Universiy of Torono From he SelecedWorks of Alexei Kocheov 27 Place assimilaion and phoneic grounding: A cross-linguisic percepual sudy Alexei Kocheov, Universiy of Torono Connie K. So, Universiy of Wesern Sydney Available a: hps://works.bepress.com/alexei_kocheov/1/
Phonology 24 (27) 397 432. f 27 Cambridge Universiy Press doi:1.117/s952675771273 Prined in he Unied Kingdom Place assimilaion and phoneic grounding: a cross-linguisic percepual sudy* Alexei Kocheov Universiy of Torono Connie K. So Universiy of Wesern Sydney This paper invesigaes predicions made by he phoneic knowledge hypohesis (Jun 1995, 24, Hayes & Seriade 24) abou he relaion beween percepibiliy of sops and common paerns of major place assimilaion. In wo percepual experimens, simuli wih Russian released and unreleased voiceless sops in clusers were presened for idenificaion of 56 liseners, naive speakers of Russian, Canadian English, Korean and Taiwanese Mandarin. Percenages of correc responses and reacion ime daa were used o deermine scales of percepual salience. Resuls reveal considerable percepual differences beween places of ariculaion, consisen across four language groups. Percepual salience of place of ariculaion was srongly affeced by presence or absence of sop releases. While he salience scale for released sops closely corresponded o cross-linguisic paerns of assimilaion, he scale for unreleased sops did no. The resuls provide parial suppor for he hypohesis, while suggesing a less direc relaion beween scales of phoneic difficuly and phonological markedness. 1 Inroducion One of he fundamenal quesions in undersanding human cogniion is how he discree knowledge of he world symbolic conceps and caegories is relaed o coninuous everyday percepions and acions. This quesion, commonly known as he SYMBOL-GROUNDING PROBLEM (Harnad 199), is relevan o all fields of cogniive science, including linguisics. * We are graeful o Juliee Blevins, Seve Winers, he Phonology ediors, an anonymous associae edior and hree anonymous reviewers for insighful commens and suggesions on earlier versions of he manuscrip. We would also like o hank Sonya Bird, Louis Goldsein, Hyung-Soo Kim, Keren Rice, Yue Wang and audiences of he Canadian Linguisic Associaion meeing and Universiy of Vicoria. This work was suppored by he Social Sciences and Humaniies Research Council of Canada. An earlier repor based on parial resuls of Experimen 1 of he curren paper appeared as Kocheov & So (25). 397
398 Alexei Kocheov and Connie K. So However, he problem has been fel paricularly acuely in generaive phonology, wih is radiionally almos exclusive focus on absrac symbolic represenaions and compuaions (Anderson 1985, Pierrehumber e al. 2, Por & Leary 25). How do phonological symbols, and he cross-linguisic sound paerns hey capure, relae o concree facs of ariculaion and percepion of speech sounds? This quesion has generally been considered o be beyond he scope of phonological heory proper. Some recen work wihin Opimaliy Theory (Prince & Smolensky 1993) has argued for a close connecion beween phonological symbols (i.e. grammaical represenaions and consrains), he cross-linguisic sound paerns hey define (markedness laws), and phoneic facs (ease or difficuly of ariculaion and percepion) (see e.g. many conribuions o Hayes e al. 24). According o Hayes & Seriade (24), facs of ariculaory and percepual difficuly, mediaed by speakers phoneic knowledge, are he ulimae source of phonological consrains and markedness laws: The hypohesis º is ha phonological consrains can be rooed in phoneic knowledge º, he speakers parial undersanding of he physical condiions under which speech is produced and perceived. The source of markedness consrains as componens of grammar is his knowledge. The effec phoneic knowledge has on he ypology of he world s sound sysems sems from he fac ha cerain basic condiions governing speech percepion and producion are necessarily shared by all languages, experienced by all speakers, and implicily known by all. This shared knowledge leads learners o posulae independenly similar consrains. The aciviy of similar consrains is a source of sysemaic similariies among grammars and generaes a srucured phonological ypology (Hayes & Seriade 24: 1). This hypohesis, furher referred o as he PHONETIC KNOWLEDGE-BASED GROUNDING (PKG) approach, is summarised in (1) (from Hayes & Seriade 24: 7). (1) a. Facs of phoneic diappleculy b. Speakers implici knowledge of he facs in (a) c. Grammaical consrains induced from he knowledge in (b) d. Sound paerns reflecing he aciviy of he consrains in (c) Universal facs of phoneic difficuly are inernalised by speakers as phoneic knowledge; he speakers use his knowledge o make symbolic generalisaions i.e. grammaical consrains and heir rankings; sound paerns observed across languages reflec he aciviy of consrain rankings in language-paricular grammars. Noe ha explici reference o phoneic knowledge disinguishes his hypohesis from oher prominen phoneically based approaches, which can be referred o as mispercepion-based grounding (Ohala 1981, 199; cf. Evoluionary Phonology; Blevins 24) and gesural dynamics-based grounding (Ariculaory Phonology; Browman & Goldsein 1989, 1992). The assumpion ha
Place assimilaion and phoneic grounding 399 grammaical consrains may be induced by speakers from acquired phoneic knowledge, raher han a priori provided o hem by Universal Grammar, ses he approach apar from Grounded Phonology (Archangeli & Pulleyblank 1994) and sandard Opimaliy Theory (Prince & Smolensky 1993). As Hayes & Seriade (24: 7) noe, speakers phoneic knowledge and grammaical consrains are no direcly accessible o a suden of phonology or phoneics and hus have o be inferred. Facs of phoneic difficuly and sound paerns, however, can be deermined hrough phoneic experimens and ypological research respecively. The relaive accessibiliy of hese wo componens makes he hypohesis empirically esable. The goal of he curren sudy is o verify predicions made by he PKG approach abou cerain aspecs of a common phonological process place assimilaion in consonan clusers. Specifically, we will examine he relaion beween facs of phoneic difficuly in percepion of syllable-final sops and common sound paerns of place assimilaion. Facs of phoneic difficuly will be invesigaed in cross-linguisic percepual experimens. 2 Place assimilaion in sop clusers Some aspecs of phoneic grounding of major place assimilaion paerns have been previously examined Jun (1995, 24; cf. Kohler 199, Byrd 1992). Jun s cross-linguisic survey of place assimilaion reveals some ineresing characerisics of he process, as shown in (2). (2) Phonological asymmeries in place assimilaion (Jun 1995) a. posiion asymmery preconsonanal (coda)>prevocalic (onse) (>=more likely arge) b. arge place coronal>labial>dorsal asymmery (>=more likely arge) c. rigger place labial/dorsal>coronal asymmery (>=more likely rigger) Regressive place assimilaion is far more common across languages han progressive place assimilaion (he posiion asymmery; (2a)). Coronal consonans appear o be he mos likely arges of assimilaion, followed by labials, and hen by dorsals (he arge place asymmery; (2b)).1 Noncoronals are more likely riggers of place assimilaion han coronals (he rigger place asymmery; (2c)). Oher asymmeries idenified by Jun include arge and rigger manner asymmeries. 1 See also Paradis & Prune (1991) on he special saus of coronals, and Prince & Smolensky (1993), Rice (1996, 1999), de Lacy (22), Hume (23) and Blevins (24: 125 129) on relaive place markedness and various heoreical approaches o i.
4 Alexei Kocheov and Connie K. So An ofen cied example of he asymmeries in (2) is Korean (casual speech) place assimilaion. Place assimilaion in Korean is always regressive. Coronals assimilae o following dorsals and labials (3a); labials assimilae o following dorsals, bu no o coronals (3b), and dorsals assimilae neiher o labials nor o coronals (3c). (3) Korean place assimilaion (Jun 1995: 51 52, de Lacy 22: 35 351) inpu oupu a. /mi+ko/ /ko+palo/ [mikko] [koppalo] believe and sraigh b. /ip+ko/ /ip+a/ [ikko] [ipa] *[ia] wear and wear+senence ender c. /ik+a/ /Cak+pHa/ [ika] [CakpHa] *[ia] *[CappHa] ripe+senence ender desrucion In he PKG approach, cross-linguisically common asymmeries found in paerns of assimilaion indirecly reflec cerain universal facs of phoneic difficuly. Thus, preconsonanal consonans should be more difficul o perceive or o produce (or boh) han prevocalic consonans. Coronals should be more difficul o perceive/produce han labials and dorsals, while labials should be more difficul han dorsals. Sops followed by non-coronals should be more difficul han sops followed by coronals. Jun s (1995, 24) survey of previous acousic sudies (of mainly English consonans) suggess ha such phoneic differences do exis. Jun posis he scales of percepual salience ( acousic srengh ) in (4). (Noe ha percepual salience is he reverse of percepual difficuly.) (4) Scales of percepual salience (Jun 1995: 125 13) a. posiion prevocalic (onse)>preconsonanal (coda) (>=more salien; less likely arge) b. arge place dorsal>labial>coronal (>=more salien; less likely arge) c. rigger place labial/dorsal>coronal (>=more salien; more likely rigger) Firs, prevocalic consonans are percepually more salien han preconsonanal consonans (4a), since he former are cued beer (by sronger CV ransiions and burss) han he laer (by weaker VC ransiions and opionally by burss) (Maléco 1958, Bladon 1986, Ohala 199, among ohers). Second, preconsonanal dorsals are he mos salien, followed by labials and hen coronals. The difference beween coronals and noncoronals is due o weaker coronal VC ransiions (as a consequence of more rapid movemen of he ongue ip; Byrd 1992). The difference beween dorsals and non-dorsals is due o a salien convergence of he second (F2) and hird (F3) formans during he VC ransiion (Sevens 1989). Third, sops followed by coronals are more salien han sops followed by
Place assimilaion and phoneic grounding 41 non-coronals (4c), an apparen effec of greaer masking of C1 cues by following non-coronals (Byrd 1992, Zsiga 1994). Jun (1995) noes ha he scales of percepual salience in (4) refer o unreleased syllable-final sops, a poin o which we will reurn laer in his paper. In many languages, however, syllable-final sops can be audibly released accompanied by a burs wih possible addiional noise (cf. opional coda sop releases in English: Henderson & Repp 1982, Lisker 1999, Zsiga 2). Releases are known o provide addiional informaion abou place of ariculaion (e.g. Maléco 1958, Winiz e al. 1972, Wrigh 24: 38), and herefore released sops are likely o be more percepually salien han unreleased sops. This gives us anoher scale of percepual salience: released sops>unreleased sops. Jun (1995) does no commen on he relaive place percepibiliy of released sops. We will herefore make a preliminary assumpion (o be revised laer) ha he relaive place percepibiliy of released sops is similar o ha for unreleased sops (4b). The PKG approach views he percepibiliy scales in (4) as par of speakers implici phoneic knowledge. Imporanly, his knowledge is shared by speakers of differen languages, gives he assumpion ha basic condiions under which speakers produce and perceive sops are he same. The shared phoneic knowledge of he percepibiliy scales leads speakers of differen languages o posi similar rankings of consrains in languagespecific phonological grammars. Jun (1995, 24) proposes he fixed hierarchies of differen PRESERVE faihfulness consrains in (5). These hierarchies reflec an assumpion ha speakers make use of heir phoneic knowledge: hey make more effor o preserve sounds ha are more percepually salien and relax heir ariculaion of sounds ha are less percepually salien (he PRODUCTION HYPOTHESIS: Jun 1995: 122; cf. Kohler 199, Byrd 1992, Hura e al. 1992, Lindblom e al. 1995). (5) Hierarchies of grammaical consrains (Jun 1995: 136) a. Pres(pl(dor))êPres(pl(lab))êPres(pl(cor)) b. Pres(pl(_V))êPres(pl(_C)) c. Pres(pl(_cor))êPres(pl(_lab/dor)) Differen rankings of he markedness consrain WEAKENING ( conserve ariculaory effor ) agains he hierarchies in (5) are capable of generaing a range of aesed paerns of place assimilaion (Jun 1995, 24). For example, he paern aesed in Korean is capured by he ranking of consrains shown in (6). (6) A fragmen of Korean grammar (Jun 1995: 146) Pres(pl(_V)), Pres(pl(dor)), Pres(pl(_cor))ê Weakeningê Pres(pl(lab)), Pres(pl(cor)), Pres(pl(_lab/dor)), Pres(pl(_C)) Ranking PRES(pl(_V) above PRES (pl(_c) ensures he regressive direcion of place assimilaion. The higher ranking of PRES(pl(dor)) and
42 Alexei Kocheov and Connie K. So PRES(pl(_cor)) ensures ha assimilaion does no affec syllable-final dorsals and consonans followed by coronals (e.g. /kp/, /k/ and /p/), while he lower ranking of he oher place-referring consrains enforces assimilaion in he oher consonan clusers (e.g. /pk/, /p/ and /k/). Jun s phoneically based accoun of place assimilaion undoubedly conribues o a beer undersanding of phoneic grounding of phonological processes. One limiaion of his accoun, however, is ha he universal scales of percepual salience in (4) are mainly deduced from previous acousic sudies, raher han being esablished in percepual experimens (cf. Hume e al. 1999, Wrigh 21, Winers 23).2 I is well known ha liseners aend o cerain acousic properies more han ohers (Bregman 199); herefore, some acousic differences may or may no be relevan percepually. Accordingly, saemens abou he universal salience of posiions and places may no be fully valid, unless verified in cross-linguisic percepual experimens. While a number of percepual sudies have confirmed he posiion percepibiliy scale (4a) (Fujimura e al. 1978, Ohala 199, Wrigh 21, Kingson & Shinya 23), relaively lile percepual work has been done o direcly verify he place salience scales and heir universaliy, as also noed by Hume e al. (1999) and Winers (23). The curren sudy is inended o conribue o he invesigaion of he arge place salience scale (4b), is universal saus and is possible relaion o grammaical consrains and paerns of place assimilaion. This is done by conducing a percepual experimen where Russian syllable-final sops /p k/ are presened o liseners, who are naive speakers of four languages: Russian, Canadian English, Korean and Taiwanese Mandarin. The PKG approach is examined by esing he specific hypoheses abou phoneic differences in sop percepion in (7). (7) a. Place arge salience hypohesis Syllable-final dorsals are more percepually salien han labials and coronals; labials are more salien han coronals (dorsal>labial>coronal). b. Release salience hypohesis Syllable-final released sops are more percepually salien han unreleased sops (released>unreleased). 2 Jun (24) cies one percepual sudy (Winiz e al. 1972) as supporing his place arge salience scale in (4b). The sudy, however, does no invesigae he percepibiliy of unreleased sops. Raher, i examines idenificaion of place in voiceless English sops based in (i) burs only and (ii) burs and ransiions (1 ms of he adjacen vowel). The auhors noe ha final /k/ depends srongly for is percepion on ransiional segmens (1972: 1313), given a considerably beer idenificaion of he segmen in he second condiion. The relaive place percepibiliy of final sops in wo condiions, (i) //>/p/>/k/, (ii) /p/>/k/>// (1972: 1311 1312), however, does no fully agree wih Jun s place salience scale.
Place assimilaion and phoneic grounding 43 c. Universal salience hypohesis The relaive salience of places of ariculaion and releases is languageindependen. Noe ha he universal salience hypohesis emphasises cross-linguisic similariy in relaive salience. Given language-paricular phonological and phoneic differences involving sop place of ariculaion, some absolue percepual differences among languages are no unexpeced (Hume e al. 1999, among ohers). While our primary focus is on he place arge salience scale, he resuls of he experimens will also allow us o evaluae Jun s predicions abou he place rigger scale (labial/dorsal>coronal). Furher, he resuls are expeced o provide evidence for or agains wo alernaive views of assimilaory sound changes. Previous sudies have produced conflicing resuls as o wha ypes of errors in place idenificaion are more common (Hura e al. 1992, Kingson & Shinya 23). Prevalence of non-assimilaory errors has been inerpreed as evidence for he role of phoneic knowledge (Hura e al. 1992), and herefore as evidence for he PKG approach (Seriade 21, Hayes & Seriade 24). Tha is, if percepual errors are predominanly non-assimilaory, hey canno, on heir own, rigger an assimilaory sound change. Therefore, here mus be anoher mechanism ha convers hese mispercepions o an assimilaory sound change. This mechanism is presumably speakers knowledge of relaive percepibiliy, of he ype assumed by he producion hypohesis (Kohler 199, among ohers). In conras, if percepual errors are predominanly assimilaory, an assimilaory sound change can arise as grammaicalisaion of hese errors, wihou reference o speakers phoneic knowledge. This will suppor he innocen misapprehension view of assimilaion (Ohala 1981, 199). Finally, he resuls of our experimens will allow us o deermine wheher he percepibiliy scales deduced from he percepual experimens do correspond o he consrain hierarchies and paerns of place assimilaion discussed in Jun (1995, 24). 3 Experimen 1 The primary goal of Experimen 1 is o es he hypoheses wih respec o place arge salience, release salience and universal salience in (7), and o examine paerns of percepual errors as possibly indicaive of he role of phoneic knowledge in place assimilaion. 3.1 Mehod 3.1.1 Maerials. The experimen was based on Russian speech simuli, carefully conrolled for he lack of caegorical or gradien place assimilaion. The simuli included sequences of he ype [ac1#c2ap], as shown
44 Alexei Kocheov and Connie K. So in (8), excised from nonsense uerances from a Russian corpus colleced using a magneic ariculomeer (EMMA: Perkell e al. 1992). (8) Targe sop labial [ap ap], [ap kap] coronal [a pap], [a kap] dorsal [ak pap], [ak ap] The uerances were produced by wo female speakers of Sandard Russian (one from Moscow and one from Perm). In each uerance, he arge sop C1 (syllable-final /p k/) was followed by a heerorganic sop C2 (syllable-iniial /p k/). Uerances where he arge consonans were followed by homorganic sops [ap pap], [a ap] and [ak kap] were also included. These were used in he ariculaory analysis as conrols for gesural reducion (see below). In he percepual experimen, simuli wih homorganic sequences were used as fillers, o ensure ha he simulus se was balanced, wih all possible combinaions of C1 and C2. In all uerances, he vowel conex was he same, he low cenral vowel /a/. The speakers were insruced o sress boh syllables. All uerances were read five imes in he carrier phrase [ EV_V pjaj] This is_again. Two okens of each uerance (he hird and fourh repeiions) were seleced for he experimen, giving a oal of 24 simuli wih heerorganic sequences (6 simulix2 okensx2 speakers) and 12 simuli wih homorganic sequences (3 simulix2 okensx2 speakers). The EMMA daa was used in he experimen in order o ensure ha all syllable-final oral gesures of he sop consonans were fully produced. A number of sudies have shown ha ariculaory gesures end o reduce spaially in syllable- and word-final posiions (Browman & Goldsein 1995, Krakow 1999). Such reducion can be eiher gradien or caegorical (he laer corresponding o deleion or assimilaion), and here are considerable differences in reducion paerns (Ellis & Hardcasle 22 on English; Son e al. 27 on Korean). Gradien and especially caegorical reducion is known o have considerable percepual consequences (Nolan 1992, Jun 1996, Surprenan & Goldsein 1998, Son e al. 27).3 Therefore, a lack of a proper conrol over gesural reducion in a sudy invesigaing percepual sources of place assimilaion would no be desirable. An ariculaory analysis of he curren simuli showed ha all syllablefinal sops were fully produced hey exhibied neiher gradien nor caegorical reducion in he magniude of he relevan gesures. Some sample okens used in he curren sudy are shown in Fig. 1, which displays he spaial posiions of EMMA receivers indicaing he posiions 3 The use of EMMA daa in percepual research is no uncommon, paricularly in sudies invesigaing he link beween producion and percepion, for example, Surprenan & Goldsein (1998), Ellis & Hardcasle (22), Pouplier & Goldsein (25) and Perkell e al. (26).
verical posiion (mm) verical posiion (mm) verical posiion (mm) 3 15 15 3 3 15 15 3 3 15 [p] in [ap kap] lips [] in [a pap] ongue ip [k] in [ak ap] Place assimilaion and phoneic grounding 45 [p] in [ap pap] ongue ongue dorsum dorsum 15 fron back 3 15 15 3 45 6 15 15 3 45 6 Figure 1 Posiions of EMMA receivers, indicaing relaive posiions of ariculaors during he producion of [p], [] and [k] in heerorganic (lef) and homorganic (righ) sequences (sample okens). lips [] in [a ap] ongue ip [k] in [ak kap] horizonal posiion (mm) of he following ariculaors (from lef o righ): he upper and lower lips, he fron of he lower eeh and he ongue, approximaely from he ip o he dorsum. Posiions of receivers during he producion of sops as C1 in heerorganic sequences (lef) are compared o posiions of he same sops as C1 in homorganic sequences (righ). (The snapshos were aken a he midpoin of he C1 consricion in heerorganic sequences and a 25% of he C1C2 consricion in homorganic sequences.) The relevan ariculaors are circled. Noice ha he lip consricion during [p] in [ap kap] has similar spaial values o ha during [p] in [ap pap] (op row). The ongue-ip consricion during [] in [a pap] is similar o ha during [] in [a ap] (middle row) (he ongue shape is somewha differen). The
46 Alexei Kocheov and Connie K. So (a) C1 release [ a k a] (b) yon yo yon yo ongue ip (c) yon yo yon yo ongue dorsum 65 7 75 8 85 9 95 1 Figure 2 A display showing he movemens of EMMA receivers, indicaing movemens of he ongue-ip and he ongue-dorsum ariculaors during he producion of he heerorganic sequence [k] of [ak ap] (b, c) and he resuling acousic signal (a). The yon and yo labels in (b) and (c) indicae onses and o ses of relevan verical movemens of he wo ariculaors. The arrow indicaes he disance beween he release of ongue-dorsum consricion for [k] and he formaion of he ongue-ip consricion for []. ongue-dorsum consricion during [k] in [ak ap] is similar o ha during [k] in [ak kap] (boom row).4 Overall, he oral ariculaory gesures of C1 and C2 in he simuli exhibied relaively lile overlap. Figure 2 shows a sample oken of [ak ap]. Noice ha he ongue-dorsum consricion for [k] in (c)) is released abou 6 ms prior o he formaion of he ongue-ip consricion for [] (b). The delay in achieving he C2 arge resuls in a fairly long audible release of C1 prior o he closure of C2 in he acousic signal (a). An acousic analysis of he simuli showed ha audible C1 releases as in Fig. 2 were found in mos of he heerorganic okens used in he experimen (as deermined on he basis of waveforms and specrograms). There were four okens wihou clear releases (wo okens of [ap ap] and one each of [ap kap] and [a pap]). In he homorganic sequences, C1 sops were always unreleased.5 Releases of [] and [k] were longer in duraion han releases of [p] (on average 36 ms and 27 ms for [] and [k] vs. 18ms for [p]). Labial releases were also lower in ampliude han coronal and 4 Spaial values for he simuli were wihin wo sandard deviaions (on average 1 mm) from he respecive conrol values, which were based on means for five okens of each homorganic sequence. 5 Similar findings for sop releases in Russian are repored by Zsiga (2) in an acousic sudy of word-boundary sop sop sequences.
Place assimilaion and phoneic grounding 47 dorsal releases. Overall, [], [p] and [k]-iniial sequences had approximaely he same closure duraion: 156 ms, 16 ms and 158 ms respecively (measured from he offse of he preceding vowel o he release of C2). F2 ransiions owards he [p] closure involved lowering (o 122 Hz, averaged for boh speakers), while hey sayed level or rose slighly before he [] and [k] closures (o 1516 Hz and 1562 Hz respecively). F3 values prior o he closure of [k] were lower han hose prior o he closures of [p] and [] (272 Hz vs. 2938 Hz and 329 Hz). Noe ha hese differences in VC ransiions are overall similar o hose repored for English sops by Sevens (1998).6 The simuli were used in wo condiions: complee and edied. Simuli for he complee condiion were he same as described above. In hese simuli, he place of C1 was cued by boh VC ransiions and he release of C1 (if presen).7 Simuli for he edied condiion were creaed from he complee simuli by removing a C1 release (if presen) from he acousic signal by seing he ampliude during he C1C2 closure o zero (cf. Wrigh 21, Winers 23). The beginning and end poins of he edied inerval were made a audio waveform zero-crossings in order o avoid inroducing any ransiens. The inensiy levels of sound files were adjused in order o mainain he same average inensiy (a he midpoin of he preceding vowel). This procedure lef he edied simuli wih VC ransiions as he only cue o he place of C1, eliminaing he oher poenial sources of place informaion sop releases, and any noise occurring during sop closures. In boh complee and edied files, 5 ms of silence were appended before and afer he signal. The inclusion of he edied se of simuli was necessary o es he ne conribuion of VC ransiions o place salience. Recall ha he originally proposed place salience scale was based on he acousic properies of vowel ransiions o unreleased sops (Jun 1995). The analysis of percepual resuls was based on he heerorganic simuli, which were mos relevan o an invesigaion of assimilaion. However, means obained from he homorganic simuli were used o provide a comparison wih hose from he heerorganic condiion. 3.1.2 Paricipans. The paricipans in Experimen 1 were 56 undergraduae sudens a Simon Fraser Universiy (Vancouver), who were naive speakers of Russian, Canadian English, Korean or Taiwanese Mandarin (14 liseners per language group). The Russian liseners (8 females, 6 males) were born in Russia or he former Sovie Union, and had arrived in Canada a an average age of 18. The Canadian English liseners 6 Based on figures of VC ransiions o sops in he conex of [A] produced by a male English speaker (Sevens 1998: 343, 357, 367), F2 values descend o approximaely 1 Hz before [b], and ascend o approximaely 155 Hz and 14 Hz before [d] and [g] respecively. F3 values are lower before [g] (approximaely 245 Hz) han before [b] and [d] (approximaely 26 Hz for boh). 7 Recall ha releases were absen in four heerorganic okens and all homorganic okens.
48 Alexei Kocheov and Connie K. So (9 females, 5 males) were born in Briish Columbia, Canada. The Korean liseners (7 females, 7 males) were born in Souh Korea (mosly in Seoul), and had arrived in Canada a an average age of 16. The Taiwanese Mandarin liseners (12 females, 2 males) were born in Taiwan (mosly in Taipei), and had arrived in Canada a an average age of 13. The average age of he paricipans was 23. All non-english naive speakers were English bilinguals. None of he liseners repored hearing problems; all were naive as o he purpose of he experimen. The reason for employing liseners/naive speakers of hese languages was o provide some conrol for differences in language-paricular phonological or phoneic knowledge. Russian is a non-assimilaing language, while English is known o exhibi coronal place assimilaion, which is mainly gradien (Browman & Goldsein 1992, Nolan 1992, Ellis & Hardcasle 22). Korean is a language exhibiing coronal and labial place assimilaion (Kim-Renuad 1991). (Boh English and Korean place assimilaion processes end o occur in casual speech.) In Sandard Mandarin, on he oher hand, he place conras in sops is absen syllablefinally, since sops are no permied in codas in general (Norman 1988, Duanmu 22). Assuming Jun s (1996, 24) accoun of place assimilaion, he grammars of he four languages differ in he relaive ranking of he WEAKENING consrain agains he fixed hierarchy of PRESERVE consrains referring o place feaures, as shown in (9).8 (9) Assumed rankings of he Weakening consrain in he grammars of Mandarin, Korean, English and Russian (sops) ê Pres(pl(dor)) ê Pres(pl(lab)) ê Pres(pl(cor)) ê Weakening in Mandarin Weakening in Korean Weakening in English Weakening in Russian Some inerlanguage differences in he phoneic implemenaion of syllable-final sops are also worh noing. While sops in (heerorganic) sequences in Russian are usually released, hey are opionally released in English (Henderson & Repp 1982, Lisker 1999, Zsiga 2) and are apparenly hardly ever released in Korean (Kim-Renaud 1991). Overall, significan similariies beween he hree groups in heir percepual performance can be aribued o facors independen of liseners naive language backgrounds, such as general facs of percepual difficuly (and possibly English bilingualism, a facor discussed below). Differences 8 The rankings for English and Korean are characerisic of he casual speech regiser. For all languages, he consrain PRES(pl(_V)) is ranked above he oher faihfulness consrains in (9), while PRES(pl(_C)) is ranked below hem. See Jun (24) on a possible OT accoun of gradien place assimilaion, such as coronal place assimilaion in English. I should be noed ha he ranking for Mandarin may no fully reflec he knowledge of some of our Mandarin liseners, who were familiar wih Taiwanese (a Min dialec of Chinese) (as discussed in w3.2.1).
Place assimilaion and phoneic grounding 49 beween he four groups may be explained by liseners languageparicular phonological/phoneic knowledge. 3.1.3 Daa collecion and analysis. The experimen used a forced-choice phoneme-idenificaion ask. The simuli were presened individually; he paricipans were insruced o lisen o speech okens and o idenify he las consonan of he firs syllable (e.g. [ap ap]) by pressing buons labelled p, and k on a response box. The paricipans were asked o respond as quickly as possible and o guess when hey were in doub. Insrucions were given in English. All simuli were presened in random order wih a four-second inersimulus inerval. Each oken was presened wice. Trials were blocked by condiion ( complee and edied ); hese blocks were presened in alernaing order in a single session. A oal of 5376 responses o simuli wih heerorganic sequences were colleced (6 uerancesx2 speakersx2 okensx2 repeiionsx2 condiionsx56 liseners). Daa were averaged across 8 responses o he same simulus ype (2 speakersx2 okensx2 repeiions) for each lisener.9 Two measures were used in he analysis of daa: percenage of correc responses (correc idenificaion responses divided by all responses) and he reacion ime for correc responses (RT), which was measured from he onse of he sound file. Higher percenages of correc and faser RT values were aken o indicae greaer percepual salience of place of ariculaion. In addiion o percen correc and RT daa, he ypes of liseners errors (incorrec responses) were examined. In paricular, i was of ineres o deermine wheher he proporions of assimilaory errors (C1 perceived as having he same place as C2) were similar o or higher han he proporions of non-assimilaory errors (C1 perceived as having a differen place from C2). 3.2 Resuls and discussion The resuls of he experimen are summarised in Table I as a confusion marix of all idenificaion responses of C1, combined for four language groups (a oal of 56 paricipans). The resuls are shown by he condiion ( complee and edied ), he arge consonan (C1: /p/, // or /k/) and he conex (C2: /p/, // or /k/). Mean percenages of correc and RT values are provided in he righmos columns. Given he subsanial qualiaive and quaniaive differences beween he wo condiions, we will examine he percenage of correc and RT resuls separaely for he complee and he edied sops (ww3.2.1 and 3.2.2). In addiion, we will examine ypes of errors in boh condiions (w3.2.3). 9 Due o a programming error, fewer okens of he released condiion simulus [ap ap] were presened. This resuled in 35 responses for his paricular simulus, compared o 45 responses for he oher simuli.
41 Alexei Kocheov and Connie K. So responses condiion arge (C1) conex (C2) p k no % reacion ime response correc (correc) p k 268 331 º57 º38 º23 77 2 77 1 73 88 1347 1332 complee p k 124 39 29 313 º34 93 3 64 73 69 87 1399 1348 k p º21 4 º26 53 4 391 1 89 29 87 28 1184 1233 means 77 1 137 p k 315 284 º91 7 º4 92 2 2 7 31 63 39 1477 138 edied p k 162 114 238 215 º46 116 2 3 53 13 47 99 1529 1518 k p 116 12 124 174 27 153 1 2 46 21 34 15 1433 1586 means 56 3 1485 Table I A confusion marix of all responses (4 language groups combined). Mean percen correc (%) and reacion ime (ms) values are given separaely by condiion, arge consonan (C1) and conex (C2). Cells wih correc responses are shaded. 3.2.1 Percen correc and reacion ime: he complee condiion. Two individual mixed-design ANOVAs were carried ou o examine he effecs of C1 Place (/p k/) and Language Group (Russian, Canadian English, Korean, Taiwanese Mandarin) on liseners mean percen correc and RT values respecively. Wih respec o percen correc, here were significan effecs of C1 Place and Language Group [F(2,14)=3. 41, p<. 1 and F(3,52)=5. 434, p<. 1 respecively]. The ineracion of C1 Place and Language Group was no significan. Pos hoc LSD ess for C1 Place revealed ha /k/ was significanly differen from /p/ and // (p<. 1 for boh) and ha /p/ was significanly differen from // (p<. 5). Pos hoc LSD ess for Language Group indicaed ha Russian liseners showed resuls significanly differen from Korean and Mandarin liseners (p<. 1 for boh). Similarly, English liseners performed differenly from Korean liseners (p<. 1). The difference beween English and Mandarin liseners was marginally significan (p=. 61). Wih respec o RT (for correc responses), here was a significan effec of C1
(a) 1 Place assimilaion and phoneic grounding 411 8 6 4 2 (b) 18 16 14 12 1 reacion ime (ms) mean % correc p k 8 Russian English Korean Mandarin Figure 3 (a) Mean percen correc and (b) mean reacion ime values by language group for complee syllable-final /p k/ (C1). Place [F(2,14)=7. 521, p<. 1]. Pos hoc LSD ess for C1 Place revealed ha /k/ was differen from /p/ and // (p<. 1 for boh). The effec of Language Group was marginally significan [F(3,52)=2. 526, p=. 68].1 The ineracion of he wo facors was no significan. Figure 3 presens mean percen correc (a) and RT (b) resuls for he complee simuli according o place of ariculaion of C1 (/p k/) and he language group (Russian, Canadian English, Korean, Taiwanese Mandarin). We can see ha he /k/ has he highes mean percenage of correc values, followed by /p/, and hen //. This paern is quie consisen 1 A re-examinaion of individual RT values for English liseners revealed ha hree ou of foureen liseners showed relaively slow RT values for // (17 21 ms, compared o RTs lower han 14 ms for he oher English liseners). Excluding he RT daa for hese hree liseners from an ANOVA analysis produced significan effecs for boh Place and Language Group [F(2,98)=5. 332, p<. 1 and F(3,49)=2. 877, p<. 5 respecively]. Pos hoc LSD ess for Place revealed ha /k/ was differen from /p/ and // (p<. 1 for boh). Pos hoc LSD ess for Language Group revealed ha he RT for Russian liseners was significanly differen from he RT for Korean and Mandarin liseners (p{. 5 for boh). Similarly, he RT for English liseners was significanly differen from ha for Korean liseners. The difference beween English and Mandarin lisener groups was marginally significan (p=. 76). The ineracion of Place and Language Group was no significan.
412 Alexei Kocheov and Connie K. So across he language groups. Russian and English liseners show higher mean percen correc values han Korean and Mandarin liseners. In erms of RT, /k/ shows shorer mean RT han he oher sops. Russian and English liseners show a endency for shorer RT han Korean and Mandarin liseners. The idenificaion of non-coronals was no necessarily beer or worse when hey were followed by a coronal. An examinaion of Table I reveals ha he idenificaion of /p/ was only 3% beer before // han before /k/, and ha of /k/ was 2% worse before // han before /p/. RT differences were also minor, bu showed somewha longer values when C2 was a coronal han a non-coronal. Overall, he resuls fully suppor he place arge salience hypohesis in (7a): /k/ is more percepually salien han /p/ and // (based on boh percen correc and RT); /p/ is more salien han // (based on percen correc). The resuls also largely suppor he universal salience hypohesis in (7c): relaive salience of places of ariculaion was he same for all four language groups (based on percen correc and in par on RT). The differences beween he language groups in he magniude of salience were in he expeced direcion: liseners whose languages exhibi no or limied place assimilaion (Russian and English) performed beer han liseners whose languages exhibi more exensive assimilaion (Korean) or a lack of place conras in coda (Mandarin). However, his difference in performance can be also aribued o language-paricular phoneic knowledge. In he curren condiion, C1 sops were predominanly released (see w3.1.1). Russian and English liseners are accusomed o released syllable-final sops in heir naive languages (alhough o differing exens), while Korean and Mandarin liseners are no used o such sounds in heir naive languages. The familiariy of he Korean and Mandarin paricipans wih English released sops may have been a facor, ye apparenly no a srong enough one o override heir naive percepual biases. The lack of significan differences beween he Korean and Mandarin groups is less expeced, given he phonoacic differences beween he wo languages (see w3.1.2; cf. Flege & Wang 1989). The lack of he effec can be aribued o wo facors. Firs, Mandarin liseners could have had greaer exposure o English han Korean liseners ( judging by heir average age of arrival in Canada). Second, as menioned earlier, our Mandarin liseners were originally from Taiwan, and some of hem were familiar wih Taiwanese (a Min dialec of Chinese), which allows (unreleased) sops in word-final posiion (Norman 1988). Finally, he resuls did no confirm he rigger place scale in (4c). However, he lack of his effec can be due o he relaively small overlap in he Russian sop clusers in he simuli (see w3.1.1). The subsanial masking effec of following non-coronals was found for English, where sop gesures show considerable overlap (Zsiga 1994, Byrd 1996). 3.2.2 Percen correc and reacion ime: he edied condiion. Two mixed-design ANOVAs were carried ou o examine he effecs of C1 Place
Place assimilaion and phoneic grounding 413 and Language Group on liseners mean percen correc and RT respecively. Wih respec o percen correc, only he effec of C1 Place was significan [F(2,14)=27. 493, p<. 1]. Pos hoc LSD ess furher indicaed ha /p/ was significanly differen from // and /k/ (p<. 1 for boh); // was significanly differen from /k/ (p<. 5). Wih respec o RT, he C1 Place effec was significan [F(2,98)=3. 585, p<. 5].11 The effec of Language Group and he ineracion of C1 Place and Language Group were no significan. Pos hoc LSD ess revealed ha /p/ was significanly differen from // and /k/ (p<. 5 for boh). Figure 4 presens percen correc (a) and RT (b) resuls for he edied sops according o place and language group. The labial /p/ has he highes mean percen correc values, followed by // and hen /k/. This paern is consisen for all four language groups. The language groups show similar overall mean percen correc values. Noe ha for Mandarin liseners, he mean percen correc value for /k/ is below chance level (32. 1% compared o chance level 33. 3%). In erms of RT, /p/ was idenified faser han // and /k/. Again, overall mean RT values are similar across he groups. Consonans followed by coronals were no idenified any beer han consonans followed by non-coronals. Thus correc idenificaion of /p/ and /k/ was on average 52% when followed by // and 55% when followed by a non-coronal (/k/ and /p/ respecively) (see Table I). RT was slower before coronals (1532 ms) han before non-coronals (147 ms). Comparing he curren resuls o hose obained for he complee condiion, i is obvious ha he edied sops were overall idenified considerably less accuraely and more slowly. Ineresingly, his effec was accompanied by place and language group differences. Wih respec o place, he decrease in percen correc values in he edied condiion was almos double for /k/ (on average 48%), while i was moderae o small for // and /p/ (17% and 9% respecively). The increase in RT values was he greaes for /k/ (on average 289 ms), followed by // (145 ms) and hen /p/ (74 ms). Wih respec o he language groups, he overall decrease in percen correc values was he greaes for Russian liseners (32%), followed by English liseners (26%) and hen Korean and Mandarin liseners (2% and 19% respecively). The increase in RT values was he greaes for Russian and English liseners (24 ms and 232 ms respecively), followed by Mandarin liseners (143 ms) and hen Korean liseners (63 ms). Recall ha he edied simuli were employed in order o deermine he ne conribuion of VC ransiions on he place percepibiliy. Thus he VC ransiion resuls only parially suppor he place arge salience hypohesis (7a). Labial /p/ is more salien han coronal // (based on percen correc and RT). Dorsal /k/, however, is percepually less salien han /p/ (based on percen correc and RT). Moreover, /k/ is percepually less salien han // (based on percen correc). The percepibiliy scale 11 Reacion ime resuls for he unreleased condiion are based on daa from 53 of he 56 paricipans (due o no responses o some of he simuli from hree paricipans).
414 Alexei Kocheov and Connie K. So (a) 1 8 6 4 2 (b) 18 16 14 12 1 reacion ime (ms) mean % correc p k 8 Russian English Korean Mandarin Figure 4 (a) Mean percen correc and (b) mean reacion ime values by language group for edied syllable-final /p k/ (C1). obained labial>coronal>dorsal is a odds wih he scale proposed by Jun (1996, 24) in (4a). The resul is paricularly surprising, since he scale was originally proposed for unreleased sops presumably based on language-independen, acousic properies of heir VC ransiions. Clearly, he liseners in he curren sudy did no ake full advanage of he place informaion provided by he convergence of F2 and F3 during he VC ransiion o /k/. The poor idenificaion of /k/ is also surprising, given he finding of he ariculaory analysis ha he ongue-dorsum raising gesures in he simuli had a considerable exen in ime and were no spaially reduced (see Figs 1 and 2). I may be he case ha unreleased /k/ is inherenly less salien in his paricular vowel conex (afer he low cenral /a/), and he resuls may be differen in oher vowel conexs.12 This quesion obviously requires more sudy. I should be noed, however, ha similarly poor idenificaion of unreleased dorsals has been found in percepual sudies invesigaing final sops in a variey of vowel conexs in a number of languages (English: Householder 1956, Maléco 1958, Lisker 1999, Wrigh 21, Winers 23; Canonese: Gao 25; Duch: Winers 23; Korean: Oh 22; Thai: Abramson & Tingsabadh 1999). 12 In a fron vowel conex, F2 before a velar sop is considerably higher, resuling in a clearer convergence of F2 and F3 (Sevens 1998: 367). This may sugges a somewha beer idenificaion of velar sops in fron vowel conexs.
Place assimilaion and phoneic grounding 415 The resuls srongly suppor he release salience hypohesis in (7b): released sops are considerably more percepually salien han unreleased sops (based on percen correc and RT). This confirms some previous findings (Maléco 1958, among ohers). Ineresingly, however, he resuls show ha releases are mos imporan for dorsals, followed by coronals and hen labials. This does seem o correlae wih he relaive acousic salience of sop releases. Recall ha in he simuli, boh dorsals and coronals had louder and longer releases han labials (see w3.1.1; cf. Cho & Ladefoged 1999 on VOT differences among places). In he absence of such releases, he lingual sops coronals, and especially dorsals seem more percepually vulnerable han labials. The finding ha he scales of percepual salience are no necessarily he same for released and unreleased sops was also obained by Winers (23). I should be noed ha releases may no be he only source of he obained differences beween he wo condiions. Recall ha he ediing procedure employed in he experimen removed all acousic noise during he C1C2 inerval (see w3.1.3). Some of his noise, resuling from incomplee closure, may provide addiional place informaion (Repp 1983a, Chioran e al. 22). The use of homorganic simuli, where C1 sops were naurally unreleased, allowed us o deermine wheher he closure noise was percepually relevan here. In his case, he difference beween he complee and edied condiions is he presence or absence of any noise during he closure. The comparison of he means for wo ses revealed ha he edied homorganic sequences had a somewha lower percen correc and slower reacion ime han he complee homorganic sequences (61% vs. 68% and 149 ms vs. 141 ms). This indicaes a relaively small effec of closure noise in he resuls (comparing he magniude of differences beween he condiions in heerorganic sequences; Table I). The difference, however, can be also aribued o he liseners overall lower confidence in idenifying less informaive simuli in he edied condiion. In general, he relaive place salience for homorganic sops in boh condiions (based on percen correc) was in he same direcion as for heerorganic sops in he edied condiion labial>coronal>dorsal.13 This shows ha he VC ransiions of naurally and arificially unreleased sops in he simuli were no subsanially differen (cf. Lisker 1999 on English sops). The resuls largely suppor he universal salience hypohesis in (7c). The relaive salience of places of ariculaion was he same for all four language groups, based on percen correc and in par on RT. The lack of significan differences beween he language groups in he idenificaion of unreleased sops was due o he considerably lower performance of Russian and English liseners. This suggess ha familiariy wih released sops in naive languages was he main facor ha conribued o he language group differences in he complee condiion (see w3.2.1). 13 The means for complee homorganic sequences were 78% for /p/, 67% for // and 62% for /k/; he means for edied homorganic sequences were 66% for /p/, 61% for // and 54% for /k/.
416 Alexei Kocheov and Connie K. So (a) 1 8 6 4 complee edied.2 Russian English Korean Mandarin (b) 1 8 6 4 p k.2 complee edied Figure 5 Proporions of assimilaory errors in he idenificaion of syllablefinal sops (C1) for he complee and edied condiions (a) by language group and (b) place of ariculaion. 3.2.3 Types of errors: he complee and edied condiions. Figure 5 presens proporions of assimilaory errors in boh condiions by language group (a) and place of ariculaion (b). Recall ha in assimilaory errors, C1 is perceived as being homorganic wih C2 (e.g. /p#/ /#/); in nonassimilaory errors C1 and C2 are perceived as being heerorganic (e.g. /p#/ /k#/). The proporions of assimilaory errors here were based on he oal numbers of errors. Values higher han. 5 indicae ha here were more assimilaory han non-assimilaory errors. As seen in Fig. 5a, assimilaory errors subsanially ounumbered nonassimilaory errors across all language groups. Proporions, however, were greaer in he complee condiion han in he edied condiion: on average 72% vs. 59%. As seen in Fig. 5b, all hree places showed a predominance of assimilaory errors, wih a raher higher value for he coronal in he complee condiion. The lower proporion of assimilaory errors for unreleased /k/ was due o is frequen confusion wih //, apparenly independenly of he place of he following consonan (see Table I). Considering all errors (assimilaory and non-assimilaory), he k response was overall more frequen han he p and responses in he complee condiion (on average 41% vs. 31% and 28% of all errors).
Place assimilaion and phoneic grounding 417 This was found for all he language groups excep Korean. In he edied condiion, he k response was consisenly disfavoured across he groups, in comparison wih he p and responses (23% vs. 41% and 36%). This seems o reflec a higher number of assimilaory errors before /k/ in he one condiion, and a higher number of confusions of he edied /k/ wih eiher /p/ or // in he oher condiion (see Table I). In sum, he analysis of errors shows ha here is a subsanial assimilaory bias in boh condiions. Srikingly, alhough he overall number of errors is greaer when sops are unreleased (compared o he released sops in he complee condiion), he proporion of assimilaory errors for he unreleased condiion is lower. This suggess ha here are several acousic sources of assimilaory bias. The firs source is sop releases. Releases are likely o conain no only he informaion abou C1 bu also, due o coariculaion, some informaion abou C2. Recall ha he releases in he simuli were fairly long (on average 28 ms); hey were longes for // and shores for /p/ (see w3.1.1). The informaion abou C2 conained in some cases in he release of C1 was inerpreed by liseners as informaion abou C1. Anoher source of he assimilaory bias is possibly relaed o C2 proper is release and CV ransiions o he following vowel. CV ransiions are known o be percepually more salien han VC ransiions (Fujimura e al. 1978, Repp 1983b, Bladon 1986, Ohala 199). Ye anoher possible source of assimilaory errors are he VC ransiions, which, due o C1C2 overlap, may conain informaion abou boh C1 and C2 (cf. Byrd 1992, Zsiga 1994, Surprenan & Goldsein 1998, Son e al. 27). The prevalence of assimilaory errors in he resuls is in conras wih some previous sudies (Hura e al. 1992),14 bu is compaible wih ohers (Kingson & Shinya 23). The findings showing such an assimilaory bias in place percepion somewha weaken he claim of he PKG approach o he role of speakers phoneic knowledge in place assimilaion: if liseners errors are predominanly assimilaory, he reference o phoneic knowledge of percepual salience may no be necessary (see w2). 3.3 Conclusion The resuls of Experimen 1 fully confirm he place arge salience hypohesis (released dorsal>labial>coronal) for he complee condiion, bu only parly confirm i for he edied condiion (unreleased labial>coronal, no dorsal>labial, coronal). While in he firs condiion place of ariculaion of C1 is primarily cued by boh VC ransiions and releases, in he second condiion i is cued by he VC ransiions only. The resuls also confirm he release salience hypohesis (released>unreleased) and he universal salience hypohesis. They also show 14 Blevins (forhcoming) noes ha he prevalence of non-assimilaory errors involving syllable-final nasals /m n N/ found by Hura e al. (1992) is probably a lexical frequency effec. Specifically, he high number of n responses in he sudy corresponds o he high frequency of his consonan word-finally in English proper names, similar o he words used in he experimen in Hura e al. (1992).