels which, as we have seen, typically reflect ablaut alternation for English and French.^[2] In Yip’s analysis, rhyme and alliterate interact with Markedness constraints. For example, for a hypothetical word hop the reduplicative expression will be hop-hop if both rhyme and alliterate dominate Markedness. Under a different hierarchy, for example, rhyme >> Markedness >> alliterate, a reduplicative expression like hop-lop would surface, since it complies with both high-ranked constraints: rhyme and Markedness (it has one “improved” unmarked onset).

While Yip’s analysis includes Input–Output Faithfulness, Minkova (2002) conducts her study exclusively on the Output level. Therefore, Base-Reduplicant correspondence is more important to her analysis, although she claims that these terms are only used for convenience and that in reality, both parts of reduplicative expressions are “equally ‘primitive’ partial copies of each other” (Minkova 2002:138). Similarly to Kentner, Minkova assumes the independence of morphological and phonological processes. Moreover, it is phonology that is deemed to underlie the reduplication template. This template in English is characterized by the alternation of a high front vowel in the first constituent and a low vowel in the second. Crucially, these two vowels are considered maximally different in vocalic quality.

On this alternation, Minkova (2002) proposes interest, defined in (3):

Note: BR indicates matching of the output Base to the Reduplicative. Minkova’s constraint favors maximal perceptual distance between the two vocalic nuclei in terms of quality (specifically, height) and phonetic duration. The difference in quality should be maximal. Thus, candidates like /rif-ræf/ and /rɪf-ræf/ (the latter being more typical of English) are potential winners given the great height difference between the nuclei in the first and second terms; on this logic, candidates like /rɪf-rʌf/ and /rɪf-rɛf/ would be marginally allowed.

The difference in duration should likewise be maximal, but non-categorical (i.e., the phonological length of the vowels should remain the same). Under the constraint final length, a phonetically long vowel is predicted to occupy the second slot.^[3] (A superordinate constraint, interest, includes final length and *ident-br(High).)

For binomials like hip-hop and flip-flop, Minkova notes that the tenseness status of the second vowel in such sequences is ambiguous. Nominally classified as tense, /ɑ/ is underlyingly bimoraic; thus under ident-br (μ) /ɑ/ would be prohibited in ablaut reduplication. Minkova acknowledges that vowels /ɔ, ɑ/ “push the threshold of acceptability with respect to the faithfulness constraint ident-br (μ) to the limit” (158), but claims that at the time when the majority of ablaut words like flip-flop entered the English lexicon, the quality of the second vowel was different from its modern counterpart; specifically, it was lower and more open.

Summarizing, in the analyses of Yip (1999), Minkova (2002) and Kentner (2017), we observe that all three emphasize phonology as the driver of reduplicative word formation. As mentioned above, we orient our hypotheses for speakers’ preferences for nonce sequences around specific elements of Minkova’s and Yip’s analyses. The findings of our experimentation constitute an empirical basis for potential refinements in OT-theoretical approaches to reduplicative structure.

1.3 Reduplication in corpus studies

Reduplication has been investigated with corpus analyses such as Arleo (2009), Cabrera (2017), Kentner (2017), Thun (1963), and Wang (2005). The volume by Thun (1963) on English reduplicatives, cited in Minkova’s analysis, contains a large corpus of reduplicative expressions along with an analysis of the sound patterns found in this corpus. Thun divides his corpus into three subgroups: (i) Copy reduplication or reduplication with no change (e.g., tick-tick) (ii) Rhyme reduplication or reduplication with the change of onset (e.g., hurly-burly) (iii) Ablaut reduplication or reduplication with the change of stressed vowel (e.g., shilly-shally). Thun points out that rhyming reduplicatives constitute “the most important group in so far as their number amounts to half the total number of reduplicatives” (Thun 1963: 212). This observed frequency suggests that, at least for English, the OT constraint rhyme may be highly ranked.

In his analysis of the ablaut group, Thun points out the prevalence of the /ɪ – æ/ alternation, with 278 reduplicatives out of approximately 490 instantiating this pattern (conjoined binomials were not included in these counts). Other vocalic combinations of his corpus include: /ɪ – ɔ/ (97), /ɪ – ʌ/ (10) and /i – æ/ (10).

Arleo (2009) analyzes reduplication in a corpus of counting-out games from 51 languages. (Counting-out sequences, for example, English tinker-tailor and inky-pinky-ponky and French bernic-bernac and pif-paf-pouf, are used to designate someone to be “it” in children’s games or in other selection contexts.) Ablaut reduplication was found in counting-out sequences in 30 of the 51 languages forming the corpus. Arleo (2009: 307) discerns a dominant ablaut pattern of “strong or maximal” contrast between two successive vocalic nuclei, the first of which tends to be a high front vowel, and the next likely being either a front low, back low or back mid vowel. Although other types of contrasts are found in this corpus, the Arleo counting-out numbers are largely consistent with Minkova’s idea of maximal perceptual difference (in terms of height) between the two vowels, with one constituent in the sequence containing some variant of /i/ and the next constituent having some variant of /a/; see also 4.4.^[4]

2.1 Objectives

As suggested above, there is a demonstrable need for evidence in English and French on current preferences for the main sound templates in binomials, Rhyme and Ablaut. This need is underscored by Minkova (2002: 166), who suggests that researchers should undertake “a comparison of Ablaut and Rhyme reduplication in English and an investigation of similar models elsewhere.” The resulting knowledge may be useful for assessing prospects for the relative productivity and survival of these templates. To this end, we oriented the objectives of our study around the three oppositions mentioned earlier.

Opposition 1: Simple Rhyme versus Ablaut. One objective is to determine whether speakers of English and French prefer Simple Rhyme templates or Ablaut templates in binomial sequences. Both rhyme and ablaut are frequent patterns in French and English binomials. Rhyme is highly productive, as suggested by its ongoing creative exploitation by poets and rappers, and by the near-limitless extent to which it can be applied in expressive language generally. Ablaut alternation, by comparison to rhyme, is a less widespread expressive mechanism, with fewer possible contexts lending themselves to its application. In the context of reduplicatives, the application of ablaut is most prevalent during the sixteenth and seventeenth centuries (Minkova 2002: 139). Only a handful of ablaut reduplicatives, e.g., hip-hop, ping-pong, ticky-tacky, emerged in the English lexicon during the twentieth century. Also, Minkova notes that the English reduplicative lexicon counts approximately 1800 items, of which only 20 % to 30 % exhibit ablaut patterns (139). Our approximate count for rhyme (based on Thun’s corpus which is also used by Minkova) is 59 % of the corpus; ablaut patterns constitute 27 % and the rest is copy reduplication. With these factors in mind, one may anticipate that simple rhyming patterns in both English and French will be more attractive than ablaut patterns. Examples: English sith and tith versus sith and sath; French ni piche ni siche versus ni piche ni pache.

Opposition 2: Simple Rhyme versus Complex Rhyme. As a permutation of Opposition 1, here we oppose Simple Rhyme with Complex Rhyme. Recall that Minkova’s analysis suggests a constraint DEP-BR that would eliminate candidates like [rɪf-brɪf], which exhibit consonant epenthesis resulting in a complex onset. By other OT analyses, *complex would likewise disfavor reduplicative candidates with consonant clusters, which are considered marked cross-linguistically. (We note, however, that examples of rhyming templates with complex onsets are attested in both English, e.g., by hook or by crook, helter-skelter, and French, e.g., am-stram-gram; la rogne et la grogne.) Due to the marked status of complex onsets, one could nevertheless hypothesize that simple rhyming would be preferred in both languages (e.g., English siz and piz vs. siz and stiz; French ni tive ni sive vs. ni tive ni trive).

Opposition 3: Complex Rhyme versus Ablaut. This opposition completes the possible permutations of comparisons. In light of the previous argumentation and evidence, it is reasonable to conjecture that the presence of a complex onset in rhyming would diminish the likeability of the rhyming template. We are interested in knowing the extent of any such diminishment, and whether similar effects would obtain for French and English. For both French and English, we might predict that ablaut would be preferred over complex rhyme, the latter being marked. At the same time, the number of permissible consonant clusters is arguably greater in English than in French,^[5] thereby raising the possibility that speakers of the two languages might differ in the degree to which ablaut is preferred or, conceivably, a preference for complex rhyme over ablaut would be exhibited by speakers of English. Examples: English fev and fov versus fev and flev; French ni tive ni tave versus ni tive ni trive.

We investigate these three oppositions with an innovative psycholinguistic method for pairing nonsense sequences. Previous work (Birdsong 1979; Green and Birdsong 2018; Pinker and Birdsong 1979) looked at preferences for constituent order in pairings of nonce sequences exemplifying Cooper and Ross (1975)-type constraints. Thus, for example, in Green and Birdsong (2018), participants were asked to indicate their degree of preference for lisket-lasket versus lasket-lisket, the first nonce sequence conforming to the Cooper and Ross Vowel Quality constraint, the second sequence violating the constraint. The results were interpreted in terms of speakers’ sensitivities to each of the individual constraints. In the present study, we do not examine speaker preferences for constituent order; rather, our objective is to assess the relative strengths of Ablaut, Simple Rhyme and Complex Rhyme by pitting nonsense exemplars of these templates against one another in a pairwise fashion. Thus, for example, participants are asked to indicate their preference for Ablaut versus Simple Rhyme by rating the Ablaut sequence fiply-faply in opposition to the Simple Rhyme sequence fiply-biply. Similar pairings of nonsense sequences are developed to elicit preferences for Simple Rhyme versus Complex Rhyme and for Ablaut versus Complex Rhyme. Details follow.

In Experiment 1, for English and French, we apply these techniques to the elicitation and analysis of preferences for disyllabic nonce binomials that contain disyllabic terms, e.g., fiply-faply versus fiply biply. In Experiment 2, these techniques are applied to items with monosyllabic constituents, e.g., siff and saff versus siff and piff.

3.1 Participants

Fifty native speakers of American English (31 f, mean age 21 years) and 51 native speakers of French (37 f, mean age 26.5 years) participated in Experiment 1. All the native speakers of French resided in France or Belgium. The English speakers were recruited primarily among undergraduate and graduate students from two American universities. Most of the French speakers were tested in Paris and were university students or working professionals; some (N = 16) were tested at the same two research sites as our English subjects. Almost all the participants in both groups were college educated and, as a consequence, had had some exposure to a second language. Candidate participants who considered themselves balanced French–English bilinguals were excluded. In the US and France, the testing conditions were identical: a quiet room, the same researcher, and instructions and interaction with the researcher (including solicitations of participation and email correspondence with participants) were exclusively in the language of testing, thereby minimizing the possibility of priming any other language.

3.2 Stimuli

As with earlier psycholinguistic work, the stimuli in Experiment 1 were specifically created nonsensical expressions, thereby enabling participants to focus solely on the form of the stimuli, without being distracted by meaning, word frequency or other factors. To the extent possible, our items are distinct from existing words in English or French. The stimuli in our experiments had been vetted by native speakers of French and English. They affirmed that our final set of stimuli were not real words and that they did not evoke strong associations with existing words.^[6] Half of the items represent the conjoined condition (e.g., feblic and teblic), and the other half the reduplicative condition (e.g., fiply-faply); both conditions are typical of lexicalized binomials. Participants were tested on 30 pairs of experimental items, with 10 items instantiating each of the three oppositions identified above. All items are minimal pairs with disyllabic elements.

For Opposition 1, we developed English-like stimuli such as fiply-biply (Simple Rhyme) and fiply-faply (Ablaut) and French-like stimuli such as fudette-budette (Simple Rhyme) and fudette-fudotte (Ablaut). Examples of Opposition 2 for English: settip and gettip (Simple Rhyme) and settip and slettip (Complex Rhyme); for French: ni saglette ni paglette (Simple Rhyme) and ni saglette ni staglette (Complex Rhyme). Examples of Opposition 3 for English: gesky and gosky (Ablaut) and gesky and glesky (Complex Rhyme); for French: ni cougui ni cougua (Ablaut) and ni cougui ni crougui (Complex Rhyme).

The phonological features of the stimuli were controlled appropriately. For Simple Rhyme pairs, the onset in the first term is always a fricative and the second onset is always a stop (e.g., English: sipret-dipret; French: sagli-dagli). For Complex Rhyme pairs, the onset of the first term is either a fricative or a stop, and the second onset combines that same consonant with either an /r/, an /l/ or a /t/ (e.g., English: fipret-flipret; French: fabi-flabi). For Ablaut items, half are contrastive in vocalic height (e.g., English: fiply-faply; French: dabi-daba) and the other half contrast in vocalic advancement (e.g., gesky and gosky; French: cudette-cudotte). Although binomial contrasts in vocalic advancement are not as common as contrasts in vocalic height, they are nevertheless present in English and French binomials (e.g., English: hem and haw, by guess and by gosh; French: bique et bouc; pin-pon); see Oakeshott-Taylor (1984), who analyzes vocalic advancement as a factor in determining directionality in English binomials.

The terms of the English items are all stressed on the first syllable, reflecting the most common metrical pattern (trochaic) in binomials with disyllabic elements (e.g., hurly-burly). The French items are all stressed on the second syllable, consistent with the fact that polysyllabic words in French receive stress on their final syllable.^[7] See Appendix 1 for items in Experiment 1.

To provide a context resembling ordinary speech, test items were inserted at the ends of carrier sentences. For English, for example, each nonsensical item appeared at the end of the sentence “He told me about…”. Male native speakers of French and English read aloud and recorded the sentences at a natural pace. To create the audio stimuli for the experimental task, individual target expressions were cropped from their carrier sentences. A trained phonetician listened to a sample of the recorded items and confirmed that they were natural-sounding and appropriate for the study.

3.3 Procedure

The study was approved by institutional review, and participants were provided with a consent form for participation in research before the experiment. The experiment was conducted in a quiet room. After a brief practice phase, the experimenter left the room and each participant completed the task individually. After the experiment, the participants were compensated and properly debriefed.

As shown in Figure 1, below, for each experimental item participants looked at two nonce binomial expressions on a laptop screen, while simultaneously listening through headsets to a recording of these expressions.^[8] Beneath the expressions, a rating scale from 1 to 6 was displayed. Participants were instructed to indicate which of the expressions sounded more acoustically pleasing to them. If they preferred the nonce binomial that appeared on the right side of the screen, they clicked on either 6 (strong preference), 5 (moderate preference), or 4 (mild preference); if they preferred the expression appearing on the left side of the screen, they clicked on either 1 (strong preference), 2 (moderate preference), or 3 (mild preference).^[9]

Figure 1:

Laptop display of a sample item and accompanying rating scale, with the movable arrow indicating a hypothetical participant response.

For the audio presentation of each item, the two expressions were separated by a 700 ms interval.^[10] The testing was not paced, and response times were not measured. Participants were not allowed to go back and review or change previously-answered items. The experiment was conducted and the responses were recorded with PsychoPy Software Package (Pierce 2007).

Potential ordering effects were minimized by counterbalancing within each Opposition. For example, for Opposition 1 half of the items represented Ablaut nonce binomials first and Simple Rhyme binomials second, and half were Simple Rhyme first and Ablaut second; a corresponding counterbalancing procedure was applied to binomial expressions in the other Oppositions. In addition, the order of presentation of the 30 items was randomized for each participant.

For each item, participants’ preferences were converted to values of −3, −2, −1, 1, 2, or 3. If participants aligned their preferences with the predicted pattern for a particular item, that item received a value (score) of either 1, 2 or 3, with 3 signifying the strongest positive valence. When participants selected the unpredicted pattern for a particular item, the values for that item were tagged with a negative sign, with −3 signifying the strongest negative valence. As an example, for the item in Figure 1, a value of 1 (signifying mild preference) was assigned, since our hypothesis was that Simple Rhyme (fiply-biply) would be preferred over Ablaut (fiply-faply) (see Jacobs et al. (2015) for an example of this same technique of assigning values to opposing preferences). Under this procedure, mean values at 0 indicate indifference.

Recall that our hypotheses for English and French were not exactly the same. Specifically, we anticipated that for both English and French, Simple Rhyme would be preferred to Ablaut on Opposition 1; and Simple Rhyme would be preferred to Complex Rhyme on Opposition 2. However, for Opposition 3, our hypotheses differed for French and English. For French, we predicted that Ablaut would be preferred over Complex Rhyme, but for English our prediction was that the Complex Rhyme would be preferred over Ablaut (see 2.1). Consequently, we adjusted the coding for English and French responses accordingly.

Keeping these hypotheses and our research objectives in mind, we conducted statistical analyses of the collected data, which allowed us to determine the following:

1. Are the means of each of the three Oppositions significantly different from the level of indifference (0 in our case)? If so, this would be considered evidence of subjects’ preference for a particular template within this Opposition.

2. Is there an effect for Condition (reduplicative vs. conjoined)? Reduplicative and conjoined binomials differ in terms of their origins (Malkiel 1959), but are similar in terms of the sound templates that they exploit (Birdsong 1979; Cooper and Ross 1975; Thun 1963). In the present study, we were able to test for a possible differential Condition effect on rhyming versus ablaut items; see the examples in (1). We note that, under a different experimental paradigm with nonce binomials that involved ordering preferences (as opposed to Oppositions in the present study), Green and Birdsong (2018) found no effect for Condition.

3.4 Experiment 1: results for native English speakers

Figure 2 below shows native English participants’ preferences for nonce binomials with disyllabic constituent terms. Simple Rhyme is preferred over Ablaut on Opposition 1 (M = 0.6, SD = 2.1). Simple Rhyme is also preferred over Complex Rhyme on Opposition 2 (M = 0.9, SD = 2.0). In contrast, participants’ sensitivities in Opposition 3 are at indifference, i.e., they exhibit no preference for Ablaut or Complex Rhyme (M = −0.03, SD = 2.1).

Figure 2:

Native English speakers’ preferences for nonce binomials with disyllabic terms, by Opposition. Opposition 1: Simple Rhyme versus Ablaut; Opposition 2: Simple Rhyme versus Complex Rhyme; Opposition 3: Complex Rhyme versus Ablaut. Black dots and horizontal lines indicate preference means and standard errors; the dotted red line represents the level of indifference.

In the first analysis, we compared the means of the three oppositions to the level of indifference, defined as 0. We used a mixed effects model with a suppressed intercept, a fixed effect of opposition and two random effects, subject and item, each with a random intercept. (The suppressed intercept was needed to gauge the difference between the means and 0.) This analysis revealed that the means for two oppositions were statistically different from 0: Opposition 1 (t(33.5) = 2.7, p = 0.01) and Opposition 2 (t(33.5) = 3.9, p < 0.001). For Opposition 3, however, no significant difference from the level of indifference was observed (t(33.5) = −0.2, p = 0.88).

Recall that for Opposition 1 we anticipated that English speakers would prefer Simple Rhyme, and that in Opposition 2 Simple Rhyme would be preferred to Complex Rhyme. The analysis corroborated these two hypotheses. For Opposition 3 we predicted Complex Rhyme > Ablaut; this prediction was not confirmed.

In the second part of the analysis, our objective was to assess the effects of Opposition and Condition, as well as to check for a possible Condition × Opposition interaction. Using lmer and emeans packages in R, we applied a mixed effects model with a random intercept for fixed effects (Opposition × Condition) and a random intercept for each of the random effects (subject and item). The analysis revealed a main effect for Opposition (F(2,24) = 3.7, p = 0.04, partial eta squared = 0.24). A post-hoc Tukey test revealed a statistically significant difference between Opposition 2 and Opposition 3 (p = 0.03), but no other significant differences between oppositions. No main effect of Condition and no interaction of Opposition and Condition were found.

3.5 Experiment 1: results for native French speakers

French native participants’ preferences for nonce binomials with disyllabic terms are shown in Figure 3 below. A visual examination reveals general patterns, where Ablaut is preferred to Simple Rhyme on Opposition 1(M = −0.5, SD = 2.2) and to Complex Rhyme on Opposition 3 (M = 1.2, SD = 1.9). For Opposition 2, Simple Rhyme is preferred to Complex Rhyme, as hypothesized (M = 1, SD = 2). Note that the negative-valence mean for Opposition 1 indicates that our French subjects’ preferences were opposite to our initial hypothesis, as they preferred Ablaut to Simple Rhyme.

Figure 3:

Native French speakers’ preferences for nonce binomials with disyllabic terms, by Opposition. Opposition 1: Simple Rhyme versus Ablaut; Opposition 2: Simple Rhyme versus Complex Rhyme; Opposition 3: Complex Rhyme versus Ablaut. Black dots and horizontal lines indicate preference means and standard errors; dotted red line represents level of indifference.

For French speakers, the first analysis revealed that the means for all three oppositions were significantly different from the level of indifference defined as 0: Opposition 1 (t(29.7) = −2.5, p = 0.02), Opposition 2 (t(29.7) = 5.04, p < 0.001) and Opposition 3 (t(29.7) = 6.5, p < 0.001).

We also looked for main effects of Opposition and Condition, and for an interaction of Opposition and Condition. The results were similar to English: a main effect of Opposition was found (F(2,24) = 4.63, p = 0.02, partial eta squared = 0.28), but no main effect of Condition was found, and no significant Condition × Opposition interaction was found. A post-hoc Tukey test revealed a significant difference between Opposition 1 and Opposition 3 (p = 0.02).

3.6 Discussion of Experiment 1

For nonce binomials with disyllabic constituent terms, the most prominent finding among English speakers is their general preference for Simple Rhyme, which is favored over both Ablaut and Complex Rhyme templates. Their preference for Simple Rhyme over Ablaut is in line with the high frequency of rhyming among English binomials (Thun 1963; see above) as well as with the frequency of rhyme across genres and varieties of expressive language; see above. English speakers’ preference for Simple Rhyme is also consistent with the paucity of ablaut alternations in recently lexicalized English reduplicatives (Minkova 2002). As we anticipated, Simple Rhyme was likewise preferred over Complex Rhyme, possibly due to the marked status of consonant clusters, and in keeping with the high ranking of *COMPLEX and DEP-BR constraints in some OT accounts (Section 2). Note that the English participants’ dispreference for Complex Rhyme items is observed relative to Simple Rhyme (Opposition 2). In the case of Opposition 3, however, there is no statistical difference in preferences between Ablaut and Complex Rhyme.

As might be predicted from the relative infrequency of consonant clusters in French (see Note 5), native speakers of that language preferred Simple Rhyme items over Complex Rhyme items. Both Simple Rhyme and Complex Rhyme templates exhibit weaker preference effects than the Ablaut template among French speakers. This result is opposite to that observed for English natives.

To summarize, for nonce binomials with disyllabic constituents, English speakers tend to prefer simple rhyming patterns over ablaut alternations and favor simple onsets over complex onsets. In contrast, for French speakers, ablaut alternations are preferred to rhymes with both simple and complex onsets. The only case where French results parallel those for English is in the comparison of Simple Rhyme with Complex Rhyme; here, speakers of both languages disfavor items with complex onsets.

Thus, for both French and English one can find evidence for *COMPLEX and DEP-BR in speakers’ dispreferences for complex onsets relative to simple onsets in binomials with disyllabic constituents. For English, the RHYME constraint of Yip (1999) is supported by results for Simple Rhyme items. No such support for RHYME is provided by the preferences of French speakers in this experiment. The relative strength of ablaut over both types of rhyme in French is not directly predictable from previous research, and is perhaps surprising given the frequency of rhyme across numerous genres, varieties and contexts of expressive language (e.g., rap, poetry, drama, song, nursery rhymes, slang, derogation, salutation), just as in English and other languages, e.g., German, Arabic, Portuguese, Polish, Spanish.

As noted in (2) and (3) above, Yip’s alliterate constraint licenses vocalic alternations. We pointed out that these are typically ablaut alternations, which are consistent with interest: br (Minkova 2002). Based on the results of Experiment 1, we are in a position to tentatively infer for binomials with disyllabic elements a relative strength hierarchy, with alliterate (interest: br) ranked over rhyme for French and rhyme over alliterate (interest: br) for English.

4.1 Methods

Experiment 2 is a partial replication of Experiment 1, using different stimuli and with several design considerations in mind. First, to explore further the differences between French and English speakers’ preferences, all nonce binomial items in Experiment 2 are composed of monosyllabic terms. In Experiment 1 French speakers consistently preferred Ablaut over both types of Rhyme, whereas English speakers found Simple Rhyme preferable to Ablaut. A possible explanation for this result may relate to metrical differences between French and English, which come into play with the disyllabic items in Experiment 1. Recall that for the English items, stress was always assigned to the first syllable of each term, and that this syllable contained the targeted feature (rhyming or ablaut alternation). For the French items, it was not always possible to place the target segments within stressed syllables, since in French the tonic accent falls on the final syllable of polysyllabic words. Thus, in rhyming items such as s aubette- d aubette the stressed syllable does not contain the contrastive segments /s/ and /d/, while in saub e tte-saub o tte the stressed syllable does contain the targeted segments /ɛ/ and /ɔ/. To address the design confound of positional variation of the target feature that results from differences in the metrical structure of French versus English, we were motivated to examine English and French speakers’ preferences using nonce binomials composed of monosyllabic elements.

An additional motivation for testing with monosyllabic-constituent items is to confirm or disconfirm the findings in Experiment 1. For example, we would like to know whether English speakers’ preferences for Simple Rhyme over Ablaut apply to binomials with disyllabic terms as well as to binomials with monosyllabic terms, or whether their preferences are conditioned by syllable structure. Similarly, for French speakers, we would like to know if, in nonce binomials composed of monosyllabic terms, participants are inclined to favor Ablaut over Simple Rhyme and Complex Rhyme and to prefer Simple Rhyme over Complex Rhyme, as they are with binomials with disyllabic constituents.

As a second departure from Experiment 1, in Experiment 2 the nonsense items representing Ablaut, Simple Rhyme, and Complex Rhyme are repeated in their respective Oppositions. For example, for English participants siff and saff (Ablaut) appear in Opposition 1 and Opposition 3, siff and piff (Simple Rhyme) appear in Oppositions 1 and 2, and siff and sliff (Complex Rhyme) in Oppositions 2 and 3. Thus for each Opposition, the resulting juxtapositions in English are exemplified as follows: Opposition 1 (Ablaut vs. Simple Rhyme): siff and saff versus siff and piff; Opposition 2 (Simple Rhyme vs. Complex Rhyme): siff and piff versus siff and sliff; Opposition 3 (Ablaut vs. Complex Rhyme): siff and saff versus siff and sliff. For French, examples of juxtapositions under Opposition 1 (Ablaut vs. Simple Rhyme) include: ni piche ni pache versus ni piche ni siche; Opposition 2 (Simple Rhyme vs. Complex Rhyme): ni piche ni siche versus ni piche ni priche; Opposition 3 (Ablaut vs. Complex Rhyme): ni piche ni pache vs. ni piche ni priche. This element of consistency over Oppositions allows for more direct comparisons of participants’ respective preferences for Ablaut, Simple Rhyme and Complex Rhyme between Oppositions, and reinforces the reliability of the results.

Likewise in the interest of consistency, all items were constructed in the conjoined condition. We justified this decision based on the results in Experiment 1, which revealed that binomial form (i.e., reduplicative vs. conjoined) was not a significant factor for English or French in any of the oppositions tested (see 3.4 and 3.5).

Finally, recalling the findings of Green and Birdsong (2018), where French participants preferred nonsense sequences whose first term contained a relatively more obstruent initial consonant than the second, in Experiment 2 the French nonsense items exemplifying Simple Rhyme are sequenced accordingly; thus, ni t ide ni f ide, ni p effe ni s effe, etc. (See Appendix 2 for items in Experiment 2).

Experiment 2 was approved by Institutional Review, and participants provided consent prior to testing. The procedure was identical to that of Experiment 1. English participants were recruited from among the students of a major American university, while French participants, similarly to Experiment 1, were recruited mostly in Paris, with some (N = 15) recruited in the U.S., among the exchange students at an American university. Fifty native English speakers (26 f, mean age 20.5 years) and fifty native French speakers (31 f, mean age 25 years) participated in Experiment 2. All French speakers were born in Europe and considered themselves native speakers of French; all English speakers were born in the U.S. and considered themselves native speakers of American English. Participants’ backgrounds in terms of education and exposure to a second language were similar to Experiment 1 and self-assessed balanced bilinguals were excluded.

In Experiment 2 the testing procedures, verification of the naturalness of the stimuli, and counterbalancing and randomization measures were all identical to those employed in Experiment 1.

In light of the results of Experiment 1, we revisited the research hypotheses for the Experiment 2. Recall that in Experiment 1 we had anticipated that Simple Rhyme would be preferred to Ablaut in Opposition 1 by both French speakers and English speakers, but French speakers in fact preferred Ablaut. Therefore, we anticipate that in Experiment 2 French participants should prefer Ablaut over Simple Rhyme. For English speakers in Experiment 1, on Opposition 3 Ablaut was preferred over Complex Rhyme, contrary to our prediction. However, we decided to retain our original hypothesis for Experiment 2 because the observed difference was not statistically significant.

4.2 Experiment 2: results for native English speakers

English participants’ preferences for nonce binomials with monosyllabic constituent terms are shown in Figure 4 below. Visual examination reveals the prevalence of Ablaut when opposed to both Simple Rhyme (M = −0.3, SD = 2.1) and Complex Rhyme (M = −0.9, SD = 1.9). In addition, Simple Rhyme wins out over Complex Rhyme in Opposition 2 (M = 0.7, SD = 1.9).

Figure 4:

Native English speakers’ preferences for nonce binomials with monosyllabic terms, by Opposition. Opposition 1: Simple Rhyme versus Ablaut; Opposition 2: Simple Rhyme versus Complex Rhyme; Opposition 3: Complex Rhyme versus Ablaut. Black dots and horizontal lines indicate preference means and standard errors; dotted red line represents level of indifference.

For native speakers of English, the first mixed-effect analysis for Experiment 2 comparing opposition means to the level of indifference yielded results for monosyllabic terms that were inconsistent with those for disyllabic terms in Experiment 1. Unlike Opposition 1 in Experiment 1, the direction of English preference in Experiment 2 is Ablaut > Simple Rhyme, with the difference approaching significance (t(34.6) = −1.9, p = 0.07). For Opposition 2, the results are consistent with Experiment 1, as Simple Rhyme is preferred over Complex Rhyme, with the difference being significant (t(34.6) = 4.6, p < 0.001). As for Opposition 3, recall that in Experiment 1 there was no difference in English natives’ preference between Ablaut and Complex Rhyme; in the case of Experiment 2, Ablaut is significantly preferred over Complex Rhyme (t(34.6) = −5.6, p < 0.001).

Since Experiment 2 does not have Condition as one of the predictors (i.e., the items do not include reduplicatives, only conjoined terms), we used a model with only one fixed effect, Opposition (with a random intercept), and two random effects, Subject and Item, each with a random intercept. The overall effect of Opposition was significant (F(2, 26.9) = 4.2, p = 0.03, partial eta squared = 0.24). A post-hoc Tukey test revealed a significant difference between Opposition 1 and Opposition 3 (p = 0.02).

To summarize, for native speakers of English, our replication of Experiment 1, using items with monosyllabic constituents and repeated exemplars of phonetic templates across oppositions, yields divergent results. In Experiment 2 Ablaut is significantly preferred over Complex Rhyme (Opposition 3), whereas there was no statistical difference in Experiment 1. For Opposition 1, English participants demonstrated a non-significant preference (p = 0.07) for Ablaut over Simple Rhyme in Experiment 2, whereas in Experiment 1 there was a significant preference for Simple Rhyme over Ablaut. For Opposition 2, the preference in Experiment 2 (Simple Rhyme > Complex Rhyme) is consistent with that of Experiment 1 and is statistically significant in both experiments.

4.3 Experiment 2: results for native French speakers

For the native speakers of French in Experiment 2, preferences for nonce binomials with monosyllabic constituents are displayed below in Figure 5. We discern here the general directions of preferences: Ablaut over Simple Rhyme for Opposition 1 (M = 0.5, SD = 2.1); Ablaut over Complex Rhyme for Opposition 3 (M = 0.6, SD = 2.1); Simple Rhyme over Complex Rhyme for Opposition 2 (M = 0.3, SD = 2.1). These preference directions align broadly with those of French speakers in Experiment 1, which involved binomials with disyllabic elements.

Figure 5:

Native French speakers’ preferences for nonce binomials with monosyllabic terms, by Opposition. Opposition 1: Simple Rhyme versus Ablaut; Opposition 2: Simple Rhyme versus Complex Rhyme; Opposition 3: Complex Rhyme versus Ablaut. Black dots and horizontal lines indicate preference means and standard errors; dotted red line represents level of indifference.

Comparing opposition means to the level of indifference with the same suppressed intercept model as in the previous experiments, we find that Opposition 1 (Ablaut > Simple Rhyme) is significant (t(51.1) = 2.99, p < 0.01), as is Opposition 3 (Ablaut > Complex Rhyme) (t(51.1) = 3.7, p < 0.001). As for Opposition 2, the observed difference is in the predicted direction (Simple Rhyme > Complex Rhyme), but falls short of statistical significance (t(51.1) = 1.7, p = 0.1). Finally, the second analysis did not reveal a main effect of Opposition.

4.4 Discussion of Experiment 2, with reference to Experiment 1

The findings from Experiments 1 and 2 are summarized in Table 1.

For French speakers, the directions of preferences in Experiment 2 align with those of Experiment 1. Specifically, for nonce binomials with both disyllabic and monosyllabic constituents, Ablaut is preferred over Simple Rhyme and Complex Rhyme, and Simple Rhyme is preferred to Complex Rhyme. However, this last preference falls short of statistical significance (p = 0.1).

For English participants, the preference in Experiment 1 for Simple Rhyme over Complex Rhyme (Opposition 2) is maintained in Experiment 2. For Opposition 1, the direction of preference is reversed, with Simple Rhyme preferred over Ablaut in Experiment 1 and Ablaut over Simple Rhyme (approaching significance) in Experiment 2. For Opposition 3, the results of Experiment 1 (Ablaut = Complex Rhyme) and Experiment 2 (Ablaut > Complex Rhyme) are dissimilar as well.

We note that in Experiment 2 the English participants and French participants display similar directional preferences for each of the three Oppositions: Ablaut > Simple Rhyme (Opposition 1); Simple Rhyme > Complex Rhyme (Opposition 2); Ablaut > Complex Rhyme (Opposition 3). Thus for nonce binomials with monosyllabic constituents, but not for nonce binomials with disyllabic terms, the preferences of native speakers of English generally resemble those of native speakers of French.

In reflecting on these results, we first recall the design features of the two experiments. For French, the nonsense items in Experiment 2 were different in two ways from those of Experiment 1. First, all items in Experiment 2 contained monosyllabic terms, with the effect that stress now coincides with the targeted contrast. Second, for Simple Rhyme items in Experiment 2, the directionality of the constituent terms in the items was altered to reflect speakers’ preferences for more obstruent initial consonants in the first term. Crucially, these changes in Experiment 2 did not result in meaningful departures from the findings of Experiment 1 for French participants.

For English on the other hand, the use of monosyllabic-term items in Experiment 2 turns out to be consequential. Not only is SR > CR the sole preference that is maintained from Experiment 1 to Experiment 2, but now, unlike Experiment 1, the preferences for English native speakers effectively align with those of French natives: ABL > SR; SR > CR; ABL > CR. These results lead us to suggest that syllable number, with its corresponding metrical structure, is a key factor in English speakers’ intuitions for phonetic form in binomials.

Our English monosyllabic items ended in a strong syllable (e.g., liss and lass), while the English disyllabic items ended in a weak syllable (e.g., kiply-kaply; kettip and kottip). The acoustic correlates of stress in English are known to be intensity, duration and pitch (e.g., Delattre 1963; Fry 1955; Lehiste 1970). Although researchers differ in their assessments of the relative prominence of these features, stressed syllables in English reliably differ from unstressed syllables by their overall greater length as well as by the longer duration of the vocalic nuclei (e.g., Crystal and House 1988; Lehiste 1970). Syllable lengthening is also an acoustic correlate of phrase structure, with final syllables (both stressed and unstressed) in a variety of phrasal domains being measurably longer than corresponding phrase-internal syllables, e.g., Klatt (1975). We note that with monosyllabic ablaut sequences like liss and lass or shiz and shaz (v. spick and span, zig and zag) the second monosyllabic term contains an inherently longer (and lower) vowel /æ/ than the vowel /ɪ/ in first term. This longer vowel is further lengthened by its occurrence phrase finally and in a stressed syllable. Data from Crystal and House (1988) show for /ɪ/ a mean duration of 75 ms and 53 ms in stressed and unstressed syllables, respectively; for /æ/ the corresponding mean durations are 159 ms and 71 ms. Similar length differences for these vowels are attested by Toivonen et al. (2014) and Umeda (1975). These three factors – stress, phrase-finality and inherent vowel duration – conspire to create a distinctly longer second syllable in Experiment 2 English items.

In this context we note that, of the phonological determinants of word order in binomials, researchers have identified short-before-long as the most general ordering principle; see Birdsong 1979 for an overview and discussion. The short-before-long generalization applies not only to ablaut alternations, but also to syllable number (fewer syllables in the first term, more in the second term), onset complexity (fewer initial consonants in the first term relative to the second), and coda complexity (fewer final consonants in the first term).

We revisit our previous discussion of the relevance of our psycholinguistic findings to Optimality Theory (see Section 3.6). With evidence from Experiment 1, we were able at that time to cautiously infer a relative strength hierarchy of alliterate (interest: br) over rhyme for French disyllabic items and rhyme over alliterate (interest: br) for English disyllabic items. Based on participants’ preferences in Experiment 2, we can now tentatively suggest a strength hierarchy of alliterate (interest: br) over rhyme for monosyllabic-term binomials in both French and English.

As in Experiment 1, in Experiment 2 we find evidence to support *complex and dep-br in both English and French speakers’ dispreferences for complex onsets relative to simple onsets. This result is consistent with the fact that complex onsets are cross-linguistically marked.^[11]

These considerations, along with speaker preferences for the oppositions we tested, allow us to propose the following OT hierarchies for reduplicatives in English and French.

We suggest that a contribution of this tentative proposal is its incorporation of syllable number in constituent terms as a phonological feature, thereby allowing OT theory to meaningfully distinguish reduplicatives across languages, in this case, English and French.