LOCUS AND CAUSE OF PASSIVE VERB STEM LENGTHENING...

Locus and cause of passive verb stem lengthening

Aldo J. Mayro

Interdisciplinary Honors Thesis

A thesis in partial fulfillment of the requirements

of the

Interdisciplinary Honors Thesis

Written under the direction of

Karin Stromswold, PhD, MD

Department of Psychology, Rutgers Center for Cognitive Science

and

Paul de Lacy, PhD

Department of Linguistics

School of Arts and Sciences, Rutgers University

2015-2016

LOCUS AND CAUSE OF PASSIVE VERB STEM LENGTHENING 1

Abstract

This production study investigated the cause and locus of longer verb stem durations in passive sentences versus progressive active sentences. Previous research by Stromswold et al. (2003, under review) and Rehrig et al. (2015) demonstrates that verb stems (e.g. kick-) are longer in passive sentences than progressive active sentences. It is possible that this difference is due to monosyllabic lengthening because the progressive suffix –ing adds a syllable to the stem but the passive suffix –ed, overwhelmingly realized as [t] or [d] in the previous studies, does not. Longer passive verb stems might also be caused by phrase-final lengthening because verbs were syntactically phrase-final in passive sentences but not in progressive active sentences, and syntactic boundaries often align with prosodic boundaries (Kreiman 1982; Lehiste & Wang 1977). In the current study, 4 monolingual English-speaking undergraduates with a Northeastern New Jersey dialect produced passive sentences (the ball was kicked around town), progressive active sentences (the boy was kicking a ball), past active sentences (the boys kicked a ball), and perfective active sentences (the boy has kicked a ball). If monosyllabic lengthening alone is affecting verb stem duration, then passive, past active, and perfective active stems should all be the same length and longer than progressive active stems. If phrase-final lengthening is also affecting verb stem duration, passive stems should be longer than all other types, and perfective active and past active stems should be longer than progressive active stems. Analyses indicated that passive verb stems were longer than past active or perfective active verb stems, which were in turn longer than progressive active verb stems, suggesting that monosyllabic lengthening and phrase-final lengthening both significantly cause passive verb stem lengthening. Additionally, the vowel was the only segment of the verb stem that was consistently and significantly longer in passives than progressive sentences.


1. Introduction Stems have a longer duration in monosyllabic words than in disyllabic words

(Lehiste, 1972). However, the locus of duration differences has not been clearly identified. The first goal of this thesis is to determine exactly where lengthening occurs in the stem – whether it is in the stem vowel, stem-final consonant, or distributed across the stem syllable’s rime. The second goal is to determine how the phonological context of the stem can affect the effect of monosyllabic lengthening at the prosodic word level or phrase-final lengthening at the prosodic phrase level.

These questions were addressed through a production study in which participants read sentences containing CVC verb stems. Stems appeared in both monosyllabic and disyllabic forms. Monosyllabic forms were created by suffixing –ed (i.e. [t]/[d]) to the stem (the stem-final C was not permitted to be an alveolar stop, otherwise the form would be disyllabic – i.e. [CVt/d-ɪd]). Disyllabic forms were created by suffixing –ing (i.e. [ɪŋ]) to the stem. To adhere to particular phonological constraints, CVC stems had to vary segmentally. However, stimuli were balanced in such a way as to allow inter-stimulus comparison and take into account inter-segmental influences (such as vowel lengthening before voiced consonants).

Analyses of stimuli used in an earlier study (Stromswold, Kharkwal, Sorkin, Zola, 2003, under review) suggest that the locus of shortening in monosyllabic words is the vowel and not the coda. Given a similar set of verb stimuli and framing sentences (e.g., the boy was pushing girl, the girl was pushed by the boy), there was no significant active/passive (i.e., -ing/-ed) difference in onset length (p = 0.18) or coda length (p = 0.83), but there was significant active/passive difference in vowel length (p < 0.005). This result disagrees with a previous study by Turk & Shattuck-Hufnagel (2000) which suggests that a lengthening or shortening effect is generally found in the rime, but primarily within the coda. However, the stimuli analyzed in the earlier experiment all contained fricative/affricate codas (kiss, push, shove, sniff, touch), and did not address the issue of whether the locus of lengthening differed depending on the stem-final consonant type.

Section 2 contains background information on Stromswold et al.’s original experiment. Section 3 discusses the experiment design and methods. Section 4 presents study results. Section 5 contains a discussion of our results, explanations of conflicting data, and future directions to proceed.

2. Background

Adult native speakers of American English tend to alter their prosody and duration of words, phones, or other linguistic segments for various reasons. Aylett and Turk (2004) have found that speakers tend to lengthen or shorten particular utterances to maintain an average predictable duration across contrasting phrases. Additionally,


infrequent words or words that have not been previously expressed by the speaker are typically articulated with a longer duration because they are less predictable. Furthermore, it has been suggested by Gahl and Garnsey (2004) that verbs that are highly syntactically predictable (i.e. highly frequent and highly probable in certain syntactic contexts) tend to have a shorter average duration. Much previous work has shown that the prosody of an utterance can also delimit a parser’s interpretation. The intonation of a sentence can change with regard to rhythm, stress, or pitch depending on the speaker’s meaning. Furthermore, hearers are often able to perceive these prosodic speech cues and correctly predict upcoming syntactic structure (see Barrow et al., 2005; Beach et al., 1996; Birch and Clifton, 2002; Carlson et al., 2001; Katz et al., 1996; Kiel-gaard and Speer, 1999; Kraljic and Brennan, 2005; Lehiste, 1973; Lehiste et al., 1976; Schafer et al., 2000; Snedeker and Trueswell, 2003; Speer et al., 1996). These low-level acoustic cues may be used subconsciously by adults to disambiguate temporarily ambiguous sentences.

Semantic ambiguity is a common feature of English sentences that can occur even when a given string is grammatical. Some sentences are permanently ambiguous, e.g. The officer shot the criminal with a gun. The sentence has two potential syntactic structures based on which constituent the prepositional phrase (PP) is attached to – either the verb phrase (VP) or a noun phrase (NP). It is impossible to determine from the information provided whether the officer is using the gun to shoot the criminal, or whether the criminal simply has the gun in his possession.

Figure 1. Trees showcasing two plausible interpretations of The officer shot the criminal with the gun. In A the officer is using the gun, whereas in B the criminal is the owner of the gun.

A B

There are also sentences that are only temporarily ambiguous (so-called “garden path” sentences). Consider the garden path sentence The door closed suddenly startled


the sleeping children. Because readers prefer to minimize the quantity of tree structure in a given parse (Frazier & Fodor 1978), they often initially parse the sentence so that closed is the main verb with door as its subject. Once startled is processed, however, they realize that in the original parsing there is no NP to fill the subject argument of startled. Thus, everything prior to startled must be the subject of the sentence and closed must be a reduced relative clause describing door. So it’s when the reader reaches the true main verb startled that she backtracks, and the sentence becomes disambiguated with only one possible meaning.

Figure 2. The leftmost tree A exhibits an unsuccessful initial parse of the sentence The door closed suddenly startled the sleeping children. Because there is no subject NP for startled, the reader must backtrack and reparse the sentence with the more complex tree structure in B where the door closed suddenly is a single NP constituent.

A B

Stromswold, Kharkwal, Sorkin, & Zola (2003, under review) conducted an eye-tracking study which suggests adults use the lower-level acoustic cue of duration to predict the syntactic structure of temporarily ambiguous utterances. In their study, adults listened to progressive active and passive sentences (e.g. The girl was pushing the boy and The boy was pushed by the girl, respectively) and chose which of two pictures displayed on a screen best matched the meaning of the sentence. Although these sentences are designed to be morphosyntactically ambiguous up until the suffix (-ing or -ed), analyses of the participants’ eye-tracking data revealed that they were beginning to look towards the correct picture before they heard the inflection. In other words, they


were able to disambiguate the sentence structure as either active or passive while the sentences were still morphosyntactically ambiguous.

One likely account of how participants in Stromswold et al.’s study were able to reliably look towards the correct picture before the sentence became morphosyntactically disambiguated is that the audio stimuli contained acoustic cues for upcoming syntactic structure and participants were able to use these acoustic cues to predict the structure of sentences while they were still morphosyntactically ambiguous. Although ToBI transcription analyses (REF) failed to reveal any consistent differences in intonational patterns for active and passive sentences, acoustic analyses of the stimuli sentences’ individual morphemes revealed that verb stems in passive sentences were significantly longer than those in active sentences (322 ms versus 289 ms). Because there were no other significant differences between the durations of the determiner the, NP, or auxiliary was preceding the verb in the active and passive sentences, Stromswold et al. concluded that the participants used the duration of the verb stem to distinguish between active and passive sentences while the two types of sentences were still morphosyntactically ambiguous.

Because the speaker of sentences used in Stromswold et al.’s study was linguistically trained and, therefore, might have introduced these normally absent cues, a follow up study was conducted by Rehrig et al. (2015) to determine if naïve adults also shorten active progressive verb stems. In Rehrig et al.’s experiment, 7 linguistically-naïve adult monolingual English speakers were recorded reading active and passive sentences similar to the ones from the eye-tracking experiment. These sentences were also temporarily ambiguous until the verb inflection, -ed or -ing. ANOVAs of the sentences’ individual morpheme duration, pitch and intensity were conducted. These analyses revealed that there was indeed a robust difference in the duration of the verb stem for all 7 participants between passive and progressive active sentences (Rehrig et al. refer to this as a passive lengthening effect). It also seems that this effect was not intentional—the participants were neither speaking to someone else nor did they believe their recordings would be used in a later comprehension study. One further follow up study using the production data from Rehrig et al. (2015) revealed that the duration of passive stems was greater for verb stems with voiced over unvoiced codas as well as greater for verb stems with non-stop rather than stop codas (Beier, Rehrig, and Stromswold, 2015). These effects are most likely due to voiced oral stops typically having a longer vowel duration, as well as it being easier to lengthen both the vowel and coda with non-stop codas.

These results raise the question of what caused the verb stems used in Stromswold et al. (2003, under review) and Rehrig et al. (2015) to be consistently and robustly shorter in progressive active sentences as compared to passive sentences. The lengthening effect of passive stems observed in Rehrig et al.’s study was most likely not intentional, and is therefore probably phonological. In their discussion of these findings, Stromswold et al.


mention Gahl and Garnsey (2004) in a first pass attempt to explain the data. Gahl and Garnsey suggest that words tend to shorten according to their relative higher frequency (Hooper, 1976) or higher predictability of occurring in a given context (Jurafsky et al., 2001). So the higher frequency or probability of verbs in the active sentences would result in their shorter duration relative to the passive sentences. This is an unlikely explanation of the data, however. Although there was a significant durational difference between the verb stems of active and passive sentences, Stromswold et al. found no difference between the durations of the auxiliary was in either sentence type. However, as Stromswold et al. point out, the progressive auxiliary was used in the active VP was pushing is more frequent and probable than the passive auxiliary was used in the passive VP was pushed, due to the higher relative frequency of active constructions in English.

Stromswold et al. go on to give three possible phonological explanations for why progressive active verb stems are shorter than passive verb stems. The first explanation asserts that passive lengthening is occurring, driven by phrase-final lengthening. This refers to the mechanism by which speakers tend to lengthen the final syllable before a major prosodic phrase boundary (Scott, 1982; Turk & Shattuck-Hufnagel, 2007). These prosodic boundaries typically align with syntactic constituent boundaries, such as NP/VP, VP/PP, NP/PP, etc. (Kreiman, 1982; Lehiste & Wang, 1977). It is grammatical for verbs such as kick that require an object to be the final word of a passive syntactic phrase (e.g. The man was kicked). Conversely, it is ungrammatical for such verbs to be the final word of an active syntactic phrase, e.g. *The man was kicking. In Stromswold et al.’s stimuli, 50% were passive sentences and 50% were progressive active sentences. Thus, prosodic phrase-final lengthening could account for some of the difference in verb stem duration in passives and progressive actives.

Stromswold et al.’s second phonological proposal for why active stems were shorter than passive stems involves speakers’ tendency to either lengthen or shorten their productions so as to maintain an average predictability among utterances (Aylett and Turk, 2004; Levy and Jaeger, 2007; Jaeger, 2010). It is possible that the longer length of the active suffix along with the higher frequency and probability of active clauses in English could account for the shortening of active stems or lengthening of passive stems.

The third phonological explanation draws on the work of Beckman and Edwards (1990) who argue that in Germanic languages such as English, stems tends to be longer in monosyllabic words than polysyllabic words. This is likely due to polysyllabic shortening (i.e. monosyllabic lengthening). The progressive participle adds an additional syllable (the progressive inflection -ing) to the progressive active verb causing the verb stem to shorten, whereas the passive participle is non-syllabic and so polysyllabic shortening does not occur.


That the length disparity occurred only within the verb stem and not in earlier segments of Stromswold et al.’s stimuli indicates that this phonological phenomenon is operating at the word level. There are at least six well-researched mechanisms that affect word-level duration. They include word-final lengthening, word-initial lengthening, accentual lengthening, monosyllabic lengthening/polysyllabic shortening, syllable-ratio equalization, and phrase-final lengthening (Turk & Shattuck-Hufnagel, 2000). Monosyllabic lengthening is the most likely candidate to account for the verb stem lengthening in passive voice constructions, due to the uniformity of Stromswold et al.’s stimuli: was kicking vs. was kicked, and the robustness of the effect. Monosyllabic lengthening/polysyllabic shortening was first proposed by Lehiste (1972) in a study where she found that the length of the first syllable of words like stick [.stɪk.] is gradually shortened as more syllables are concatenated, as in sticky [.stɪ.ki.] and stickiness [.stɪ.ki.nɪs.].

Almost all the verbs used in Stromswold et al. and Rehrig et al.’s studies have monosyllabic stems and become disyllabic feet only when the -ing inflection is added. One exception is pat which has an alveolar stop coda and therefore takes the full syllabic passive suffix, [ɪd]. Monosyllabic lengthening was not present in the passive verb stem as a result. Another exception is tickle which is disyllabic and may remain so or become tri-syllabic in progressive active constructions depending on the speaker’s pronunciation: [.tɪ.klɪŋ.] or [.tɪ.kə.lɪŋ]. If it remained disyllabic, then no lengthening effect was found in passive verb stems as compared to progressive active verb stems. Other than these two exception cases, the monosyllabic passive verb stem was always longer than the disyllabic active verb stem in both Stromswold et al. and Rehrig et al.’s data.

Monosyllabic lengthening/polysyllabic shortening operates at the prosodic word level, rather than the syntactic level. Accordingly, the robust lengthening effect exhibited in Stromswold et al. (2003, under review) and Rehrig et al. (2015) should occur in a variety of sentence constructions, regardless of the syntactic structure. Under this assumption, we set out to investigate two separate questions. First, is the longer duration of passive verb stems versus active verb stems caused by monosyllabic lengthening, phrase-final lengthening, or both? The second question investigated was in which segment of the verb stem is the effect of monosyllabic lengthening and/or phrase-final lengthening localized?

3. Methods

3.1 Participants

The participants in the current study were 4 undergraduate (18-21 years old) monolingual English speakers who spoke with a Northeastern New Jersey dialect. This


regional dialect was determined via the online New York Times dialect survey. All participants were similar to the ones in Stromswold et al. (2003, under review) and Rehrig et al. (2015)’s studies. Participants with a history of hearing, speech, language, learning, or other neurological disorders were excluded, and all participants were naïve to the purpose of the experiment.

3.2 Stimuli

3.2.1 Sentences

As shown in Figure 3 below, the study included four types of sentences. Like Stromswold et al. (2003, under review) and Rehrig et al. (2015), we included passive (1A) and progressive active sentences (1D). We also included past active sentences (1B) and perfective active sentences (1C) as well. A list of all sentences used in the study is provided in Appendix B.

Figure 3. All framing sentences were of the four types listed below. In order to compensate for the progressive active type containing the only –ing participle, 2 additional tokens of this structure were added per verb.

Sample sentences:

A. Passive: The mother was hugged around the waist. B. Past active: The school girls hugged a favorite teacher. C. Perfective active: The parent has hugged a newborn baby. D. Progressive active (x3): The baby was hugging a tiny puppy.

We chose to include simple past and perfective actives in order to investigate whether passive lengthening is solely a word-level phonological process or whether higher level phonological processes (e.g., prosodic final lengthening or syntactic processes) are also involved. The addition of these two sentence types, however, gave us three instances of verbs with -ed participles and only one instance of verbs with the -ing participle. In order to achieve balance between -ed and -ing verbs, we included two additional tokens of the progressive active sentence per verb. This resulted in a total of six frame sentences per verb.

There is no way to avoid the phrase-finality of verbs that could contribute to stem lengthening in the passive sentences. Both full by-phrase passives and truncated passives contain VPs which may or may not be immediately followed by a PP (e.g., The ball was kicked by the girl vs. The ball was kicked again). The truncated passive was chosen for this study because the full passive contains the preposition by, which we have observed causes English speakers to delete the preceding -ed suffix when it is realized as a alveolar stop, [t] or [d].


Because intonational tone, or pitch accent, can cause lengthening (Turk & Sawusch, 1997), we attempted to control for this in the verb stem by keeping the number of syllables and phonemes preceding the verb relatively consistent across verb stems. Analyses of variance revealed there were no significant differences in number of phonemes or syllables across the 4 sentence types or the manners of articulation of the stem codas (all p’s > .10). To avoid unwanted list intonation, we varied the number of post-verb syllables in the sentences, still keeping the number of syllables constant across the 4 sentence types.

In all sentence types, all verbs immediately preceded the vowel [ə], either as the determiner a in the active sentences (see 1B-1D) or as the first segment of a preposition in the passive sentences (see 1A). The reason for this is that [ə] is the most articulatorily ‘neutral’ of all vowels, given its mid-central location. Additionally, we chose not to use frame sentences in which the determiner the followed the verb, as the voiced fricative [ð] tends to delete preceding -ed inflections. In all 4 sentence types, the verb was preceded by the voiced sibilant [z] because [z] can be reliably segmented in most phonological contexts (Turk et al., 2006). In the passive and progressive sentences, [z] was the coda of was, and in perfective active sentences, [z] was the coda of has. Because past active constructions do not contain either auxiliary morpheme, we selected only subject NPs which epenthesize a [z] in the plural form (e.g., girls).

Additionally, we chose only high written-frequency NPs which are commonly understood by psychology major undergraduates in 2015. Nouns varied according to the semantic requirements of the chosen verbs. We attempted to select different NPs for each sentence to keep every sentence as novel as possible.

There were a total of 216 target sentences (3 coda types x 6 verbs x 6 sentence types x 2 instances) per participant.

3.2.2 Verbs

Phonological constraints:

Eighteen verbs were selected for the study, with each verb appearing in each syntactic construction. Verbs were selected because they met phonological constraints including the features of voicing, continuance, vowel segment duration, and manner of articulation. We attempted to maintain a similar number of syllables and phonological segments preceding the verb stem across verbs, vowels, and coda features. The online MRC Psycholinguistics Database (Wilson, 1988) was used to find English verbs that fit the criteria. All verbs were monosyllabic and had the form CVC. Verbs containing consonant clusters, i.e. CCVC or CVCC, were excluded in order to allow easier comparison between verbs and minimize possible locations of shortening within the rime.


The stem-final consonants were limited to three manners of articulation: oral stop, fricative, and nasal stop. Stems with liquid final consonants were excluded due to a lack of high frequency English verbs containing this segment, and because of the difficulty of accurately segmenting vowel+liquid sequences (Turk et al., 2006). Verbs containing a velar nasal [ŋ] coda were excluded due to a lack of high frequency verbs. Fricative codas containing the voiced consonants [v] or [ð] were also avoided because of the difficulty of segmenting them from vowels (Turk et al., 2006). Furthermore, none of the selected verbs had codas that were alveolar stops, [d] or [t], as this would result in the [ɪd] passive inflection, thereby adding an unwanted syllable.

There were six verbs for each different manner of articulation coda. Half of the verb stem-final consonants were [+voice] (all 6 nasals and 3 stops) and the other half were [�voice] (all 6 fricatives and 3 stops). This was done to balance the number of allomorphs of the –ed inflectional suffix: i.e. in regular productive morphology, -ed is realized as [t] after [�voice] segments and [d] after [+voice] segments (and [ɪd] after coronal stops, which were excluded here).

All CVC verbs contained one of the short lax vowels [ɪ], [æ], and [ʌ] to allow for easy comparison between verbs containing vowels which are all approximately the same phonological length. We chose short vowels over long vowels (e.g. [i], [u], or diphthongs) because long vowels tend to have a much more variable duration, making it difficult to tell how much they have been shortened or lengthened. Peterson & Lehiste (1960) found that the short vowels [ɪ] and [ʌ] (their [ə]) are approximately the same length at 161 ms and 181 ms respectively. The American English [æ], however, is much longer at 284 ms. In fact, this is longer than English long vowels, e.g. [iː] or [uː]. Nonetheless, [æ] is still a phonologically short vowel in American English.1 The treatment of [ɪ] and [ʌ] as shorter short vowels and of [æ] as a longer short vowel allows us to consider the first two as a verb set to be contrasted with [æ] verbs. For this reason, 9 of the 18 verbs contain the [æ] vowel while the remaining 9 are either [ɪ] or [ʌ]. Unfortunately, there were too few verbs with an [æ] vowel and fricative coda that met our constraints, so we could only include two fricative coda [æ] verbs, and we included an extra nasal coda [æ] verb to compensate.

Onset consonants have very little effect on a following vowel’s duration (Van Santen, 1992), and so were not controlled, though sibilant-initial stems were avoided due to the difficulty of segmenting these verbs from the preceding auxiliary was.

Semantic constraints:

1 [æ]’s phonological status is clearly seen in its behavior, as in the American English minimal word

restriction (e.g. *[bɪ], *[bʊ], *[bæ], cf. [bi], [bu], [bɑ]).


The verbs all have relatively high frequencies according to Thorndike-Lorge spoken, fiction, and magazine frequency counts (Spoken mean, range: 3583.89, 18957; written fiction mean, range: 7179.22, 29380; written magazine mean, range: 4082.83, 19018 - See Appendix A for a complete list of verbs with their frequencies). Verbs with obvious lewd or sexual connotations (e.g. bang) were excluded.

Our verbs differed from Stromswold et al. and Rehrig et al.’s in the following way: there was no longer a need for the verbs to be semantically reversible or exclusively actional. The frame sentences are not an exclusive progressive active vs. passive binary, and we did not intend the same noun and verb pair to be used in the binary.

Syntactic Constraints:

All verbs were transitive. We excluded verbs that are likely to become adjectival passives rather than verbal passives, such as serve or load (Levin & Rappaport, 1986). Such verbs might be processed as a different lexical category (adjective rather than verb) and so yield incomparable production data.

3.3 Catch Trials and Fillers

In addition to the total 216 experimental sentences each participant read, there was an additional 36 catch trial questions (one for each of the 18 verbs x 2 instances). These were introduced to ensure that participants processed the sentences that they read aloud. Catch trials were spaced out across the study, presented to participants after every 4-6 sentences. Each catch trial consisted of a single yes or no comprehension question about the sentence that the participants had just read. For example, after reading aloud the sentence, The worker was cashing a welfare check, the participant was asked, Was a welfare check being written by the worker?. The syntactic form of the questions varied pseudorandomly across verbs. The participant would be prompted by these questions to respond either yes or no by pressing the Q or P key on the keyboard in front of them. The correspondence between Q/P and yes/no varied by question.

After each of the 36 catch trials, participants read filler sentences. Filler sentences were introduced because comprehension questions can affect participant’s production data when they immediately precede target sentences. The filler sentences were designed to be similar to the experimental sentences in that they alternated between the 4 syntactic structures as well as the number of syllables preceding the verb. However, none of the NPs and verbs selected for the fillers appeared in the experimental sentences.

3.4 Equipment and Materials

The speakers were recorded in a sound-attenuated booth within the Rutgers University Phonology Lab using an AKG C420 Close Talk, head-mounted microphone and a Marantz PMD 670 solid state digital audio recorder. The experiment was coded


using E-Prime 2.0 Professional and run on a Lenovo ThinkPad W530 laptop within the booth. Participants read from an HP Pavilion w1907 LCD wide-screen flat panel monitor and navigated the experiment on an HP K1500 keyboard.

3.5 Procedure

Prior to beginning the experiment, each participant was given a short survey in which they were asked about speech, hearing, and other physiological disorders. In addition, they were evaluated aurally by the experimenter for speech characteristics that would make segmentation difficult (e.g. creakiness or turning nasal codas into nasalized vowels).

Participants performed the experiment in a sound attenuated booth, wearing a head mounted microphone. To familiarize themselves with the experimental protocol, participants read 5 practice sentences and answered a practice catch trial question. During the practice phase, the experimenter answered participants’ questions and adjusted the microphones and volumes as necessary. Participants were instructed to say each sentence once silently in their heads before reading them aloud. To reduce the chance of undesirable intonation, we asked all participants to produce each sentence in their ‘normal’ speech, as if casually talking to a close friend or sibling. If they felt that they misread a sentence, they were instructed to take a deep breath and try again. After each successful production, participants proceeded to the next trial by pressing the spacebar, followed by a 500 ms fixation crosshair. All 108 experimental sentences were presented in pseudorandom order. Throughout the experiment, participants were in direct communication with the experimenter via high quality Sennheiser headphones

Every 4-6 experimental sentences, participants were presented with a binary catch trial question. They then selected their yes/no answer by key press of either Q or P. The yes/no equivalent of Q and P was also pseudorandomized.

After participants read the first 108 experimental sentences, they took a 10 minute break outside the booth. During this time, they drank water and completed a brief online questionnaire on regional dialects from The New York Times. After they completed the questionnaire, the participants reentered the booth and began a second set of the same 108 experimental sentences and catch trials in a different pseudorandom order. In total, the experiment took 1 hour to 1 hour and 20 minutes to complete.

3.6 Acoustical Analyses

Production data was recorded using GoldWave digital audio recording software and saved as .WAV files (sampling rate: 44.1 kHz, quantizing rate: 16-bit mono). Boundaries between the relevant features of the verb stem were demarcated by hand by a single coder. Individual phonemes were segmented as follows: for slices preceding vowels, the wav method (i.e., the zero-crossing before the fist upswing of the first


undistorted period); for slices preceding fricatives, at the zero-crossing of the first upswing of the first smooth low-amplitude period following a vowel2; for slices preceding stops, at the final zero-crossing before closure; for slices preceding nasals and rhotics, at the zero-crossing before the first upswing of the first highly contrastive period. Duration lengths were determined and captured in text grids using Praat, audio software used by phoneticians to analyze acoustic properties of speech. SPSS was used for all statistical analyses, including ANOVAs designed to determine the effect of sentence type and phonological segments on verb stem duration.

4. Results

4.1 Sentence Type Effects on Verb Stem

If monosyllabic lengthening is the only factor that affects verb stem duration, we would expect the three -ed sentence types (passive, past active, and perfective active) to have verb stems of equal length and significantly longer than the one -ing sentence type (progressive active). If there is an effect of prosodic phrase-final lengthening in addition to monosyllabic lengthening, then the stem of passive sentences should be the longest because, of the 4 sentence types, only passives have a major syntactic phrase boundary after the verb, and syntactic boundaries often align with prosodic boundaries (Kreiman 1982; Lehiste & Wang 1977). Both the past active and perfective active sentences should have stems of equal length and shorter than the passive. The progressive actives should, again, have the shortest stem duration.

In order to determine the effect of sentence type on verb stem duration, we performed a one-way analysis of variance (ANOVA) with subject and verb as random factors on collapsed participant data. We found a main effect of sentence type on verb stem length (passive = 309 ms, past active = 305 ms, perfective active = 295 ms, progressive active = 270 ms; F(3, 14) = 8.98, p = .002). As we would expect from monosyllabic lengthening, planned comparisons revealed that the verb stems of progressive active sentences were significantly shorter than the stems of passive, past active, and perfective active sentences (all p’s < .0005 with Bonferroni correction for multiple comparisons).

2 It should be remarked that this method, as such, is not optimal for segmenting vowels from fricatives, as

it is a fairly subjective criteria. A better method would have been to put a boundary at the zero-crossing nearest to where F2 drops significantly. As a measure of mouth openness, F2 is a good indicator of the transition from a vowel to a fricative. Fortunately, the resulting boundary from the method used in this study very often converged on a drop in F2.


Figure 4. The verb stems of progressive active (-ing) sentence types were significantly shorter than any of the other three (-ed) sentence types. **** denotes significance level p < .0005. Error bars denote standard errors.

In the same ANOVA, we also found a main effect of verb (F(17, 44) = 11.50, p < .0005) and a main effect of subject (F(3, 11) = 30.59, p < .0005), in addition to a significant interaction between sentence type and verb (F(51, 153) = 1.78, p = .004) and sentence type and subject (F(9, 154) = 2.76, p = .005). We would expect the main effect of verb on verb stem duration, as each verb chosen for this experiment has an intrinsically different length – e.g. pick has two unvoiced stop consonants and a phonetically short vowel, while ram has two voiced consonants with a phonetically long vowel. Additionally, we would expect to find the main effect of subject because it is uncontroversial that different people talk at different speech rates. The interaction between sentence type and verb accords with Rehrig et al. (2015) in which it was found that the lengthened duration of passive stems was greater for verb stems with voiced versus unvoiced codas and for non-stop versus stop codas. The interaction between sentence type and subject reveals that the participants displayed different arrangements of verb stem lengthening for -ed sentence constructions.

In order to more precisely determine these differences, we performed separate ANOVAs for each participant’s verb stem data, again using sentence type as the fixed factor and verb as a random factor. There was a significant main effect of sentence type for all 4 participants: participant 1 (F(3, 51) = 3.40, p = .024), participant 2 (F(3, 51) = 3.99, p = .013), participant 3 (F(3, 51) = 13.53, p < .0005), participant 4 (F(3, 52) = 17.06, p < .0005). Planned comparisons using Bonferroni corrections were performed. In general, progressive active verb stems were shorter than the three -ed sentence types. For participants 3 and 4, planned comparisons revealed that the progressive verb stems were significantly shorter than all 3 types of -ed verb stems. For participant 1, the progressive

200 220 240 260 280 300 320 340

Passive Past active Perfective active Progressive active

Mea

n St

em D

urat

ion

(ms)

Sentence Types

Stem Duration by Sentence Type

**** ****

****


active verb stems were only significantly shorter than the passive verb stems. For participant 2, the progressive active verb stems were significantly shorter than both passive and perfective active stems (see Figure 5 below). See Appendix C for each participant’s individual mean differences and p values for each comparison.

Figure 5. In general, progressive active verb stems were significantly shorter for each participant. Error bars denote standard errors.

4.2 Stem Duration across -ed Sentence Types

In order to investigate further differences among the three -ed types, we first performed an ANOVA of collapsed participant data excluding progressive verb stems, and again using sentence type as the fixed factor and verb as a random factor. We found no effect of sentence type (F(2, 8) = 1.69, p = .245). There was, however, a main effect of verb (F(17, 18) = 8.18, p < .0005). Separate ANOVAs for each participant’s data were subsequently performed. No effect of sentence type was found for participant 1 (F(2, 34) = 1.75, p = .19), participant 2 (F(2, 34) = 1.21, p = .31), or participant 3 (F(2, 34) = .366, p = .70). There was a main effect of sentence type for participant 4 (passive = 365 ms, past active = 383 ms, perfective active = 349 ms; (F(2, 35) = 3.37, p = .046)). Planned pair-wise comparisons of all 3 –ed types in participant 4 revealed that the only significant

240

260

280

300

320

340

360

380

400

1 2 3 4

Mea

n St

em D

urat

ion

(ms)

Subject

Stem Duration by Sentence Type

Passive

Past active

Perfective active

Progressive active


difference was that past active stems were significantly longer than perfective active stems (p = .015 with Bonferroni correction for multiple comparisons).

Figure 6. We only found a significant difference in participant 4’s past active and perfective active verb stems. Error bars denote standard errors.

We also found a main effect of verb for participant 2 (F(17, 34) = 5.60, p < .0005), participant 3 (F(17, 34) = 7.64, p < .0005), and participant 4 (F(17, 34) = 5.02, p < .0005), and a marginal effect of verb in participant 1 (F(17, 34) = 1.76, p = .076). This is an unsurprising find for reasons listed in section 3.1.

Finally we found an interaction of sentence type and verb for participant 1 (F(34, 52) = 2.99, p < .0005) and participant 3 (F(34, 54) = 2.44, p = .002), but not for participant 2 (F(34, 54) = 1.44, p = .116) or participant 4 (F(34, 51) = 1.49, p = .098). Visual inspection (see Appendix D) suggests that participant 1’s sentence by verb interaction is due to cash being extremely long in one passive sentence (one token of 612 ms). No clear outlier to explain the interaction could be identified in participant 3.

4.3 Locus of Stem Shortening

240

260

280

300

320

340

360

380

400

1 2 3 4

Mea

n St

em D

urat

ion

(ms)

Subject

Stem Duration by Sentence Type (3 -ed types)

Passive

Past active

Perfective active


One of the primary goals of this experiment was to locate the constituent within the verb stem which is primarily affected by monosyllabic lengthening or any other phrase-boundary-related duration effect. We segmented the onset, vowel, and coda of each CVC verb stem and performed separate ANOVAs for each subject and each segment (i.e. onset, vowel, coda), with verb choice as a random factor. In these analyses we compared only the passive and progressive active sentence types, as these types most consistently resulted in the greatest stem duration disparity for each verb.

4.3.1 Onsets

There was no effect of sentence type on onset duration when participant data was collapsed (F(1, 5) = 3.62, p = .112). Additionally, there was no consistent pattern across subjects. There was a main effect of sentence type for participant 2 (passive = 103 ms, progressive active = 92 ms; (F(1, 17) = 5.82, p = .027)) and a marginal effect for participant 1 (passive = 105 ms, progressive active = 90 ms; (F(1, 17) = 3.43, p = .082)), but neither participants 3 or 4 showed any effect of sentence type (F(1, 17) = .831, p = .375 and F(1, 18) = .013, p = .911, respectively). There was a main effect of verb for all participants: participant 1 (F(17, 17) = 2.28, p = .050), participant 2 (F(17, 17) = 9.11, p < .0005), participant 3 (F(17, 17) = 4.47, p = .002), participant 4 (F(17, 17) = 4.02, p = .003). Finally, there was a significant interaction between sentence type and verb for participant 1 (F(17, 106) = 5.13, p < .0005) and participant 2 (F(17, 107) = 1.70, p = .054). However there was only a marginal interaction for participant 3 (F(17, 106) = 1.57, p = .087) and no interaction for participant 4 (F(17, 106) = 1.22, p = .265).

The main effect of verb is unsurprising. Onset phones varied in their inherent lengths due to differences in voicing and continuance – features known to cause greatly varying durations. The interaction in participant 1 is, again, likely driven by the outlier cash. The interaction in participant 2 might be due to apparent duration disparities across sentence types in the onsets of fan, kiss, or tug.


Figure 7. No significant difference. Error bars denote standard errors

Figure 8. Passive and progressive active onset duration was significantly different for participant 2 (p = .027), but only marginally significant for participant 1 (p = .082), and not significant for participant 3 or 4. Error bars denote standard errors.

4.3.2 Vowels

For all 4 participants, vowel duration was longer in passives than progressives: participant 1 passive = 114 ms, progressive active = 84 ms, F(1, 17) = 21.52, p < .0005; participant 2 passive = 117 ms, progressive active = 87 ms, F(1, 17) = 34.57, p < .0005; participant 3 passive = 123 ms, progressive active = 95 ms, F(1, 17) = 110.07, p < .0005;

20

40

60

80

100

120

Passive Progressive active

Mea

n O

nset

Dur

atio

n (m

s)

Sentence Type

Onset Duration by Sentence Type

70

80

90

100

110

120

130

1 2 3 4

Mea

n O

nset

Dur

atio

n (m

s)

Subject

Onset Duration by Sentence Type

Passive

Progressive active

ns


participant 4 passive = 139 ms, progressive active = 106 ms; F(1, 17) = 70.10, p < .0005). We also found a main effect of verb for every participant: participant 1 (F(17, 17) = 7.93, p < .0005), participant 2 (F(17, 17) = 11.33, p < .0005)), participant 3 (F(17, 17) = 30.50, p < .0005), participant 4 (F(17, 17) = 22.77, p < .0005)). There was a significant interaction between sentence type and verb for participant 1 (F(17, 106) = 6.40, p < .0005), participant 2 (F(17, 107) = 2.90, p < .0005), and participant 4 (F(17, 106) = 1.96, p = .02), but not participant 3 (F(17, 106) = 1.15, p = .321).

Once again, the main effect of verb is unsurprising because we chose the three vowels [æ], [ɪ], and [ʌ] due to their inherently varying lengths: [æ] had a mean duration of 158 ms, [ʌ] had a mean duration of 98 ms, and [ɪ] had a mean duration of 77 ms. The interaction in participant 1 is likely due to the outlier cash. The interaction in participant 2 is possibly due to apparent duration disparities across sentence types in can, tag, or tug (See Appendix D). No clear outlier could be identified to explain the interaction in participant 4.

Figure 9. Mean vowel duration is about 31% longer in passive verb stems than active verb stems. **** denotes significance level of p < .0005. Error bars denote standard errors.

0

20

40

60

80

100

120

140

Passive Progressive active

Mea

n Vo

wel

Dur

atio

n (m

s)

Sentence Type

Vowel Duration by Sentence Type

****


Figure 10. The difference in duration between passive and progressive active vowels was significant for all participants. Error bars denote standard errors.

4.3.3 Codas

There was no effect of sentence type on coda duration when participant data was collapsed (F(1, 5) = .294, p = .612). Additionally, there was no consistent pattern across participants. Consistent with Turk & Shattuck-Hufnagel (2000), participant 4’s codas were longer in monosyllabic passives than in disyllabic progressives, (passive = 108 ms, progressive active = 90 ms; (F(1, 17) = 7.38, p = .015)). But participant 2’s progressive active codas were longer than passive codas (passive = 63 ms, progressive active = 74 ms; (F(1, 17) = 5.65, p = .030)). No effect of sentence type was found in participant 1 (F(1, 17) = .116, p = .737)) or participant 3 (F(1, 17) = 2.88, p = .108)). For all participants, there was a main effect of verb: participant 1 (F(17, 17) = 6.81, p < .0005), participant 2 (F(17, 17) = 8.87, p < .0005), participant 3 (F(17, 17) = 6.47, p < .0005), participant 4 (F(17, 17) = 6.10, p < .0005). There was also a significant interaction between sentence type and verb choice for all participants: participant 1 (F(17, 107) = 5.36, p < .0005), participant 2 (F(17, 106) = 2.80, p = .001), participant 3 (F(17, 106) = 5.69, p < .0005), participant 4 (F(17, 106) = 7.09, p < .0005).

We were unsurprised by the main effect of verb on coda duration for the same reason as the main effect of verb on onset duration. The interaction between sentence type and verb stem is consistent with the findings of Rehrig et al. (2015).

70

80

90

100

110

120

130

140

150

1 2 3 4

Mea

n Vo

wel

Dur

atio

n (m

s)

Subject


Passive

Progressive active


Figure 11. No significant difference. Error bars denote standard errors.

Figure 12. We found a significant effect of sentence type on the coda constituent for participants 2 and 4, but in opposite directions. No significant effect for participants 1 and 3. Error bars denote standard errors.

4.4 Vowel Duration by Coda

Another primary motivation for this experiment was to determine how various phonetic features of coda might affect verb stem duration in passive versus progressive active sentences. We looked at three coda features: voicing, manner of articulation, and continuancy. The analysis in section 4.3 revealed that vowels were the only segment that

0 10 20 30 40 50 60 70 80 90

100

Passive Progressive Active

Mea

n C

oda

Dur

atio

n (m

s)

Sentence Type

Coda Duration by Sentence Type

0

20

40

60

80

100

120

140

1 2 3 4

Mea

n C

oda

Dur

atio

n (m

s)

Subject

Coda Duration by Sentence Type

Passive

Progressive active

ns


all 4 participants consistently and significantly lengthened in the passive versus progressive active sentence types. Accordingly, we only performed ANOVAs of the vowel duration of passive and progressive active sentences.

Separate ANOVAs were performed for each participant with coda voicing, coda manner of articulation, or coda continuancy as the random factor.

4.4.1 Coda Voicing

When participant data was collapsed, there was a main effect of coda voicing with voiced codas resulting in longer vowel durations (F(1, 565) = 47.12, p < .0005). There was also a marginally significant interaction between sentence type and coda voicing (F(1, 565) = 3.27, p = .071). Individual participant analyses revealed that for all 4 participants, there was a large and significant effect of coda voicing with stems with voiced codas having longer vowel durations (between 14.7-27.8% longer). Participant 1 voiced = 101 ms, unvoiced = 79 ms, F(1, 138) = 12.26, p = .001; participant 2 voiced = 105 ms, unvoiced = 85 ms, F(1, 139) = 16.69, p < .0005; participant 3 voiced = 109 ms, unvoiced = 95 ms, F(1, 138) = 6.71, p = .011; participant 4 voiced = 127 ms, unvoiced = 101 ms, F(1, 138) = 14.82, p < .0005. Consistent with Rehrig et al.’s (2015) finding that verbs with voiced codas lengthened more when produced in passive sentences than verbs with unvoiced codas, for participant 2 there was a significant interaction between sentence type and coda voice (F(1, 139) = 4.96, p = .028). However, there was no significant interaction between sentence type and coda voicing for the other 3 participants (all F’s < 1).

4.4.2 Coda Manner of Articulation

When participant data was collapsed, there was a main effect of coda manner of articulation on vowel duration (F(2, 563) = 17.6, p < .0005). There was no significant interaction between sentence type and coda manner of articulation (F(2, 563) = .813, p = .444). Individual participant analyses revealed that for all 4 participants, nasal codas produced the longest vowels, followed by stops, and finally fricatives. This effect was significant for participant 1 (nasal = 100 ms, stop = 94 ms, fricative = 77 ms; (F(1, 136) = 4.08, p = .019)), participant 2 (nasal = 119 ms, stop = 98 ms, fricative = 89 ms; (F(1, 137) = 7.23, p = .001)), and participant 4 (nasal = 130 ms, stop = 113 ms, fricative = 100 ms; (F(1, 136) = 6.83, p = .001)), but not participant 3 (nasal = 108 ms, stop = 106 ms, fricative = 104 ms; (F(2, 136) = 2.14, p = .122)). This finding supports de Lacy (1998) in which the magnitude of coda effect on vowel duration was described as {[t], [ʧ]} < [s] < {[d], [ʤ]} < {[n], [l]} < [z]. In the study from de Lacy (1998), nasal codas produced longer vowel durations than [+voice] stops, then [–voice] fricatives, then [–voice] stops. In our stimuli, all fricative codas were [–voice], while stop codas were divided half [+voice] and half [-voice]. We found no significant interaction between sentence type and manner of articulation for any participant (participant 1 (F(2, 136) = .096, p = .909),


participant 2 (F(2, 137) = 1.166, p = .315), participant 3 (F(2, 136) = .058, p = .943), participant 4 (F(2, 136) = .207, p = .813)).

4.4.3 Coda Continuancy

When participant data was collapsed, there was a main effect of coda continuancy on vowel duration (F(1, 565) = 26.77, p < .0005). There was no significant interaction between sentence type and coda continuancy (F(1, 565) = 1.07, p = .301). Individual participant analyses revealed that for all 4 participants, verb stems that had continuant codas had significantly shorter mean vowel duration (participant 1 continuant = 77 ms, noncontinuant = 97 ms, F(1, 138) = 7.49, p = .007; participant 2 continuant = 84 ms, noncontinuant = 100 ms, F(1, 139) = 7.43 p = .007; participant 3 continuant = 94 ms, noncontinuant = 107 ms, F(1, 138) = 4.23, p = .042; participant 4 continuant = 100 ms, noncontinuant = 121 ms, F(1, 138) = 9.80, p = .002. It is unsurprising that continuant codas resulted in the shortest adjacent vowels, as all such codas in our stimuli were also unvoiced. Conversely, the verbs with the longest vowel durations had a voiced coda, which includes nasals and half of the stops (all of which are noncontinuant). We found no significant interaction between sentence type and continuancy for any participant (participant 1 (F(1, 138) = .112, p = .739), participant 2 (F(1, 139) = 1.01, p = .316), participant 3 (F(1, 138) = .104, p = .748), participant 4 (F(1, 138) = .405, p = .526)).

4.5 Sentence Type Effects on Vowel Duration

In previous analyses, only vowels were consistently and significantly longer in passive verb stems versus progressive verb stems. If monosyllabic lengthening is the only factor affecting vowel duration, we would expect to find a significant difference between the vowels in progressive active stems and the vowels in the three –ed sentence types (passive, past active, perfective active), but no vowel duration differences among the three -ed sentence types. To determine if this was the case, we collapsed the data from all 4 participants into a single ANOVA in which vowel duration was the dependent variable, sentence type was the fixed factor, and participant and verb were random factors.

Figure 13. Phonological effects theorized to be present in verb stems of varying sentence types.

Effect Passive Past active

Perfective active

Progressive active

Phrase-Final Lengthening

Yes No No No

Monosyllabic Lengthening

Yes Yes Yes No


This ANOVA revealed a main effect of sentence type (passive = 123 ms, past active = 117 ms, perfective active = 114 ms, progressive active = 93 ms; (F(3, 19) = 18.85, p < .0005)). Planned comparisons with Bonferroni corrections revealed significant differences between passive sentences and all other types, as well as between progressive active sentences and all other types (See Appendix E and Figure 14). However, there was no significant difference in vowel duration between the past active and perfective active sentence types (p = .633). The greater vowel length in passives than past active or perfective active sentences could reflect phrase-final lengthening in passive sentences. If the syntactic phrase boundary in the passive sentences between V and PP corresponds with a prosodic phrase boundary, we would expect the verb stem vowel in passive sentences to be longer than the vowel in the other two -ed sentence types, in which there is no such syntactic (and presumably no prosodic) phrase boundary. This would result in the vowel duration ranking we see in the data: passive > past active = perfective active > progressive active.

Figure 14. Post hoc analyses revealed a significant difference between progressive active and all other types and passive and all other types, but no difference between perfective active and past active. **** denotes significance value p < .0005. ** denotes significance value p < .05. Error bars denote standard errors.

The same ANOVA revealed a significant interaction between sentence type and subject (F(9, 154) = 2.62, p = .008), which suggests that participants behaved differently from one another for verb stem vowel duration. As depicted in Figure 15, participants 1, 2, and 3 all follow the same pattern with passive verb stems having the longest vowel durations, followed by the simple and perfective active verb stems, and finally the

60 70 80 90

100 110 120 130 140 150 160

Passive Past active Perfective active Progressive active

Mea

n Vo

wel

Dur

atio

n (m

s)

Subject

Vowel Duration by Sentence Type ****

**** ****

**** **

ns


progressive active verb stems having the shortest vowel durations. Participant 4, however, is unlike the others. This participant’s past active sentences have longer verb stem vowel durations than those of his passive sentences. Explanations that might account for this result will be discussed in section 5. These include fatigue, selective verb focus, or failure to have actively processed the production sentences.

Figure 15. Notice participant 4 is the only case where the passive vowel duration is not the longest, rather it is the past active vowel duration. Error bars denote standard errors.

5. Discussion

As in Stromswold et al. (2003, under review) and Rehrig et al. (2015), we were able to find monosyllabic lengthening of passive verb stems (or polysyllabic shortening of progressive verb stems). Both previous studies compared only progressive active and passive sentence types, while we tested these as well as past active and perfective active sentence types. Moreover, our stimuli contained only monosyllabic verb stems with no consonant clusters and -ed participles which were exclusively coronal stops, [t] and [d]. In two of our participants, we found that the progressive active -ing sentence type yielded verb stem durations which were significantly shorter than those within all three -ed sentence types. For another participant, the progressive active verb stems were

70

80

90

100

110

120

130

140

150

160

1 2 3 4

Mea

n Vo

wel

Dur

atio

n (m

s)

Subject


Passive

Past active

Perfective active

Progressive active


significantly shorter than both passive and perfective active ones, and for one other participant the progressive active verb stems were significantly shorter than passive stems only.

This study went one step further and determined that it was the vowel constituent of each verb stem was being consistently and significantly shortened when an additional syllable was added to the rightmost edge. To discover this, we compared verb stems between the passive and progressive active sentence types only, as these were found to have the greatest and most consistent disparity in total stem duration for all participants in the previous analysis. For onsets, we were only able to find a significant difference across sentence types in one participant, but this disparity was small – only about 11 ms. The data for codas also yielded unencouraging results. Only two participants significantly lengthened their codas across sentence types, but they did so in opposite directions. Participant 2 unexpectedly produced codas which were longer in the progressive sentences. The only viable candidate for the locus of the duration disparity was the vowel. On average, the vowel constituent in passive stems was approximately 30 ms longer than in progressive active stems; this equates to a duration change of 35.71% for participant 1, 34.48% for participant 2, 29.47% for participant 3, and 31.13% for participant 4. In a category judgment experiment, Klatt and Cooper (1975) found that the just-noticeable difference (JND) for a change in the duration of a single segment is 25 ms or greater. Furthermore, the JND of duration is known to follow Weber’s Law, so it is more accurate to say that humans can perceive a duration change of 20% (Klatt, 1976). The participants might have used this vowel difference as a perceptual cue for disambiguating active and passive sentences in Stromswold et al. (2003, under review).

If monosyllabic lengthening alone was having an effect on vowel duration within the verb stem, then we would expect all three -ed sentence types to yield approximately the same duration as one another, and longer than the progressive active. When we collapsed all participants together, we found that the vowel in progressive active types was indeed significantly shorter than in all other types. However, contrary to the theory we also found that the vowel in passive sentences was significantly longer than all other sentence types. In addition, there was no significant difference between the past active and perfective active types. Clearly there must be another effect of sentence type present to account for this trend. It is possible that relative frequency of these sentence structures could contribute to this pattern. Aylett and Turk (2004) found that speakers of English tend to maintain a constant average predictability when producing language. In English, passive sentence structures are comparatively infrequent and as a result, speakers might have spent more time on them than the more predictable progressive, past, and perfective active types.

Another explanation appeals to the presence of phrase-final lengthening as a result of inherently different syntactic structures between passive sentences and the three


active types. It is believed that verbs in active sentences are present in the middle of phrases, while verbs in passive sentences exist in phrase-final position. With the exception of rush, all verbs in our stimuli are ditransitive and require two arguments (a subject, or agent, and an object) to be grammatical in active sentences, e.g. *The boy was kicking vs. The boy was kicking a ball. In passive sentences, however, an agent is not required, e.g. The ball was kicked is just as grammatical as The ball was kicked by the boy. Such data is used as evidence for the existence of a major syntactic boundary following the verb in passive sentences, but not in active sentences. Furthermore, according to Kreiman (1982) and Lehiste & Wang (1977), lexically-headed syntactic phrase boundaries will often align with prosodic phrase boundaries. As a result we would expect the presence of phrase-final lengthening within the passive verb stem only. Phrase-final lengthening could account for the generally longer vowel duration in passive stems, as well as the lack of significant duration difference between the past active and perfective active vowels.

Participant 4 was an outlier among our subjects. He was the only participant who did not follow the vowel duration trend of passive > {simple active = perfective active} > progressive active. This participant’s past active vowel duration was found to be far longer than those in passive sentences (see Figure 15). One theoretical explanation for this result is that the past active sentences were the only type that lacked an auxiliary (was or has) within the VP. Because speakers generally try to adjust the duration of segments to make their productions approximately equal in length (Aylett and Turk, 2004), we might expect speakers to lengthen past active stems to compensate for not having an auxiliary. However none of the other participants’ data accord with this theory. In fact, their past active vowel durations are generally shorter than those of the perfective active sentence and well as the passive sentence type, both of which contain auxiliaries.

There are several reasons to regard this participant’s data as perhaps unreliable. Participant 4 performed somewhat worse on the catch trials than all other participants tested with only an 89% correct response rate. Conversely, participants 1 and 3 both had a 100% correct rate and participant 2 had a 97% correct rate. This suggests that participant 4 may not have been processing the sentences he was producing sufficiently. Another reason to doubt the validity of the data is that participant 4 spoke far slower than any other participant. His mean verb stem duration was 337 ms, as opposed to the other participants’ mean durations of 288 ms, 264 ms, and 281 ms, respectively. This might have been due to fatigue or presence of verb stem focus. Participant 4 frequently yawned during the study, which suggests he was tired while producing the sentences. If focus was the cause of the longer mean duration, it implies he was hyper-articulating some verbs. We reviewed this participant’s audio recordings and did find evidence of focus for a number of sentences. The diagnostic for identifying focus included the existence of a closure preceding stop onsets, a greater relative voice onset time in onsets, and a smooth


but major drop in pitch (F0) across the vowel. Conversely, non-focused vowels will begin with a lower pitch contour that rises across the vowel. See Appendix F for an example found in participant 4.

The consistency and significance of the lengthening effect in our results suggest that two factors influence the greater verb stem duration in passive sentences compared to progressive active sentences found in Stromswold et al. (2003, under review) and Rehrig et al. (2015): monosyllabic lengthening and phrase-final lengthening (although it is still unclear whether this is syntactic or prosodic). If all speakers lengthen vowels to the same degree across sentence types as our data indicates, this could have an effect on the listener as a cue to syntactic structure. More generally, monosyllabic and phrase-final lengthening may serve a demarcative function. They could potentially aid listeners in parsing a stream of syllables into individual words.

Further research into the scope and quantification of monosyllabic and phrase-final lengthening could improve the accuracy of current models of online language processing and production. We may wish to proceed by determining whether speakers of different languages exhibit the same effect. Stems are known to be longer in monosyllabic words for Germanic languages (Beckman & Edwards, 1990), but it is unclear if this finding holds true for other language families. Furthermore, we might improve the current study in three ways. First by having a coder familiar with Praat, but naïve to the purpose of the experiment, re-slice the data. Unconscious biases on the part of the single coder who was familiar with the experiment design and purpose might have contributed to the results. Next we would like to design stimuli in a way that would allow us to correct for speech rate between participants. Finally, it would be interesting to determine if the effects we observed change or hold for spontaneous speech, as opposed to the stimuli-reading speech we analyzed in our experiment. This might be done through corpus analysis or by eliciting active and passive sentences naturally in another experiment.


References

Aylett, M., & Turk, A. (2004). The smooth signal redundancy hypothesis: A functional explanation for relationships between redundancy, prosodic prominence, and duration in spontaneous speech. Language and Speech, 47(1), 31-56.

Barrow, I.M., Givens, G., Stuart, A., Kalinowski, J., & Rastatter, M. P. (2005). Influence of duration and location of pauses on comprehension of a temporarily ambiguous utterance in adults. Perceptual and Motor Skills, 100(1), 142-152. Beach, C. M., Katz, W.F., & Skowronski, A. (1996). Children's processing of prosodic cues for phrasal interpretation. Journal of the Acoustical Society of America, 99(2), 1148-1160. Beckman, M. E. and Edwards, J. (1990). Lengthenings and shortenings and the nature of

prosodic constituency. In Kingston, J. and Beckman, M. E., editors, Papers in laboratory phonology: Between the grammar and the physics of speech, pages 152–178. Cambridge: Cambridge University Press.

Beier, E., Rehrig, G., Stromswold, K. (2015). Acoustic Cues for Active and Passive Strucure Vary for Different Verbs. EPA Poster. Birch, S., & Clifton, C. (2002). Effects of varying focus and accenting of adjuncts on the comprehension of utterances. Journal of Memory and Language,47, 571-588. Carlson, K., Clifton Jr., C., & Frazier, L. (2001). Prosodic boundaries in adjunct attachment. Journal of Memory and Language, 45(1), 58-81. Dahan, D., Tanenhaus, M. K., & Chambers, C. G. (2002). Accent and reference resolution in spoken-language comprehension. Journal of Memory and Language, 47, 292-314. De Lacy (1998). The effect of consonant clusters on vowel duration in English, ms. University of Massachusetts, Amherst. Frazier, Lyn & Fodor, Janet Dean (1978). The sausage machine: A new two-stage parsing model. Cognition, 6 (4):291-325.


Gahl, S., & Garnsey, S. M. (2004). Knowledge of grammar, knowledge of usage: Syntactic probabilities affect pronunciation variation. Language, 748-775. Hooper, J. B. (1976). Word frequency in lexical diffusion and the source of morphophonological change. Current progress in historical linguistics (pp. 96- 105). Amsterdam: North-Holland Pub. Co. Jaeger, T. (2010). Redundancy and reduction: Speakers manage syntactic information density. Cognitive Psychology, 61(1), 23-62. Jurafsky, D., Bell, A., Gregory, M., & Raymond, W. D. Probabilistic relations between words: Evidence from reduction in lexical production. Katz, W. F., Jenouri, K., Verma, S., & Beach, C.M. (1996). Duration and fundamental frequency correlates of phrase boundaries in productions by children and adults. Journal of the Acoustical Society of America, 99(5), 3179-3191. Kjelgaard, M.M., & Speer, S. R. (1999). Prosodic facilitation and interference in the resolution of temporary syntactic closure ambiguity. Journal of Memory and Language, 40(2), 153- 194. Klatt, D. H., & Cooper, W. E. (1975). Perception of segment duration in sentence contexts. In Structure and process in speech perception (pp. 69-89). Springer Berlin Heidelberg. Klatt, D. H. (1976). Linguistic uses of segmental duration in English: Acoustic and perceptual evidence. The Journal of the Acoustical Society of America, 59(5), 1208-1221. Kraljic, T., & Brennan, S. E. (2005). Prosodic disambiguation of syntactic structure: For the speaker or for the addressee? Elsevier Inc. Kreiman, J. (1982). Perception of sentence and paragraph boundaries in natural conversation. Journal of Phonetics, 10(2), 163-175. Lehiste, I. (1972). The timing of linguistic utterances and linguistic boundaries. J. acoust. Soc. Am. 51: 2018-2024.


Lehiste, I. (1973). Phonetic disambiguation of syntactic ambiguity. Journal of the Acoustical Society of America, 53(1), 380. Lehiste, I., Olive, J. P., & Streeter, L. A. (1976). Role of duration in disambiguating syntactically ambiguous sentences. Journal of the Acoustical Society of America, 60(5), 1199. Lehiste, I., & Wang, W. S. (1977). Perception of sentence boundaries with and without semantic information. In W. U. Dressler, & O. E. Pfeidder (Eds.), (pp. 277-283). Innsbruck: Inst. fur Sprachwissenschaft der Univ. Innsbruck. Levin, B., & Malka Rappaport. (1986). The Formation of Adjectival Passives. Linguistic Inquiry, 17(4), 623–661. Jaeger, T. F., & Levy, R. P. (2007). Speakers optimize information density through syntactic reduction. In Advances in neural information processing systems (pp. 849-856). Peterson, G., & Lehiste, I. (1960) Duration of syllable nuclei in English. Journal of the Acoustical Society of America, 32(6), 693-703. Rehrig, G., Beier, E., Chalmers, E., Schrum, N., Stromswold, K. (2015). Acoustic correlates of future syntactic structure in active and passive sentences. CUNY poster. Schafer, A. J., Speer, S. R., Warren, P., & White, S. D. (2000). Intonational disambiguation in sentence production and comprehension. Journal of Psycholinguistic Research, 29(2), 169-182. Scott, D. R. (1982). Duration as a cue to the perception of a phrase boundary. The Journal of the Acoustical Society of America, 71(4), 996-1007. Snedeker, J., & Trueswell, J. (2003). Using prosody to avoid ambiguity: Effects of speaker awareness and referential context. Journal of Memory and Language, 48, 103-130. Speer, S. R., Kjelgaard, M. M., & Dobroth, K. M. (1996). The influence of prosodic structure on the resolution of temporary syntactic closure ambiguities. Journal of Psycholinguistic Research, 25(2), 249-271.


Stromswold, K., Kharkwal, G., Sorkin, J.E., & Zola, S. (under review). Tracking the elusive passive: The processing of spoken passives. Turk, A., Hufnagel, S. (2000). Word-boundary-related duration patterns in English. Journal of Phonetics, 28, 397-440. Turk, A., Hufnagel, S. (2007). Multiple targets of phrase-final lengthening in American English words. Journal of Phonetics, 35, 445-472. Turk, A., Nakai, S. & Sugahara, M. Jul 2006 Methods in Empirical Prosody Research. Sudhoff, S., Lenertova, D., Meyer, R., Pappert, S., Augurzky, P., Mleinek, I., Richter, N. & Schliesser, J. (eds.). Mouton de Gruyter, p. 1-28 (Language, Context and Cognition; vol. 3) Turk, A., Sawusch, J. (1997). The domain of accentual lengthening in American English. Journal of Phonetics, 25(1), 25-41 Van Santen, J.P.H. (1992). Contextual effects on vowel duration. Speech Communication, 11(6), 513-546. Wilson, M.D. (1988) The MRC Psycholinguistic Database: Machine Readable Dictionary, Version 2. Behavioural Research Methods, Instruments and Computers, 20(1), 6-11.


Appendix A

Verb List and Thorndike-Lorge Verb Frequencies

Fricatives

Word CCAE_Spoken CCAE_Fiction CCAE_Magazine miss 12058 16559 10528

pass 19027 29604 19136

kiss 1698 12626 2077

cash 591 391 820

rush 2537 8070 3525

hush 70 669 118

Stops

Word CCAE_Spoken CCAE_Fiction CCAE_Magazine pick 17697 31233 18027

pack 2008 5211 6336

wrap 2531 8866 4835

tug 129 3150 533

tag 330 777 1060

hug 1097 4854 1311

Nasals

Word CCAE_Spoken CCAE_Fiction CCAE_Magazine fan 705 1449 1031

pin 614 2399 1102

ban 2837 343 2130

ram 221 545 266

hum 197 2278 506

can 163 224 150


Appendix B

Sentence List

Verb Sentence Type

Coda Type Coda voice

Coda continuancy

Vowel Length

Sentence

pass Passive Fricative - + Long The note was passed around the classroom.

pass Prog. active Fricative - + Long The truck was passing a black limousine.

pass Prog. active Fricative - + Long The guest was passing a bowl of salad.

pass Prog. active Fricative - + Long The driver was passing another red car.

pass Past active Fricative - + Long The super spies passed a secret message.

pass Perf. active Fricative - + Long The taxi has passed another pedestrian.

kiss Passive Fricative - + Short The young girl was kissed under the starry sky.

kiss Prog. active Fricative - + Short The suitor was kissing a potential bride.

kiss Prog. active Fricative - + Short The mother was kissing a newborn child.

kiss Prog. active Fricative - + Short The president was kissing another baby.

kiss Past active Fricative - + Short The brave schoolgirls kissed a couple of boys.

kiss Perf. active Fricative - + Short The daughter has kissed another old aunt.

cash Passive Fricative - + Long The check was cashed around eleven o'clock.

cash Prog. active Fricative - + Long The landlord was cashing a rent check.

cash Prog. active Fricative - + Long The woman was cashing a big paycheck.

cash Prog. active Fricative - + Long The worker was cashing a welfare check.

cash Past active Fricative - + Long The two bankers cashed a check for their new client.

cash Perf. active Fricative - + Long The teenager has cashed a birthday check.

rush Passive Fricative - + Short The mayor was rushed around the busy town.

rush Prog. active Fricative - + Short The train was rushing across the steel bridge.

rush Prog. active Fricative - + Short The taxicab was rushing along the city street.

rush Prog. active Fricative - + Short The soldier was rushing across the battlefield.

rush Past active Fricative - + Short The truck drivers rushed across the highway.

rush Perf. active Fricative - + Short The student has rushed around his campus.

miss Passive Fricative - + Short The deadline was missed about a month ago.

miss Prog. active Fricative - + Short The first grader was missing a front tooth.

miss Prog. active Fricative - + Short The new baby was missing a clean blanket.

miss Prog. active Fricative - + Short The new student was missing a notebook.

miss Past active Fricative - + Short The suspended girls missed a day of school.

miss Perf. active Fricative - + Short The old woman has missed a doctor’s appointment.

hush Passive Fricative - + Short The student was hushed a dozen times today.

hush Prog. active Fricative - + Short The nanny was hushing a noisy baby in the crib.

hush Prog. active Fricative - + Short The teacher was hushing a prankster in her class.


hush Prog. active Fricative - + Short The mother was hushing a disobedient toddler.

hush Past active Fricative - + Short The angry nuns hushed a student at morning prayer.

hush Perf. active Fricative - + Short The politician has hushed a rude heckler at the rally.

pack Passive Stop - - Long The picnic was packed a couple of days ago.

pack Prog. active Stop - - Long The player was packing up his equipment.

pack Prog. active Stop - - Long The traveler was packing a large suitcase.

pack Prog. active Stop - - Long The camper was packing a bag for his hike.

pack Past active Stop - - Long The college seniors packed a snack for the trip.

pack Perf. active Stop - - Long The rock star has packed up his instruments.

pick Passive Stop - - Short The red apple was picked a couple of hours ago.

pick Prog. active Stop - - Short The little girl was picking a pumpkin to carve.

pick Prog. active Stop - - Short The farmer was picking a few ripe pears.

pick Prog. active Stop - - Short The cowboy was picking a new horse to ride.

pick Past active Stop - - Short The careful players picked a new teammate.

pick Perf. active Stop - - Short The new husband has picked a honeymoon spot.

wrap Passive Stop - - Long The package was wrapped a while ago.

wrap Prog. active Stop - - Long The skier was wrapping a scarf around himself.

wrap Prog. active Stop - - Long The soldier was wrapping a bad injury.

wrap Prog. active Stop - - Long The scout was wrapping a rope around the tree.

wrap Past active Stop - - Long The caring sons wrapped a birthday gift for their mom.

wrap Perf. active Stop - - Long The big snake has wrapped around the tree branch.

tug Passive Stop + - Short The girl's hair was tugged a couple of times.

tug Prog. active Stop + - Short The toddler was tugging a nurse's arm.

tug Prog. active Stop + - Short The puppy was tugging a red chew toy.

tug Prog. active Stop + - Short The jogger was tugging a dog by the leash.

tug Past active Stop + - Short The two huskies tugged a large dog sled.

tug Perf. active Stop + - Short The rescuers has tugged a boat to port.

tag Passive Stop + - Long The girl was tagged about a dozen times.

tag Prog. active Stop + - Long The boy was tagging along with his sister.

tag Prog. active Stop + - Long The grocer was tagging a bag of apples.

tag Prog. active Stop + - Long The child was tagging a friend at recess.

tag Past active Stop + - Long The workers tagged another tree to remove.

tag Perf. active Stop + - Long The mailman has tagged another package.

hug Passive Stop + - Short The mother was hugged around the waist.

hug Prog. active Stop + - Short The nurse was hugging a sick patient.

hug Prog. active Stop + - Short The racecar was hugging a tight corner.

hug Prog. active Stop + - Short The baby was hugging a tiny puppy.

hug Past active Stop + - Short The school girls hugged a favorite teacher.


hug Perf. active Stop + - Short The parent has hugged a newborn baby.

fan Passive Nasal + - Long The rich man was fanned aboard his yacht.

fan Prog. active Nasal + - Long The gambler was fanning a deck of cards.

fan Prog. active Nasal + - Long The butler was fanning a houseguest.

fan Prog. active Nasal + - Long The kind maid was fanning a princess.

fan Past active Nasal + - Long The gusty winds fanned across the fields.

fan Perf. active Nasal + - Long The servant has fanned a wealthy lady.

ram Passive Nasal + - Long The castle wall was rammed until it collapsed.

ram Prog. active Nasal + - Long The young knight was ramming a castle gate.

ram Prog. active Nasal + - Long The mountain goat was ramming another goat.

ram Prog. active Nasal + - Long The angry bull was ramming a wood fence.

ram Past active Nasal + - Long The brave warriors rammed a hostile fortress.

ram Perf. active Nasal + - Long The huge truck has rammed a taxi on the highway.

ban Passive Nasal + - Long The new novel was banned across the nation.

ban Prog. active Nasal + - Long The mayor was banning a rally in the city limits.

ban Prog. active Nasal + - Long The governor was banning a movie in his state.

ban Prog. active Nasal + - Long The clerk was banning a customer from the store.

ban Past active Nasal + - Long The strict teachers banned a controversial book.

ban Perf. active Nasal + - Long The pilot has banned a passenger from smoking.

hum Passive Nasal + - Short The song was hummed around town.

hum Prog. active Nasal + - Short The girl was humming a catchy tune.

hum Prog. active Nasal + - Short The actor was humming a theme song.

hum Prog. active Nasal + - Short The mother was humming a lullaby.

hum Past active Nasal + - Short The singers hummed a beautiful melody.

hum Perf. active Nasal + - Short The choir has hummed a church song.

can Passive Nasal + - Long The fruit was canned a day after harvest.

can Prog. active Nasal + - Long The maid was canning a quart of peaches.

can Prog. active Nasal + - Long The wife was canning a large amount of yams.

can Prog. active Nasal + - Long The chef was canning a few tomatoes.

can Past active Nasal + - Long The farmers canned a bunch of potatoes.

can Perf. active Nasal + - Long The cook has canned a pound of mushrooms

pin Passive Nasal + - Short The criminal was pinned against a brick wall.

pin Prog. active Nasal + - Short The teenager was pinning a post to the webpage.

pin Prog. active Nasal + - Short The president was pinning a medal on the soldier.

pin Prog. active Nasal + - Short The new intern was pinning a note on the wall.

pin Past active Nasal + - Short The scoutmasters pinned a badge on the boy.

pin Perf. active Nasal + - Short The grandmother has pinned a skirt for the girl.


Appendix C

Participant (1) Sentence Type

(2) Sentence Type

Mean Diff. in sec. (1-2) Sig.

1 Passive Progressive Active

0.03510574 0.000

Past active 0.017949857 0.269 Perfective Active 0.02637421 0.022

Progressive active

Truncated Passive -0.0351057 0.000 Past active -0.01715588 0.111 Perfective Active -0.00873153 1.000

Past active Truncated Passive -0.01794986 0.269 Progressive Active

0.017155883 0.111

Perfective Active 0.008424356 1.000 Perfective active

Truncated Passive -0.0263742 0.022 Progressive Active

0.008731527 1.000

Past active -0.00842436 1.000 2 Passive Progressive

Active 0.02911546 0.000


Progressive active


Past active Truncated Passive -0.01698802 0.085 Progressive Active

0.012127447 0.191

Perfective Active -0.00966202 0.961 Perfective active


0.02178946 0.001

Past active 0.009662015 0.961 3 Passive Progressive

Active 0.04096546 0.000


Progressive active


Past active Truncated Passive -0.00162763 1.000 Progressive 0.03933783 0.000


Active Perfective Active 0.006193619 1.000

Perfective active


0.03314421 0.000

Past active -0.00619362 1.000 4 Passive Progressive

Active 0.05509058 0.000

Past active -0.01741466 0.461 Perfective Active 0.016410983 0.585

Progressive active


Past active Truncated Passive 0.017414659 0.461 Progressive Active

0.07250524 0.000

Perfective Active 0.03382564 0.004 Perfective active


0.0386796 0.000

Past active -0.00032799 0.004


Appendix D

SPSS plot of Participant 1 (Sentence Type 1 = Passive, Sentence Type 2 = Progressive active, Sentence Type 3 = Past active, Sentence Type 4 = Perfective active)

SPSS plot of Participant 2 (Sentence Type 1 = Passive, Sentence Type 2 = Progressive active)


Appendix E

(1) Sentence Type

(2) Sentence Type

Mean Diff. in sec. (1-2) Sig.

Bonferroni Passive Progressive active 0.02908439 0.000 Past active 0.00506324 0.013 Perfective active 0.00773653 0.000

Progressive active

Passive -0.0290844 0.000 Past active -0.0240212 0.000 Perfective active -0.0213479 0.000

Past active Passive -0.0050632 0.013 Progressive active 0.02402115 0.000 Perfective active 0.002673289 0.633

Perfective active

Passive -0.0077365 0.000 Progressive active 0.02134786 0.000 Past active -0.00267329 0.633


Appendix F

Participant 4 focus diagnostic example with highlighted vowel:

The workers tagged another tree to remove. (focus)

The workers tagged another tree to remove. (no focus)

LOCUS AND CAUSE OF PASSIVE VERB STEM LENGTHENING...

Documents

Transcript of LOCUS AND CAUSE OF PASSIVE VERB STEM LENGTHENING...