Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2013 Jan 1.
Published in final edited form as: J Phon. 2011 Nov 21;40(1):82–91. doi: 10.1016/j.wocn.2011.08.003

The production and phonetic representation of fake geminates in English

Grace E Oh 1,*, Melissa A Redford 1
PMCID: PMC3352589  NIHMSID: NIHMS324857  PMID: 22611293

Abstract

The current study focused on the production of non-contrastive geminates across different boundary types in English to investigate the hypothesis that word-internal heteromorphemic geminates may differ from those that arise across a word boundary. In this study, word-internal geminates arising from affixation, and described as either assimilated or concatenated, were matched to heteromorphemic geminates arising from sequences of identical consonants that spanned a word boundary and to word-internal singletons. Word-internal geminates were found to be longer than matched singletons in absolute and relative terms. By contrast, heteromorphemic geminates that occurred at word boundaries were only longer than matched singletons in absolute terms. In addition, heteromorphemic geminates in two word phrases were typically “pulled apart” in careful speech; that is, speakers marked the boundaries between free morphemes with pitch changes and pauses. Morpheme boundaries in words with bound affixes were very rarely highlighted in this way. These results are taken to indicate that most word-internal heteromorphemic geminates are represented as a single long consonant in the speech plan rather than as a consonant sequence. Only those geminates that arise in two word phrases exhibit phonetic characteristics that are fully consistent with the representation of two identical consonants crossing a morpheme boundary.

1. Introduction

True geminates are phonetically long segments that contrast with phonetically short segments in a phonemic inventory. Fake geminates are phonetically long segments that are not contrastive. The latter segments arise when morpheme concatenation results in a sequence of two identical segments. The current study focuses on these accidentally long segments in English to investigate the phonetic representation of segmental duration in a language that does not have a length contrast.

1.1. True versus fake geminates

There are three types of geminates in the world’s languages: lexical, assimilated, and concatenated. Lexical geminates are given in the lexicon and are part of the phonemic inventory (e.g. [papːa] ‘mush’ vs. [papa] ‘pope’ in Italian). Assimilated geminates arise when one segment takes on the identity of the preceding or following segment at a word-internal morpheme boundary (e.g. cheon+li→[ ʨ hʌlːi] ‘the law of nature’ in Korean). Concatenated geminates arise from identical consonant sequences that span a morpheme boundary within a word or in a phrase (e.g. un+named, fun name). Gemination can also arise through morphological derivation (e.g., Arabic).

True geminates are distinguished from fake geminates in that, even when spanning a morpheme boundary, segmental length in one word distinguishes it from an otherwise identical word in the inventory. For example, the Korean word [ʨhʌlːi] ‘the law of nature’ is distinguished from [ʨhʌli] ‘handling’ by segmental length rather than by morphology because the archaic Chinese morphemes from which [ʨhʌlːi] and [ʨhʌli] were derived historically are not accessed by native Korean speakers during comprehension (see Cho, 2001). This example from Korean can be contrasted with the situation in English where segmental length never serves to distinguish meaning in otherwise phonologically identical words.

Autosegmental phonology captures the difference between true and fake geminates as a difference in the number of underlying feature bundles involved. True geminates are represented as a single feature bundle that is linked to two timing units, whereas fake geminates are represented by a sequence of feature bundles with each bundle linked to its own unit (see McCarthy, 1986). The representations of true and fake geminates are illustrated in (a) and (b), respectively.

graphic file with name nihms324857u1.jpg

An autosegmental phonological representation of geminates suggests that differences between true and fake geminates will not be manifested in absolute timing differences at the phonetic level, since both are associated with two timing units (though see Payne (2005), for an alternative view). Such a suggestion is supported by phonetic studies that have compared the absolute duration values of true and fake geminate consonants in Estonian (Lehiste, Morton, & Tatham, 1973), Levantine Arabic (Miller, 1987), Bengali (Lahiri & Hankamer, 1988), and Tashlhiyt Berber (Ridouane, 2010) and found no difference between the two geminate types.

Although true and fake geminates may be indistinguishable from one another in terms of absolute consonant duration, a few studies indicate that they may be distinguished by relative duration (i.e., singleton-to-geminate duration (Miller, 1987) or vowel-to-consonant duration (Ridouane, 2010)). For example, Ridouane found that vowels that precede true geminates are shorter than those that precede fake geminates in Tashlhiyt Berber. Since Ridouane also found that absolute consonant duration was the same for true and fake geminates, the preceding vowel duration differences should have translated into relative duration differences between true and fake geminates. If a ratio of consonant to preceding vowel duration (C:V1) would have been used, then the true geminates of Tashlihiyt Berber would be were longer than fake geminates. The possibility that true and fake geminates might be distinguished by a C:V1 ratio is intuitively satisfying because it parallels a finding in the literature on phonemic length perception; namely, that C:V1 duration provides a robust higher-order cue to the singleton/geminate contrast in languages with a phonemic length contrast, such as Italian and Japanese (Idemaru & Guion, 2008; Pickett, Blumstein, & Burton, 1999).

As interesting as the Tashlhyit Berber results are, they nonetheless conflict with those reported by Lahiri and Hankamer (1988), who found no preceding vowel duration differences for true and fake geminates in Bengali. The conflicting results may, however, be due to a difference in the types of fake geminates that were examined in the two studies. Whereas Ridouane (2010) examined fake geminates that arose across a word boundary, Lahiri and Hankamer examined fake geminates that arose through affixation. It is likely that a word boundary would trigger boundary-adjacent syllable lengthening, which would explain the longer preceding vowel durations that Ridouane observed for fake geminates. If boundary-adjacent lengthening does not occur, then true and fake geminates could have the same preceding vowel durations, just as Lahiri and Hankamer observed for Bengali. If true and fake geminates do not differ in preceding vowel duration or in absolute consonant duration, then they will also have the same relative (C:V1) durations.

Along with many others (e.g., Byrd & Saltzman, 2003; Keating, Cho, Fougeron, & Hsu, 2003; Klatt, 1976; Lehiste, 1970; Turk & Shattuck-Hufnagel, 2000; Wightman, Shattuck-Hufnagel, Ostendorf, & Price, 1992), we assume that boundary adjacent lengthening occurs because boundaries are represented in the speech plan. We define the speech plan in the way that Levelt (1989, chap. 8) and many others have done; namely, as the multi-word phonetic plan that allows for fluent speech.

When our assumptions regarding boundaries are applied to make sense of Ridouane’s (2010) and Lahiri and Hankamer’s (1988) conflicting results, then the prediction is that word boundary fake geminates will differ from word-internal fake geminates even though both types emerge from morpheme concatenation. This prediction follows from the view that the different types of fake geminates are differently represented in the speech plan. Specifically, fake geminates that cross a word boundary are represented as a sequence of identical consonants, consistent with the traditional view of these segments. Such a representation will be phonetically indicated by juncture cues, such as boundary-adjacent lengthening, which result in fake geminates having shorter relative durations than true geminates. Word-internal fake geminates are represented as long consonants in the speech plan, in the same way as true geminates. This representation will be phonetically indicated by relative durations that are longer than those associated with word-boundary fake geminates and by a lack of phonetic boundary cues.

Our proposal that morpheme boundaries are absent in the phonetic representation that guides speech production would seem to be at odds with the finding of phonetic length differences of affixes in heteromorphemic words that vary systematically in their relative decomposability (Hay, 2007). Such a finding adds to many others that support a large class of dual-route processing models, which assume lexical representations that are both holistic and morphologically decomposed (Hay, 2002; Niswander-Klement & Pollatsek, 2006; Schreuder & Baayen, 1995; Wurm, 1997). Lexical items are accessed either directly in their holistic form or indirectly via their constituent morphemes depending on how decomposable they are. Lexical items that have higher decomposability ratios are preferentially accessed via their parts than those that have lower decomposability, which are in turn preferentially accessed as whole items. If the dual route model is relevant to the phonetic representation of items selected for implementation, then we might expect that the phonetic behavior of word-internal fake geminates will be the same as that for word-boundary fake geminates when the word-internal geminates occur in heteromorphemic words with high decomposability ratios. This would suggest that not all word-internal fake geminates have the same speech plan representation and so they may not all have the same phonetic signature. Word internal fake geminates that emerge in highly decomposable words may be represented as a sequence of two consonants, and so should pattern with word boundary geminates. Fake geminates that occur in less decomposable words may be represented as a single long consonant, and so should pattern differently from word boundary fake geminates.

1.2. Current study

The current study focused on the production of fake geminates across different boundary types in English to investigate the hypothesis that word-internal fake geminates may differ from those that arise across a word boundary. Word-internal fake geminates arising from affixation were compared to phrases with word boundary geminates and to words with singletons. Speech style was manipulated as a means of exploring representation via the phonetics. This method of exploration follows from Johnson, Flemming, and Wright’s (1993) assumption that careful, or hyperarticulated, speech makes obvious speech plan representations, which are often otherwise obscured through phonetic reduction in normal speech. Whereas Johnson et al. are explicit in their statement of this assumption, they note that most of the work that has investigated style-dependent production differences is guided by the same assumption (e.g., Kang & Guion, 2008; Moon & Lindblom, 1994; Smiljanic & Bradlow, 2005; Van Son & Pols, 1999). Our expectation was that speakers’ careful productions would highlight the representation underlying fake geminates and that these productions would therefore result in either greater consonantal length differences between singletons and fake geminates, or in the phonetic marking of morpheme boundaries in words or phrases with fake geminates, or both depending on the representation.

The effect of decomposability on production was also tested. Specifically, assimilated word-internal fake geminates were compared to concatenated word-internal fake geminates. Assimilated geminates arise in English with the Latinate prefix in- (e.g., immoral), whereas concatenated geminates arise with the Germanic prefix un- (e.g., unnamed). The Latinate and Germanic prefixes differ in that the former are more idiosyncratic and structure-changing than the latter, consistent with their division in Lexical Phonology into level 1 versus level 2 affixes (see Hay, 2002; Plag, 2003). The Lexical Phonological division of affixes into different levels references differences in “the productivity of participating morphemes, order of attachment, and their phonological interaction with the roots” (Vannest & Boland, 1999, p. 324). A corpus study by Baayen and Renouf (1996) confirms that the level 1 Latinate prefix in- is indeed less productive than the level 2 Germanic prefix un-. The Latinate prefix in- also has a smaller prefix likelihood value (0.094) than the Germanic prefix un- (0.283), as calculated by the number of English words beginning with the sequence in- or un- divided by the number of words with in- or un- prefixes (Wurm, 1997). Words with prefixes that are unproductive and have low prefix likelihoods are more likely to be accessed in composite form (as whole items) than words with prefixes that are productive and have high prefix likelihoods (Wurm, 1997). If this suggestion is correct, then in- and un- prefixed words may have different speech plan representations with regards to boundary information, which should lead to different phonetic outcomes: the production of fake geminates in un- prefixed words should pattern with the production of word boundary fake geminates, whereas the production of fake geminates in in- prefixed words should resemble that associated with true geminates.

2. Methods

2.1. Participants

Eight University of Oregon undergraduate students (4 males and 4 females) participated. All of the participants were native speakers of American English who reported normal hearing. Participants were compensated with course credit for their time.

2.2. Stimuli

The stimuli were 24 English words, all of which were found in the Webster’s Online Dictionary (2008). The words with singleton consonants and the phrases with word boundary geminates were chosen to match the words with word-internal geminates in terms of lexical stress and spelling—all geminate and singletons were spelled with double m’s or n’s (see Table 1).

Table 1.

Word-internal singletons and word boundary geminates were chosen to match words the in- and un- prefixed words with word-internal geminates. The table shows the stimulus set and the phonetic transcription of the target words. The ratio values shown for the target words are the mean frequency of the base morpheme was divided by the mean frequency of the whole word.

Word-internal singletons
Word-internal geminates
Word boundary geminates
ammonia [əmoʊnyə] (0) immoveable [ɪmuvəbəl] (0.02) dim morning [dɪm mɔrniŋ] (1)
immensely [ɪmɛnsli] (0) immoral [ɪmɔrəl] (0.11) grim magic [gɹɪm mædʒɪk] (1)
immunity [ɪmyunɪti] (0) immemorial [ɪməmɔ ɹiəl] (0.55) prim memorial [pɹɪm məmɔriəl] (1)
immigrational [ɪmɪgɹeɪ ʃənəl] (0.5) immeasured [ɪmmɛ ʒɚ1d] (1) slim moviedom [slɪm muvidəm] (1)
annex [ənɛks] (0) unnoticed [ʌnnoʊtɪst] (0.09) one nail [wʌn neɪl] (1)
innate [ɪneɪt] (0) unnamed [ʌnneɪmd] (0.24) fun name [fʌn neɪm] (1)
annoyed [ənɔɪd] (0) unnerve [ʌnnɝv] (0.9) fun noise [fʌn nɔɪz] (1)
innerve [ɪnɝv] (1) unnail [ʌnneɪl] (1) one nurse [wʌn nɝs] (1)
Open in a new tab

The words annoyed and unnamed have been previously identified as containing singleton and fake geminate /n/s, respectively (Kaye, 2005). The other words, including the matching words with singletons, were selected and categorized by the first author, who is a native speaker of Korean—a language with a phonemic durational contrast (Choi & Jun, 1998; Han, 1996). This author’s perceptions of nasal length were confirmed by asking 10 other native Korean speakers to judge nasal length in the selected words, using the following procedure: A female native American-English speaker, who was naïve to the purpose of the study, was asked to produce the stimuli listed in Table 1 in the frame sentence, I said ____ again. The speaker’s productions were recorded and then played back in random order to the 10 Korean listeners, who were instructed to rate the relative duration of /n/ and /m/ in each word on a 7 point scale (1=extremely short, 7=extremely long). The rating results were consistent with the first author’s categorizations of the stimuli. Korean listeners rated singleton nasals as much shorter than word-internal geminate nasals (singletons, M=1.9, SD=0.56; word-internal geminates, M=5.2, SD=0.68). Word boundary geminate nasals were determined to be even longer than word-internal geminate nasals (word boundary geminates, M=6.1, SD=0.63).

The reader will note that at least two of the 8 words in Table 1 that were determined to contain singletons are obviously prefixed (immigrational and innerve). That is, these words contain unbound stems, just like the words with geminates. The words immigrational and innerve were chosen, like all the others, to match the words with fake geminates in lexical stress and spelling. However, their status as singleton consonants suggests that our notion of decomposability and its effects on representation and production may be simplistic. In particular, it could be that decomposability is gradient rather than categorical (see, e.g., Hay & Baayen, 2005). If this is the case, then a simple comparison of in- prefixed to un- prefixed word-internal geminates may not provide a complete test of the effect of decomposability. The analyses took this into account by entering relative decomposability as a covariate.

Along these lines, it is worth noting that when we calculated relative decomposability following Hay (2001) we found that the mean decomposability ratio for in- prefixed words was lower than that for un- prefixed words (0.42 versus 0.56). We also found substantial variability within each category. This variability is shown in Table 1. The relative decomposability of individual lexical items in Table 1 is expressed on a scale from 0 to 1, with 0 representing non-decomposable items and 1 representing completely decomposable items. The exact values were obtained by calculating the frequency ratio of the base item to that of the whole item and then dividing by 100. Heteromorphemic words with base items that were at least 100 times more frequent than the whole word were assigned a decomposability value of 1. Base and whole item frequency ratios were calculated separately based on word frequencies in the Corpus of Contemporary American English (COCA).

2.3. Procedure

The 24 words shown in Table 1 were embedded in the carrier phrase, I said ____ again, randomly ordered into 3 different lists of phrases, with each list representing a single repetition of all 24 items. Each speaker was given several minutes to look over the stimuli before recording began. Unfamiliar words were defined and produced by the experimenter for the speaker. The experimenter then explained that speakers would be producing each phrase in their normal speaking style as well as in a careful speaking style. The careful speaking style was explained as being the clear speech style one might use with a foreigner. Once speakers were comfortable with the stimuli list and understood the style instructions, recording began. Speakers first produced the phrase in a normal style. The experimenter then asked, What did you say?, which was the speaker’s cue to produce the same sentence using a careful style. In this way, all speakers produced 72 phrases twice, yielding a total of 144 productions per speaker.

All recordings were made in the sound-insulated booth in the University of Oregon’s Phonetic-Phonology Laboratory using a SHURE SM10A microphone and Marantz PMD660 solid state recorder. The recorded utterances were saved in digital format, for later analysis.

2.4. Measurements

The spoken phrases were displayed in Praat as oscillograms and spectrograms. Acoustic duration of the first vowel and of the relevant nasal consonant was measured to obtain measures of absolute and relative duration. Non-durational boundary cues were also investigated. Fundamental frequency (F0) contours were tracked across the relevant VC(C)V sequences and the presence or absence of pauses was noted. The segmentation criteria that guided acoustic measurement and pause identification were as follows.

Nasal-vowel boundaries were identified in the waveform by a drop in acoustic energy, a drop in F1, the significant diminishment of F2 and near absence of higher formants. The absolute duration of the preceding vowel and subsequent nasal was calculated based on the segmentation procedure. When a pause intervened at a word or morpheme boundary, total nasal duration was calculated as the sum of preceding and following nasal duration. The relative duration of nasals was calculated using the measure that best distinguishes singleton and nasals in languages with phonemic duration (Idemaru & Guion, 2008; Pickett et al., 1999); namely, as the proportion of total nasal duration to preceding vowel duration (C:V1).

F0 was measured at 8 points across the relevant VC(C)V sequence: preceding vowel onset, midpoint, offset, syllable 1 offset, syllable 2 onset, following vowel onset, midpoint, and offset. The syllable 1 offset and syllable 2 onset measures were taken at the offset and onset of nasal duration, if a pause occurred, or before and after a discontinuity in the nasal portion of the waveform, if no pause occurred, or on either side of a midpoint in nasal duration, if no pause or discontinuity were present. As shown in Fig. 1, the discontinuity in the waveform served as a reference to determine the offset of the first syllable and the onset of the second syllable.

Fig. 1.

Fig. 1

Open in a new tab

Syllable offset and onset measures on either side of a syllable boundary, which was defined either by a pause, a discontinuity in the waveform or as the midpoint of the nasal. The first and the second vertical lines indicate the syllable 1 offset and syllable 2 onset, respectively. The example VCCV sequence shown here is from the word immovable.

2.5. Analyses

We used a mixed effects model with consonant type (singleton, word-internal geminate, word boundary geminate) and speech style (normal and careful) to evaluate the differences between singletons, word-internal geminates, and word boundary geminates on (1) the absolute nasal duration, (2) the relative nasal duration, and (3) the F0 differences between the syllable 1 offset and syllable 2 onset. Test item was nested within speaker and both were treated as random effects. A separate mixed effects model examined the differences between word-internal in- prefixed and un- prefixed geminates in order to account for the potential effect of different decomposability on duration and F0 disjuncture. Prefix type and speech style were treated as fixed factors and base frequency and decomposability (base-to-whole word frequency ratio) as covariates that were nested within speaker. Speaker was treated as a random effect. For each analysis, log-likelihood tests for the random effects were reported.

3. Results

The results suggest that speech plan representations encode word-internal singletons and word-internal geminates in in- prefixed words similarly and distinctly from word boundary geminates. The absolute duration of all word-internal fake geminates was greater than that of singletons in normal and careful speech, but only the relative duration of word-internal fake geminates was greater than that of word-internal singletons across speaking styles. All word-internal geminates also differed in absolute duration in normal and careful speech, but not in relative duration regardless of speaking style. F0 contours showed clear disjuncture in phrases with word boundary geminates, but not in words with internal geminates and singletons. In careful speech, the majority of two word phrases were uttered with a pause between the first and second word as were a small percentage of un- prefixed words with concatenated geminates. No pauses occurred in words with singletons or in those with an in- prefixed words with assimilated geminates. These results are presented in more detail below.

3.1. Geminate duration

The first set of analyses focused on the absolute and relative duration of word-internal geminates and their consonant-matched singleton and word boundary counterparts. The expectation was that all fake geminates would be longer than singletons, but that word-internal geminates would differ from word boundary geminates if their speech plan representations differed.

3.1.1. Absolute nasal duration

The results on absolute duration are presented in Table 2. There were significant effects of consonant type [F(2, 189)=227.649, p < 0.000], speech style [F(1, 957)=2522.288, p < 0.000] and a significant interaction between consonant type and speech style [F(2, 957)=155.810, p < 0.000]. Word-internal geminates were longer than singletons, but the same length as word boundary geminates. The difference between word-internal geminates and singletons was greater in careful speech than in normal speech.

Table 2.

Results from the mixed effect model that tested the effects of consonant type, speech style, and their interaction on absolute duration. The factor consonant type1 tested the effect of word-internal geminates versus singletons. The factor consonant type2 tested the effect of word-internal geminates versus word boundary geminates.

Random effects
Groups Estimate Std. error
Residual 614.569 28.095
Item|Subject 386.603 50.524
Fixed effects
Estimate Std. error t-Value Sig.
(Intercept) 122.057 3.039 40.151 0.000
Consonant Type1 −40.333 4.299 −9.382 0.000
Consonant Type2 8.026 4.299 1.867 0.063
Speech Style 86.380 2.530 34.140 0.000
Cons.Type1 × Speech Style −49.026 3.578 13.701 0.000
Cons.Type2 × Speech Style 9.979 3.578 2.789 0.005
AIC BIC logLik
10941.585 10951.673 10937.585
Open in a new tab

The effect of prefix type and speech style on the absolute duration of word-internal geminates was also examined. Results from this analysis are presented in Table 3. The effects of prefix type [F(1, 667)=63.143, p < 0.000], speech style [F(1, 748)=1897.763, p < 0.000], and their interaction [F(1, 748)=21.425, p < 0.000] were significant. The absolute duration of word-internal geminates in un- prefixed words was longer than the duration of word-internal geminates in in- prefixed words. This difference was again larger in careful speech than in normal speech, as shown in Fig. 2. The matched singleton and word boundary geminate values are shown in Fig. 2 for comparison.

Table 3.

Results from the mixed effect model that tested the effects of prefix type, speech style, and their interaction on the absolute duration of word-internal geminates.

Random effects
Groups Estimate Standard error
Residual 844.626 43.668
Basefreq|Subject 4.334 8.033
Decomratio|Subject 183.932 107.959
Fixed effects
Estimate Std. error t-Value Sig.
Intercept 123.827 2.897 42.742 0.000
Prefix Type −7.099 2.979 −2.383 0.017
Speech Style 101.078 2.966 34.077 0.000
Prefix Type × Speech Style −19.416 4.195 −4.629 0.000
AIC BIC logLik
7367.189 7381.105 7361.189
Open in a new tab
Fig. 2.

Fig. 2

Open in a new tab

Absolute nasal duration for word-internal singletons, word-internal geminates and word boundary geminates produced in normal and careful speech.

It is possible that the longer duration of word-internal geminates in un- prefixed words compared to those in in- prefixed words may be attributed to intrinsic differences in the duration of /n/ versus /m/ rather than to prefix type. This possibility is undercut, however, by the similar mean durations of /n/ and /m/ when these occur as singleton or word boundary geminates (singleton /n/=101.4 ms (SD=42.3); singleton /m/=99.4 ms (SD=30.4); word boundary geminate /n/=180.94 ms (SD=58.4); word boundary geminate /m/=175.58 ms (SD=53.5)). Thus, the durational difference between /n/ and /m/ word-internal geminates may have been due to the prefixes involved.

3.1.2. Relative nasal duration

The results on relative duration are presented in Table 4. There were significant effects of consonant type [F(2, 189)=296.448, p < 0.000], speech style [F(1, 957)=60.200, p < 0.000], and their interaction [F(2, 957)=42.821, p < 0.000]. Word-internal geminates were longer than either singletons or word boundary geminates. Again, this difference was greater in careful speech than in normal speech, as shown in Fig. 3.

Table 4.

Results from the mixed effect model that tested the effects of consonant type, speech style, and their interaction on relative duration. The factor consonant type1 tested the effect of word-internal geminates versus singletons. The factor consonant type2 tested the effect word-internal geminates versus word boundary geminates.

Random effects
Groups Estimate Standard error
Residual 0.041 0.002
Item|Subject 0.015 0.002
Fixed effects
Estimate Std. error t-Value Sig.
(Intercept) 1.793 0.021 85.304 0.000
Consonant Type1 −0.425 0.030 −14.310 0.000
Consonant Type2 −0.434 0.029 −14.624 0.000
Speech Style 0.247 0.021 12.031 0.000
Cons.Type1 × Speech Style −0.228 0.029 −7.854 0.000
Cons.Type2 × Speech Style −0.238 0.291 −8.166 0.000
AIC BIC logLik
−165.517 −155.429 −169.517
Open in a new tab
Fig. 3.

Fig. 3

Open in a new tab

Relative nasal duration for word-internal singletons, word-internal geminates and word boundary geminates produced in normal and careful speech.

In contrast to word-internal geminates, word boundary geminates were not different from singletons in relative duration [t(766)=1.99, SE=0.26, p=0.47]. The results on relative duration were therefore different from those on absolute duration, where word boundary geminates were found to be as long as word-internal geminates and so significantly longer than singletons.

The effect of prefix type on relative duration was again examined, but no significant effects were found (see Table 5). These results were also substantially different from those on absolute duration, where differences due to prefix type were found. The absence of prefix-dependent differences in relative duration in the presence of prefix-dependent differences in absolute duration indicates that the pre-consonantal vowel in the un- prefix was longer than the pre-consonantal vowel in the in- prefix. In other words, the absolute differences in /m/ and /n/ duration in in- and un-prefixed words were due to overall differences in the duration of the entire in- and un- prefix rather than to differences the consonants associated with these prefixes.

Table 5.

Results from the mixed effect model that tested the effects of prefix type, speech style, and their interaction on the relative duration of word-internal geminates.

Random effects
Groups Estimate Standard error
Residual 0.113 0.006
Basefreq|Subject 0.000 0.000
Decomratio|Subject 0.171 0.089
Fixed effects
Estimate Std. error t-Value Sig.
Intercept 1.869 0.035 53.446 0.000
Prefix Type 0.041 0.034 1.199 0.231
Speech Style 0.123 0.034 3.584 0.000
Prefix Type × Speech Style 0.012 0.048 0.244 0.807
AIC BIC logLik
562.886 576.802 556.886
Open in a new tab

To summarize, the absolute duration of word-internal and word boundary geminates were similar and substantially different from singleton duration; but once preceding vowel length was taken into account, word-internal and word boundary geminates differed and only word-internal geminates were longer than singletons. In this way, word-internal fake geminates behaved more like true geminates than did word boundary geminates.

3.2. Non-durational boundary cues

If word-internal geminates behave like true geminates and unlike word boundary geminates, then we might expect that word-internal geminates are represented in the speech plan in the same way as true geminates; namely, as a single long consonant rather than as a sequence of consonants. Such a representation implies that word-internal boundary information is absent in the plan that guides the execution of fluent speech. If this is correct, then the conclusion that word-internal fake geminates behaved like true geminates predicts that intonational and pausing patterns that cue boundaries will be absent in words with internal singletons and geminates, but not in phrases with word boundary geminates. This prediction was largely upheld in an analysis of the F0 contours and pausing patterns associated with the relevant VC(C)V sequences across the consonant types under investigation.

The results on F0 differences are shown in Table 6. There were significant effects of consonant type [F(2, 189)=107.823, p < 0.000], speech style [F(1, 957)=216.823, p < 0.000], and their interaction [F(2, 957)=198.616, p < 0.000]. The significant two-way interaction was due to the second component in the interaction. The first component of the interaction was not significant. The effect of prefix type on F0 differences was also examined, but no significant results were found (see Table 7).

Table 6.

Results from the mixed effect model that tested the effects of consonant type, speech style, and their interaction on F0 difference at the syllable boundary (offset F0 – onset F0). The factor consonant type1 tested the effect of word-internal geminates versus singletons. The factor consonant type2 tested the effect of word-internal geminates versus word boundary geminates.

Random effects
Groups Estimate Standard error
Residual 40.822 1.866
Item|Subject 6.311 1.384
Fixed effects
Estimate Std. error t-Value Sig.
(Intercept) 0.184 0.558 0.019 0.985
Consonant Type1 −0.017 0.789 −0.022 0.982
Consonant Type2 0.173 0.788 0.220 0.826
Speech Style 16.146 0.652 24.760 0.000
Cons.Type1 × Speech Style −0.578 0.922 −0.627 0.531
Cons.Type2 × Speech Style 15.621 0.922 16.939 0.000
AIC BIC logLik
7662.556 7672.644 7658.556
Open in a new tab

Table 7.

Results from the mixed effect model that tested the effects of prefix type, speech style, and their interaction F0 difference at the syllable boundary (offset F0 – onset F0) in the words with internal geminates.

Random effects
Groups Estimate Standard error
Residual 13.211 0.969
Basefreq|Subject 5.690 3.258
Decomratio|Subject 0.000 0.000
Fixed effects
Estimate Std. error t-Value Sig.
Intercept −0.532 0.442 −1.203 0.230
Prefix Type 0.170 0.529 0.332 0.748
Speech Style 1.062 0.525 2.025 0.044
Prefix Type × Speech Style 1.075 0.742 −1.448 0.148
AIC BIC logLik
2100.773 2112.593 2094.773
Open in a new tab

Fig. 4 shows the F0 contours associated with the normal and careful productions of VC(C)V sequences in items with word-internal singletons, in- and un- prefixed word-internal geminates, and in items with word boundary geminates. The figure shows that the source of the significant two-way interaction between type and style is a clear disjuncture between offset and onset for word boundary geminates in careful speech. Also evident in Fig. 4 are the smooth transitions between offset and onset for the other consonant types in the two speaking conditions.

Fig. 4.

Fig. 4

Open in a new tab

F0 measures taken at V1 onset, midpoint, offset, and syllable offset followed by F0 measures at syllable onset, V2 onset, midpoint, offset as a function of consonant type and speech style.

The pause data were consistent with the patterns evident in the F0 data. In careful speech, speakers almost always inserted a pause at the word boundary in phrases with fake geminates (95%). When speaking carefully, speakers also occasionally inserted a pause at the morpheme boundary in un- prefixed words (10%), and between the first and second syllable in words with singletons (3%), but never at the morpheme boundary in in- prefixed words. Pause duration in word boundary geminates was longer than in word-internal geminates or singletons (mean pause duration for word boundary geminates=273 ms, SD=115.3; word-internal geminates=202 ms, SD=58.1; and singletons=213 ms, SD=55.6). When speaking normally, speakers never inserted pauses within a word, but occasionally inserted pauses at word boundaries in the phrases (8%).

Together, the intonational contours and pausing patterns associated with production of the stimuli in normal and careful speech show that word boundaries are highlighted in production, and that the word-internal boundary associated with un- prefixed words may be more likely to be represented than that associated with in- prefixed words.

4. Discussion

The results from the current study suggest that the same sound may be represented with different relative lengths in different English words, even though segmental length is non-contrastive in English. Similar to true geminates, the assimilated fake geminates that occurred in in- prefixed words were longer in absolute and relative terms than matched singletons, especially in careful speech. These fake geminates were also never pulled apart in production, even in careful speech. Assuming that careful speech productions highlight speech plan representations, the combined results suggest that English in- prefixed word-internal geminates may be represented as long consonants.

The absence of phonetic boundary cues at the in-/root boundary in words with assimilated fake geminates suggests that these boundaries are not represented in the speech plan. This conclusion is consistent with the Lexical Phonological convention known as bracket erasure (Kiparsky, 1982; Mohanan, 1986). According to the bracket erasure convention, word-internal morphological information is removed once the (lexical) phonological rules that interact with morphological structure have applied. In the case of the level 1 in- prefix, the morpheme boundary would have been erased after the final nasal of the prefix once it had been assimilated to the place of the initial nasal in the root and before being passed out of the lexicon.

Whole word and many dual route models of lexical access also allow for the possibility that morpheme boundaries are absent in the plan that guides fluent speech (e.g., Butterworth, 1983; Bybee, 1985; Laudanna, Burani, & Cermele, 1994; Niswander-Klement & Pollatsek, 2006; Schreuder & Baayen, 1995; Wurm, 1997). For example, Hay (2002) argues that level 1 affixes, such as the in- prefix of the current study, are preferentially accessed via a direct route, which would entail that the holistic representations of these words guide their production. Such representations need not in themselves contain information about the constituent structure of the item. If the holistic representations do not contain information about constituent structure, then morpheme boundaries for these items would be absent in the speech plan.

The finding that the in- prefixed word-internal fake geminates in English, like true geminates, are longer than singletons in both absolute and relative terms can be interpreted to suggest either (1) that the different representations of true and fake geminates cannot be phonetically distinguished or (2) that, like true geminates, in- prefixed words with assimilated geminates are intrinsically long consonants. Whereas a Lexical Phonology framework encourages the first interpretation, whole word and dual route models of lexical access allow for the second.

In Lexical Phonology, words are always accessed via their constituent morphemes. Assimilated fake geminates are therefore seen as arising in English when the final sound of a level 1 prefix is assimilated, and so made identical to, the initial sound of the root to which it is attached. Once the morpheme boundary is deleted from the representation following the assimilation process, there may be no way to phonetically distinguish between length due to the production of a consonant sequence and length due to a single long consonant. At least in the current study, we assumed that the only way to establish a phonetic distinction between the two types of representation is by exploiting the possibility that a boundary is or is not represented in the speech plan.

In contrast to Lexical Phonology, whole word and dual route models of lexical access might be interpreted to support a view in which fake geminates are represented in the same way as true geminates; namely, as a single long consonant. This suggestion follows from the proposal that multimorphemic lexical items are either accessed directly (whole word models) or accessed following either a whole-route or decomposed route (dual route models), which means that lexical items are represented both as whole items and as decomposed items. If we make the further assumption that holistic representations do not make reference to constituent structure, and encode the phonological and phonetic information necessary for accurate production, then we might conclude that a long sound (i.e., fake geminate) may be straightforwardly represented in holistic representations as a segment with intrinsic length.

Fowler (1980) was an early advocate for the view that temporal information is an intrinsic part of speech sound representation in all languages—a view that was formalized by Browman and Goldstein in Articulatory Phonology (Browman & Goldstein, 1986, 1990, 1992) and is actively advocated for in the phonetics literature. In Articulatory Phonology, phonological representations are based on linguistically relevant vocal tract constrictions, referred to as gestures. The phonological specification of a lexical item is the pattern of relative timing of gestures corresponding to the sequencing and allophonic realization of the target sound pattern. These patterns are known as gestural scores, and every lexical entry is presumably associated with its own gestural score.

Gestural scores provide a straightforward specification of articulatory movement, which can be used to build speech plans. For example, in words such as immensely and immoral with singleton and fake geminate consonants, respectively, the gestural scores differ not only in the patterns associated with the second and third syllables, but also in the relative timing of their similar initial vowel-nasal-vowel sequences. The tongue body gesture associated with the second vowel would be phased later with respect to the bilabial and velic gestures associated with the /m/ in immoral when compared to the /m/ in immensely. Again, such information would be part of the lexical representation; a suggestion that seems compatible with the notion of holistic representation in whole word and dual route models of lexical access.

In contrast to the possibility that in- prefixed words with assimilated geminates in English are represented with length just like true geminates, our results strongly suggest that word-boundary fake geminates are represented as consonant sequences. Consider the result that word-boundary fake geminates were only ever longer than matched singletons in absolute terms. The finding that these geminates had the same relative durations as singletons suggests that preceding vowels were lengthened along with the consonants as part of a boundary-dependent lengthening process. Such a suggestion is consistent with the unsurprising finding that word boundary geminates were frequently pulled apart in careful speech via F0 changes and pause insertion. Both findings argue for a representation in which a consonant sequence spans a boundary that is also realized in the speech plan.

The results on un- prefixed words with concatenated geminates are more opaque. Like the in- prefixed geminates in this study, un-prefixed geminates were also longer than singletons in absolute and relative terms. The concatenated fake geminates of un- pre-fixed words nonetheless differed from the assimilated fake geminates in in- prefixed words in two respects: the absolute duration of the /n/ in un- prefixed words was longer than the absolute duration /m/ in in- prefixed words. And, more intriguingly, speakers occasionally marked the morpheme boundary between the un-prefix and root in careful speech via pause insertion.

The finding that un- prefixed geminates behaved very similarly to in- prefixed geminates in normal speech, but occasionally like word boundary geminates in careful speech, could indicate that un- prefixed words are sometimes represented with a morpheme boundary in the speech plan and sometimes without one. Whereas Lexical Phonology and whole word models of lexical access do not allow for different representations of the same class of words, dual route models of lexical access do.

The longer absolute duration of the un- prefixed geminates compared to the in- prefixed geminates was likely due specifically to the morpheme un- rather than to intrinsic differences in the duration of /n/ and /m/. Like Hay (2002, 2007), we assume that the durational differences mark differences in affix semantic transparency and productivity, which in turn may suggest differences in how words with such affixes are accessed. Whereas words with level 1 prefixes may be preferentially accessed as whole items, words with level 2 prefixes may be preferentially accessed via their constituent morphemes, especially when the lexical frequency of the base is high relative to the frequency of the whole word. A word that is accessed in a decomposed form will likely also have a segmented form representation. Thus, fake geminates that arise at a morpheme boundary in words and are accessed in a decomposed form may be represented as a consonant sequence, just like word boundary geminates. When this happens, un- prefixed words with concatenated geminates will behave more like word boundary geminates than like in- prefixed words with assimilated geminates. In the present study, this happened only occasionally in the careful speaking condition.

Under normal speaking conditions, the un- prefixed fake geminates behaved almost exactly like the in- prefixed fake geminates in the present study. Although the un- prefixed words may be preferentially accessed in decomposed form, it is conceivable that speakers in the present study would have defaulted to accessing the whole forms of these words for two reasons. First, all stimuli were repeated several times over the course of the experiment. This repetition would have boosted the frequency with which the stimulus words were accessed, which would have facilitated whole route access. Secondly, the stimulus set did not include items that would have encouraged the semantic decomposition of any of the prefixed words, nor did the frame sentences support the compositional semantics of the target words. For these reasons, we conclude that the similar behavior of in- and un- prefixed word-internal fake geminates in this study suggests similar representations in the speech plan.

5. Conclusion

Although English fake geminates are not phonemic, those that occurred in prefixed words had longer absolute and relative durations that those that occurred across a word boundary. With few exceptions, prefixed words were also produced without boundary cues, in contrast to two word phrases where the boundary was regularly identified via F0 changes and pause insertion in careful speech and often identified in the same way in normal speech. Assuming a dual-route model of lexical access and lexical-phonological representations of the sort proposed in Articulatory Phonology, the results suggest that word-internal fake geminates, and especially the assimilated fake geminates of in- prefixed words, are represented and accessed as long consonants in English. Only fake geminates in two word phrases exhibit phonetic characteristics that are wholly consistent with the representation of a consonant sequence.

Acknowledgments

We are grateful to Susan Guion-Anderson and Kaori Idemaru for commenting on previous versions of this paper, to Volya Kapatsinski and Tyler Kendall for help with formulating and executing the analyses, and to the anonymous reviewers who identified many ways in which the exposition and discussion of the results could be improved. This work was supported in part by NIH Grant R01 HD061458.

This research was supported in part by Award Number R01HD061458 from the Eunice Kennedy Shriver National Institute of Child Health & Human Development (NICHD). The content is solely the responsibility of the authors and does not necessarily represent the official view of NICHD or of the National Institutes of Health.

References

  1. Baayen RH, Renouf A. Chronicling the Times: Productive lexical innovations in an English newspaper. Language. 1996;72(1):69–76. [Google Scholar]
  2. Browman CP, Goldstein L. Towards an articulatory phonology. Phonology Yearbook. 1986;3:219–252. [Google Scholar]
  3. Browman CP, Goldstein L. Gestural specification using dynamically-defined articulatory structures. Journal of Phonetics. 1990;18:299–320. [Google Scholar]
  4. Browman CP, Goldstein L. Articulatory phonology: An overview. Phonetica. 1992;49:155–180. doi: 10.1159/000261913. [DOI] [PubMed] [Google Scholar]
  5. Butterworth B. Lexical representation. In: Butterworth B, editor. Language production. Vol. 2. London: Academic Press; 1983. pp. 257–294. [Google Scholar]
  6. Bybee J. Morphology: A study of the relation between meaning and form. Amsterdam: John Benjamins; 1985. [Google Scholar]
  7. Byrd D, Saltzman E. The elastic phrase: Modeling the dynamics of boundary-adjacent lengthening. Journal of Phonetics. 2003;31(2):149–180. [Google Scholar]
  8. Cho T. Effects of morpheme boundaries on intergestural timing: Evidence from Korean. Phonetica. 2001;58:129–162. doi: 10.1159/000056196. [DOI] [PubMed] [Google Scholar]
  9. Choi S, Jun J. Are Korean fortis and aspirated consonants geminates? Language Research. 1998;34(3):521–546. [Google Scholar]
  10. Fowler C. Coarticulation and theories of extrinsic timing. Journal of Phonetics. 1980;8:113–133. [Google Scholar]
  11. Han JI. Perception of Korean fortis and lenis consonants: Evidence for a geminate analysis of fortis consonants. In: Akatsuka N, Iwasaki S, Strauss S, editors. Japanese/Korean linguistics. Vol. 5. Akatsuka: CSLI Publications and SLA; 1996. [Google Scholar]
  12. Hay J. Lexical frequency in morphology: Is everything relative? Linguistics. 2001;39(6):1041–1070. [Google Scholar]
  13. Hay J. From speech perception to morphology: Affix-ordering revisited. Language. 2002;78(3):527–555. [Google Scholar]
  14. Hay J. The phonetics of ‘un’. In: Munat J, editor. Lexical creativity, texts and contexts. Amsterdam: John Benjamins Publishing Company; 2007. pp. 39–57. [Google Scholar]
  15. Hay J, Baayen RH. Shifting paradigms: Gradient structure in morphology. Trends in Cognitive Sciences. 2005;9(7):342–348. doi: 10.1016/j.tics.2005.04.002. [DOI] [PubMed] [Google Scholar]
  16. Idemaru K, Guion S. Acoustic covariants of length contrast in Japanese stops. Journal of International Phonetic Association. 2008;38(2):167–186. [Google Scholar]
  17. Johnson K, Flemming E, Wright R. The hyperspace effect: Phonetic targets are hyperarticulated. Language. 1993;69:505–528. [Google Scholar]
  18. Kang KH, Guion SG. Clear speech production of Korean stops: Changing phonetic targets and enhancement strategies. Journal of the Acoustical Society of America. 2008;124:3909–3917. doi: 10.1121/1.2988292. [DOI] [PubMed] [Google Scholar]
  19. Kaye AS. Gemination in English. English Today. 2005;21:43–55. [Google Scholar]
  20. Keating P, Cho T, Fougeron C, Hsu C-S. Domain-initial articulatory strengthening in four languages. In: Local J, Ogden R, Temple R, editors. Papers in laboratory phonology VI. Cambridge, UK: Cambridge University Press; 2003. pp. 143–161. [Google Scholar]
  21. Kiparsky P. Lexical morphology and phonology. In: Yang I-S, editor. Linguistics in the morning calm. Seoul: Hanshin; 1982. pp. 3–91. [Google Scholar]
  22. Klatt D. Linguistics uses of segmental duration in English: Acoustic and perceptual evidence. Journal of the Acoustical Society of America. 1976;59:1208–1221. doi: 10.1121/1.380986. [DOI] [PubMed] [Google Scholar]
  23. Lahiri A, Hankamer J. The timing of geminate consonants. Journal of Phonetics. 1988;16:327–338. [Google Scholar]
  24. Laudana A, Burani C, Cermele A. Prefixes as processing units. Language and Cognitive Processes. 1994;9(3):295–316. [Google Scholar]
  25. Lehiste I. Suprasegmentals. Cambridge, MA: MIT Press; 1970. [Google Scholar]
  26. Lehiste I, Morton K, Tatham M. An instrumental study of consonant gemination. Journal of Phonetics. 1973;1:131–148. [Google Scholar]
  27. Levelt W. Speaking: From intention to articulation. Cambridge, MA: MIT Press; 1989. [Google Scholar]
  28. McCarthy JJ. OCP effects: Gemination and antigemination. Linguistic Inquiry. 1986;17:207–263. [Google Scholar]
  29. Merriam-Webster Online Dictionary. 2008 Retrieved January 25, 2008, from 〈 http://www.merriam-webster.com/〉.
  30. Miller A. Phonetic characteristics of Levantine Arabic geminates with differing morpheme and syllable structures. Ohio State Papers from the Linguistics Laboratory. 1987;36:120–140. [Google Scholar]
  31. Mohanan KP. The theory of lexical phonology. Dordrecht: Reidel; 1986. [Google Scholar]
  32. Moon SJ, Lindblom B. Interaction between duration, context, and speaking style in English stressed vowels. Journal of the Acoustical Society of America. 1994;96:40–55. [Google Scholar]
  33. Niswander-Klement E, Pollatsek A. The effect of root-frequency, word frequency, and length on the processing of prefixed English words during reading. Memory and Cognition. 2006;34:685–702. doi: 10.3758/bf03193588. [DOI] [PubMed] [Google Scholar]
  34. Payne E. Phonetic variation in Italian consonant gemination. Journal of the International Phonetic Association. 2005;35(2):153–189. [Google Scholar]
  35. Pickett R, Blumstein S, Burton M. Effects of speaking rate on the singleton/geminate contrast in Italian. Phonetica. 1999;56(3–4):135–157. [Google Scholar]
  36. Plag I. Word formation in English. Cambridge: Cambridge University Press; 2003. [Google Scholar]
  37. Ridouane R. Geminates at the junction of phonetics and phonology. In: Fougeron C, Kühnert B, D’Imperio M, Vallée N, editors. Laboratory phonology 10: variation, phonetic detail and phonological representation. 2010. pp. 61–90. [Google Scholar]
  38. Schreuder R, Baayen RH. Modeling morphological processing. In: Feldman L, editor. Morphological aspects of language processing. Hillsdale, NJ: Erlbaum; 1995. pp. 131–154. [Google Scholar]
  39. Smiljanic R, Bradlow AR. Production and perception of clear speech in Croatian and English. Journal of the Acoustical Society of America. 2005;118(3):1677–1688. doi: 10.1121/1.2000788. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Turk A, Shattuck-Hufnagel S. Word boundary-related duration patterns in English. Journal of Phonetics. 2000;28:397–440. [Google Scholar]
  41. Vannest J, Boland JE. Lexical morphology and lexical access. Brain and Language. 1999;68:324–332. doi: 10.1006/brln.1999.2114. [DOI] [PubMed] [Google Scholar]
  42. Van Son R, Pols L. Perisegmental speech improves consonant and vowel identification. Speech Communication. 1999;29:1–22. [Google Scholar]
  43. Wightman CW, Shattuck-Hufnagel S, Ostendorf M, Price PJ. Segmental durations in the vicinity of prosodic phrase boundaries. Journal of the Acoustical Society of America. 1992;91:1707–1717. doi: 10.1121/1.402450. [DOI] [PubMed] [Google Scholar]
  44. Wurm LH. Auditory processing of prefixed English words is both continuous and decompositional. Journal of Memory and Language. 1997;37:438–461. [Google Scholar]

RESOURCES