Stability of Phonetic Features of Czech Plosives in Spontaneous Speech

S abili y o Phone ic Fea u es o Czech Plosi es in Spon aneous Speech Pa el machač — Radek Ska ni zl ABSTRACT: This s udy examines speech educ ions in he spon aneous Czech o six young adul speake s. Spe- ci ically, in e ocalic plosi es a e analysed om he pe spec i e o phone ic ea u es, wi h he aim o disco e hese ea u es’ ela i e s abili y. Audi o y analysis was used o de e mine he ealisa- ion ypes o plosi es, and hese ypes we e hen e i ied by acous ic analyses o du a ion, in en- si y ange, ha monici y, and oicing p o ile. The esul s show ha phonologically oiced plosi es unde go educ ion p ocesses mo e (40%), wi h semi- ocalised ealisa ion being he mos equen , while oiceless plosi es a e educed less o en (20%), wi h ica i e-like ealisa ion being he mos equen educ ion. The leas s able phone ic ea u e o Czech plosi es is hus closu e, as con i med by all he analysed acous ic pa ame e s. ABSTRAKT: S udie se zabý á řečo ými edukcemi e spon ánních p oje ech šes i mladých mlu čích češ iny. Konk é ně jsme analyzo ali ealizace in e okalických explozi zhlediska one ických ysů scí- lem popsa ela i ní s abili u ěch o ysů. Na základě audi i ní analýzy jsme u čili ypy ealizací a ak o zniklou ka ego izaci jsme o ěřo ali p os řednic ím akus ické analýzy ání, ozdílu nej- yšší anejnižší in enzi y u ni ř explozi y, ha monici y aznělos ního p o ilu. Výsledky ukazují, že onologicky znělé explozi y podléhají edukcím čas ěji (40 %) než explozi y neznělé (20 %). Uzně- lých explozi je nej ek en o anější edukcí polo okalická ealizace, uneznělých ealizace ika i- izo aná. Všechny akus ické pa ame y ukazují, že nejméně s abilním one ickým ysem českých explozi je jejich zá ě o os . KEY WORDS: speech p oduc ion, speech educ ions, phone ic ea u es, plosi es, Czech KLÍČOVÁ SLOVA: p odukce řeči, řečo é edukce, one ické ysy, explozi y, češ ina 1 INTRODUCTION One o he mos signi ican sound cha ac e is ics o spoken speech is he ich a i- abili y o wo d and sen en ial o ms in he acous ic signal (Blache & Meunie , 2004). Anumbe o educed o ms belong he e which di e om he ull p onuncia ion in ag ea e o smalle deg ee (e.g. he Czech wo d úplně “comple ely” [ʔuːplɲɛ] > [uplɲɛ]> [upəɲɛ] > [upəɛ] > [upɛː] > [pɛ ː] > [pɛ ]). To unde s and he p inciples o e- duc ions a he wo d le el, i is necessa y o know he p inciples a he le el o indi- idual segmen s and hei classes; ha is he opic o he p esen s udy. E en hough accumula ing speech educ ions may be ypical o spon aneous speech in pa icula , i is also he phone ic and sociolinguis ic s udies ha show ha speech educ ions OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 17 a e o be ound in e e y speech s yle, ac oss all social g oups, wi h each speake (e.g. G eenbe g, 1999; Keune, E nes us, Van Hou & Baayen, 2005; Schupple , E nes us, Scha enbo g & Bo es, 2011; Cangemi & Niebuh , 2018). I is he spon aneous speech and speech educ ion esea ch ha is essen ial o disco e ing and modelling he p inciples o speech communica ion, he ela ionship be ween speech p oduc ion and pe cep ion, men al s o age o wo d o ms (e.g. La- hi i & Ma slen-Wilson, 1991; Mi e e & Blome , 2003; Ranbom & Connine, 2007), o sound change (e.g. Ohala, 1981, 1989; Hamann, 2009). E en hough he e may be only asligh di e ence be ween he educed and he ull (canonical) o ms, i is no unusual o he deg ee o educ ion o be so high ha i p e en s na i e speake s om adequa ely iden i ying indi idual wo ds isola ed om he na u ally p oduced and commonly comp ehensible sen ences (e.g. Johnson, 2004; E nes us & Wa ne , 2011; Niebuh & Kohle , 2011), e.g. pan kolega “colleague” [pan kolɛga] > [βɰla]; samozřejmě “o cou se” [samoz ɛjmɲɛ] > [sˑ ɛˑ]. Resea ch in o he dynamics o sen ence educ ions in he Czech language (in p og ess) p es- en s an example ha ea u es an easily comp ehensible compound sen ence consis - ing o 19 wo ds. Mo e han 90% o lis ene s iden i ied only 5 isola ed wo ds co ec ly, o e ahal o he lis ene s ecognised only 9 wo ds and he e we e 3 wo ds ha none o he lis ene s we e able o iden i y. Wi hin an expe imen , iden i ica ion o indi- idual wo ds in acon inuous sen ence is made possible by means o linguis ic con ex . To be able o unde s and s ongly educed sen ence o ms, i is also he ex alinguis ic con ex which is necessa y o he lis ene ’sunde s anding, e.g. he educed sen en- ial o m už o jede “i ’sal eady going” [ʃ ɛˑɾɛ], canonically [ˈʔuʃ oˈjɛdɛ], is impossible o deciphe wi hou he si ua ional con ex (wai ing a abus s op). Du ing his o he speech, he speake , usually unknowingly and au oma ically, chooses om anumbe o lea ned po en ial a icula ion possibili ies (a bo h he segmen al and sup asegmen al le el in he o m o phone ic de ail) and de e mines he equency and he deg ee o hei use in ime and space (see e.g. B owman & Golds ein, 1992; Fowle & Sal zman, 1993). In he case ha he speake does no pu - sue pa icula p agma ic o aes he ic aims, he o she na u ally inclines o exe ing only he speech e o necessa y o adequa e unde s anding unde he gi en ci - cums ances (Lindblom, 1990), i.e. unde s anding ‘on he i s a emp ’. The esul ing acous ic o m o wo ds and sen ences and he deg ee o educ ion in apa icula speech ac depends on he “coo dina ion” be ween he communica ing pa ne s: he speake o e s ace ain phone ic o m while being di ec ed in eal ime by he lis- ene , who guides he speake in he case o incomp ehension o misunde s anding and hus de e mines he maximum accep able deg ee o educ ion in he pa icu- la si ua ion (wha we e e o as minimal phone ic in o ma ion; see he Discussion o mo e in o ma ion). As he communica ing pa icipan s s i e o ind abalance be ween “speake laziness” and he need o be immedia ely unde s ood, as well as he need o unde s and one’scommunica ing pa ne easily on he basis o acqui ed expe ience o speech (see he s ochas ic models o speech pe cep ion, such as Pie - ehumbe , 2003), hey de elop adynamic sys em o speech educ ions whose unc- ionali y is e i ied in e e yday con e sa ions and is u he modi ied by he p ac ice o speech. I s undamen al p inciples a e hen sha ed by he espec i e linguis ic OPEN ACCESS 18 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 communi y. Asys em o educ ions comp ises massi e educ ions a he le el o wo ds and ph ases (see he examples abo e), as well as educ ions a he le el o seg- men s (speci ically plosi es o his s udy). I is logical ha he la e will pa icipa e in educ ions a highe le els in ai ly p edic able ways. 2 PHONETIC FEATURES AND THEIR STABILITY In o de o be e unde s and he p inciples o speech educ ion a he le el o se- man ic uni s, i is use ul o conside he inhe en endencies owa ds educ ions in di e en ypes o segmen s ( o mo e in o ma ion, see he Discussion). The aim o his s udy is o documen he ypes o ealisa ions o one class o consonan s (plo- si es) in he Czech language and, a he same ime, o de e mine how equen ly hese ypes appea in spon aneous speech and o de e mine hei selec ed acous ic p ope - ies. Based on he quan i ica ion o canonical o ms and he ypes o educed o ms, we deduce he ela i e inhe en s abili y o phone ic ea u es o aphone o ag oup o phones. Acco ding o Bybee (2003, p.3) “all ca ego y membe s need no ha e all he ea u es cha ac e ising he ca ego y, bu amembe is mo e cen al o mo e ma ginal depending on he numbe and na u e o sha ed ea u es”. The educed ealisa ions o plosi es pa icula ly conce n he absence o weakening o some o he inhe en pho- ne ic ea u es (hence o h PFs; see below) and show cha ac e is ics o o he g oups o phones (e.g. semi owels, nasals). Fo p ac ical easons, he desc ip ion o he educ ions is based on he “ideal phone” which we app oach as acomplex o po en ial phone ic ea u es (i.e. asys em o a icula o y ges u es, acous ic acks and audi o y pe cep ions), ypical o a ull, canonical p onuncia ion o aphone. Following Machač and Ska ni zl (2009) we call hem he inhe en phone ic ea u es. The “ideal phone” is a heo e ical cons uc used as as a ic lis ing o he inhe en PFs o aphone o hei compa ison wi h he “ eal phone”, i.e. he ac ually p onounced one. The po en iali y o phone ic ea u es is based on he ac ha he e is no need o implemen all o he inhe en PFs in aspeci ic ealisa ion o aphone. Qui e he opposi e— mos ly unde he in luence o he pho- ne ic en i onmen — i will come o PFs no belonging o he inhe en PFs in en o y (see e.g. a icula o y p osody, Kohle , 1999), which can ha e asigni ican impac on he inal o m o he phone (Machač & Zíko á, 2015). PFs bo owed om hei (no necessa ily immedia e) neighbou s a e called ex insic; an example o an ex insic acous ic PF in plosi es can be con inuous noise o ull o man s uc u e. No using all o he inhe en PFs (e.g., closu e in plosi es) and using he ex insic PFs (e.g., ull o man s uc u e in plosi es) usually, hough no always (compa e wi h he de oic- ing o Czech in e ocalic ica i es, Machač, 2008), esul s in aweakened acous ic con as be ween he neighbou ing phones. Al hough he iden i ica ion o he phone i sel becomes mo e di icul , and some imes e en impossible, he adequa e pe cep- ion o seman ic uni s in con inuous speech does no end o be a ec ed. I is also he his o ical de elopmen o he Czech language which documen s he loss o some o he inhe en PFs and he es ablishmen o he ex insic phone ic ea- u es in a ious ypes o phones. In he a ea o plosi es, i was, o example, pala ali- OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 19 sa ion leading o he es ablishmen o pala al plosi es /c ɟ/ o ica i isa ion /g/ > /ɣ/ (in con empo a y Czech > /ɦ/, in common speech he “non- ica i e” /ɦ/ o 0, e.g. oho “ ha ” [ ogo] > [ oɣo] > [ oɦo] > [ oo] > [ oˑ]). Va ious PFs o phones can a y signi ican ly in hei s abili y ( o examples, see Table 1), whe eby he mo e s able and hus mo e equen ones can be ega ded as mo e impo an o app op ia e pe cep ion han hose ha a e o en absen o weak- ened. The s abili y o ea u es pa icula ly designa es he abili y o esis he in lu- ence o he neighbou ing segmen s (c . Bladon & Al-Bame ni’s, 1976, concep o co- a icula o y esis ance). A he same ime, he mos s able ea u es end o a ec he neighbou ing segmen s he mos and he deg ee o s abili y hus posi i ely co ela es wi h he so-called coa icula o y agg ession (Fowle & Sal zman, 1993). speech sounds mo e s able ea u es less s able ea u es owels open ocal ac , p esence o 0, o man s uc u e quan i y, quali y, o al cha ac e nasals nasali y occlusion, place o a icula ion oiced sibilan s place and manne o a icula ion p esence o 0 oiceless sibilan s place and manne o a icula ion absence o 0 Table 1:Examples o mo e s able and less s able PFs in Czech (based on Machač & Ska ni zl, 2009; Machač & Zíko á, 2015). The aim o his s udy is o de e mine he ela i e s abili y o PFs o Czech plosi es in he in e ocalic posi ion on he basis o audi o y ca ego isa ion o hei ealisa ions, e i ied by acous ic analyses. The ype and equency o de ia ions om canonical p oduc ion o plosi es is ela ed o wo ac o s which may play a ole in he ealisa- ions: lexical s ess and wo d equency. The Czech language has eigh plosi es: bilabial /p b/, al eola / d/, pala al /c ɟ/ and ela /kɡ/. Phonological oicing in Czech is implemen ed by means o phone ic oicing, i.e. he p esence o absence o ocal old ib a ion h oughou he phone. The inhe en PFs o plosi es, conside ed om h ee in e connec ed phone ic pe - spec i es (p oduc ion, acous ics and pe cep ion), a e a) a icula o y ea u es: closu e, s op elease, place o a icula ion, p esence o ab- sence o ocal old ib a ion, aised so pala e; b) acous ic ea u es: lowe ed in ensi y, noise phase, p esence o absence o unda- men al equency ( 0), absence o ull o man s uc u e; c) pe cep ual ea u es: lowe ed loudness, elease bu s , oicing o oicelessness, o ali y, place o a icula ion (bilabiali y, al eola i y…). This di ision o PFs co esponds o he e na y phone ic pe spec i e o he same sound subs ance and o he objec i es o his s udy. He e, we i s use pe cep ual anal- ysis o examine he s abili y o a icula o y ea u es (ins umen al me hods o a icu- la o y phone ics a e no sui able o s udying spon aneous speech), and second, we OPEN ACCESS 20 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 iden i y acous ic cha ac e is ics o di e en ways o plosi e a icula ion. Fo ins ance, oicing/ oicelessness may be analysed pe cep ually; based on ha , we may in e a - icula o y cha ac e is ics (p esence/absence o ocal old ib a ion) which, in u n, a e mani es ed acous ically as p esence/absence o undamen al equency. When assessing he s abili y o PFs, we use adi ionally used and unambiguous ca ego ical e ms like oicing/ oicelessness and closu e e c., a he han he mo e clumsy p esence/absence o ocal old ib a ion and lowe ed loudness, espec i ely. 3 METHOD In his s udy, we i s conduc ed audi o y analysis o he eco ded da a o de e mine he ealisa ion ypes o plosi es. The ealisa ion ype co esponds o he g oup o plosi es cha ac e ised by an iden ical manne o ealiza ion o PFs: canonical, semi- ocalised, ica i ised, nasalised (see sec ion 4.1 o mo e de ails). Subsequen ly, we s udied hei selec ed acous ic cha ac e is ics (du a ion, in ensi y ange, ha monic- i y, oicing p o ile) and e i ied he ele ance o he pe cep ually iden i ied ca ego- ies. Then, we in es iga ed he dependence o he occu ence o he ypes o ealisa- ion o plosi es on oicing, place o a icula ion (as he essen ial inhe en PFs), wo d s ess and wo d equency (as ex e nal ac o s). Based on all hese indings, we s a ed he ela i e s abili y o PFs o plosi es in Czech. 3.1 mATERIAL Six highly spon aneous dialogues we e used o he analyses. Each dialogue pai consis ed o aCha les Uni e si y s uden who in i ed agood iend o hei s o an in e iew. I was he speech o he iends (i.e., n = 6) ha was analysed. The speak- e s ( ou emales, wo males) we e in he age ange o 19 o 25. Each in e iew las ed app ox. 40 minu es and was eco ded in he sound- ea ed s udio o he Ins i u e o Phone ics in P ague. In he analyses, we ocused on in e ocalic plosi es, ini ially o medially wi h espec o wo d bounda ies; i is p ecisely he in e ocalic posi ion whe e he high- es le el o using inhe en PFs may be an icipa ed. Table 2 shows he numbe o he indi idual plosi es based on lexical s ess o he gi en syllable. The smalle numbe o okens o pala al plosi es is due o hei o e all ela i ely low equency in Czech; in he case o /ɡ/, he low numbe is due o hei occu ence (in he in e ocalic posi ion) only wi hin loan wo ds in Czech. In o al, 2,187 plosi es we e analysed in his s udy. 3.2 AUDITORY ANALYSIS The aim o he audi o y analysis was o de e mine he ealisa ion ypes o he a ge plosi es. Bo h au ho s e alua ed he plosi es by means o ca e ul lis ening (wi h he use o wa e o m and spec og am displays; see Figu e 1 and he ele an commen s in sec ion 4.1 o mo e de ail). Ra e disc epancies in he wo au ho s’ e alua ions we e OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 21 add essed by join analysis. In he case o educed o ms, he absence o inhe en and p esence o ex insic PFs was eco ded; in he case o g adual educ ion, he deg ee o hei p esence ( he deg ee o opening o he ocal ac , exp essed in phone ically meaning ul ca ego ies) was eco ded. The moni o ed PFs a e lis ed in Table 3. Nei he he place o a icula ion no i s po en ial change a e lis ed in Table 3, be- cause he co esponding inhe en audi o y PFs— bilabiali y, al eola i y, pala ali y and ela i y— we e p ese ed e en in he educed ealisa ions. The place o a icula- ion hus, in line wi h p e ious in o mal obse a ions, p o ed o be ahighly s able PF o in e ocalic plosi es (con a y o nasals, o ins ance). 3.3 ACOUSTIC ANALYSIS The aim o he acous ic analysis was o in es iga e he ele ance o he ca ego ies es ablished on he basis o he audi o y e alua ion. Fi s ly, he segmen bounda ies we e au oma ically aligned using HMM (Pollák, Volín & Ska ni zl, 2007) and hen adjus ed manually acco ding o he ules s a ed in Machač & Ska ni zl (2009). The acous ic analyses we e pe o med in P aa (Boe sma & Weenink, 2011), ocusing on he ollowing acous ic pa ame e s (see sec ion 5 o mo e de ails): a) plosi e du a ion; b) di e ence be ween he highes and he lowes in ensi y wi hin he plosi e ( he di e ence would be la ge in acanonical plosi e, especially a oiceless one); s essed uns essed o al p 179 178 357 b 134 183 317 184 184 368 d 183 184 367 c 21 147 168 ɟ52 198 250 k 103 187 290 ɡ23 47 70 Table 2:Numbe o in e ocalic plosi es a he beginning o s essed and uns essed syllables. inhe en PFs ex insic PFs closu e semi- ocali y ica i i y lap (only o /d/) ← opening o he ocal ac — la ge ← opening o he ocal ac — smalle ← ex eme sho ening o closu e oicing / oicelessness oicelessness / oicing o ali y nasali y Table 3:Phone ic ea u es moni o ed du ing he audi o y analysis. OPEN ACCESS 22 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 c) mean ha monici y (HNR; Boe sma, 1993) wi hin he plosi e (acanonical plosi e would mani es low HNR alues, co esponding o he p e alence o noise-like, a he han one-like componen s in he spec um); d) oicing p o ile (Möbius, 2004), which enables he s udy o dynamic cha ac e is- ics o oicing h oughou aphone. Speci ics, as well as he pu pose o measu ing pa ame e s b) o d), a e p o ided in sec ions 5.2 o 5.4, along wi h he esul s and hei in e p e a ion. Fundamen al equency ( 0), ha monici y and in ensi y we e ex ac ed in P aa using he de aul se ings. Fo he oicing p o ile, 0 was ex ac ed in i e equidis an s eps h oughou each plosi e. 4 RESULTS I– AUDITORY EVALUATION 4.1 VARIABILITY IN THE mANNER OF ARTICULATION The mos impo an ypes o non-canonical plosi e ealisa ions a e, om he pe - spec i e o he manne o a icula ion, hose which we e semi- ocalised and ica- i ised: he a icula ion o gans do no each he a ge posi ion, asmalle o la ge gap emains be ween hem and he exhaled ai low is no in e up ed. As a esul , he leas s able inhe en PF is closu e, and he mos common ex insic PF is opening o he ocal ac . Examples o wa e o m and spec og aphic displays a e shown in Figu e 1. Asmalle opening o he ocal ac esul s in acon inuous ic ion noise and a ica i ised plosi e (Fig. 1a, b). Ala ge opening o he ocal ac and he p esence o one ( o man s uc u e) in phone ically oiced plosi es is ypical o semi- ocali- sed ealisa ions (Fig. 1c, d). I is clea ha he g adien na u e o his ype o ges u al educ ion may p e en eliable iden i ica ion o he s a ed ca ego ies. Ano he ex insic PF eco ded in in e ocalic plosi es is nasali y. In spi e o aclosu e o med in he o al ca i y, he so pala e lowe s, o ins ance due o de- c eased speech e o and/o o easons o dis an coa icula ion (e.g., [nɛbudɛmɛ] > [nɛmuɾɛmɛ] nebudeme “we will no ”). Wi h ai lowing h ough he nasal ca i y, he gi en phone loses he inhe en PF o ali y and gains he ex insic PF nasali y. The oiced plosi es hus ans o m in o nasals wi h he same place o a icula ion: [b] > [m], [d] > [n], [ɟ] > [ɲ], [ɡ] > [ŋ]. Fo nasalised oiceless plosi es, i is possible o eco d only a e y ain ic ion noise o igina ing du ing he passage o ai h ough he nasal ca i y: [p] > [m ], [ ] > [n ], [c] > [ɲ], [k] > [ŋ]. Figu e 1d shows an example o a oiced nasalised “plosi e”; i s nasal cha ac e is clea ly audible and is also mani- es ed h ough he nasal o man a ound 2 kHz. The al eola [d] is aplosi e wi h he sho es du a ion in Czech, wi h he a e age du a ion o he closu e phase being 47 ms (Machač, 2006). I s u he sho ening esul s in a lapped ealisa ion, ha is, aquick, ballis ic ongue mo emen wi hou as able closu e phase. OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 23 4.1.1 FREQUENCY OF REALISATIONS TYPES BASED ON VOICING The ae odynamic di e ences o oiced and oiceless plosi es esul in adi e se e- quency o he ypes o ealisa ions. Figu e 2 shows some in e es ing aspec s. Voiced plosi es a e educed signi ican ly mo e o en (38.1%) han he oiceless plosi es (18.7%), as con i med by he Ya es-co ec ed chi-squa ed es : χ2 (1; n = 2174)= 100.54; p < 0.0001. This esul co esponds o he mo e di icul p oduc ion o phone- ically oiced plosi es: du ing hei p oduc ion, he e is acon lic c ea ed be ween he need o keep asu icien ansglo al p essu e di e en ial and, a he same ime, iGu e 1:Examples o changes in he manne o a icula ion: a. ica i ised /d/; b. ica i ised /k/; c.semi- ocalised /d/; d. semi- ocalised and nasalised /ɟ/. The a ge ex insic phone ic ea u es a e indica ed by he ellipses. OPEN ACCESS 24 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 he need o acomple e closu e o he o al ca i y. This con lic may be mi iga ed o elimina ed applying a ious s a egies. Fi s o all, speake s may inc ease he olume o he ocal ac o accommoda e oicing o alonge ime (Wes bu y, 1983; S i sky e al., 1997). In e ms o educ ion s a egies, i seems ha Czech speake s p e e he elimina ion o he closu e as inhe en PF— as a esul , he di e ence in subglo al and sup aglo al p essu e can be main ained mo e easily, and he ocal olds ib a e h oughou he plosi e. This is impo an especially in languages such as Czech in which phonological oicing co ela es mainly wi h phone ic oicing, ha is wi h he p esence/absence o 0 o he whole du a ion o he phone. On he o he hand, en- seness appea s o be me ely aseconda y ac o o pe cep ion, wi h he excep ion o whispe (Ska ni zl, Š u m & Machač, 2013). Modi ica ion o he manne o a icula ion has been shown o be he p e alen ype o educ ion o plosi es. Almos 40% o all oiced plosi es lack he inhe en PF closu e. The mos common ype o educ ion is semi- ocalisa ion (59.1% o all non- canonical ealisa ions), which is well ahead o ica i isa ion (17.4%) and o ealisa- ion o /d/ as a lap (18.7%); nasalisa ion occu s ma ginally (4.7%). In compa ison, in he oiceless plosi es, he closu e as inhe en PF disappea s “only” in less han one i h o all cases, and he mos common ype o educ ion is ica i isa ion (82.3%). In he case ha he ex insic ea u e opening o he ocal ac (la ge) is joined by he ex insic PF one-like cha ac e , he esul is semi- ocalisa ion (13.6%). The e was also ama ginal occu ence o “ oiceless” nasalisa ion (4.1%). 4.1.2 FREQUENCY OF REALISATION TYPES BASED ON PLACE OF ARTICULATION As seen in Figu e 3, he esul s show aclea endency owa ds ahighe p obabili y o occu ence o educ ions a mo e pos e io places o a icula ion and, a he same ime, in plosi es which equi e ala ge mass o he ac i e a icula o ; his does no seem o hold s ic ly only o /ɡ/. The explana ion o he o e all endency is ob ious: iGu e 2:F equency o educ ion ypes based on oicing. OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 31 which we e de oiced; see below) and also in semi- ocalised phonologically oiceless plosi es. The emaining ealisa ion ypes o oiceless plosi es a e cha ac e ised by aseemingly complica ed and, a he same ime, la gely simila cou se o oicing p o- babili y: i dec eases apidly du ing app oxima ely he i s hal o he segmen due o he o se o 0, oicing is leas p obable be ween 50 and 75% o he du a ion, and hen oicing p obabili y ises apidly again as a esul o 0 onse associa ed wi h he ollowing owel. These di e ences con i m he ele ance o he ca ego y semi- ocali- sed in phonologically oiceless plosi es. Voicing p o iles o plosi es wi h espec o hei phone ic oicing a e depic ed in Figu e 10. The e a e clea simila i ies in e ms o phone ic oicing: he cou se o oic- ing p obabili y is simila , on he one hand, ac oss canonically ealised phonologically oiced plosi es and phonologically oiceless bu phone ically oiced plosi es, and on he o he hand, ac oss canonically ealised oiceless plosi es and phonologically oiceless bu de oiced plosi es. An excep ion among he phonologically oiced plo- si es is ep esen ed by he canonically ealised /ɡ/ which pa ially loses i s oicing; his is in acco dance wi h he smalles space be ween he ocal olds and he closu e and he esul ing in ense ae odynamic con lic being a i s mos in ense (see also, e.g., Ohala, 1983; Kea ing, 1984). iGu e 9:Voicing p o ile o plosi es ealised cano- nically, as semi- ocalised, ica i ised, and nasalised. iGu e 10:Voicing p o i- les in di e en changes o plosi e oicing. OPEN ACCESS 32 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 The abo e-s a ed indings suppo he impo ance o he pe cep ually s a ed ca e- go ies o phone ic oicing and de oicing, al hough i is necessa y o epea ha de oi- cing ea u es only spo adically in ou ma e ial. 5.5 SUmmARY OF RESULTS— ACOUSTIC ANALYSIS The subsequen acous ic analysis was used (1) o iden i y acous ic cha ac e is ics co - esponding o he indi idual ypes o ealisa ions, (2) o con i m he meaning ulness o he ca ego ies iden i ied pe cep ually and wi h he aid o acous ic displays and, a he same ime, (3) o es ways o iden i ying he educed o ms in compa ison o he canonical o ms (du a ion, in ensi y ange, ha monici y, oicing p o ile). All he ou examined acous ic pa ame e s p o ed o be impo an o dis inguish- ing be ween canonical and semi- ocalised ealisa ions, and also o he iden i ica ion o oicing changes (i.e., o dis inguishing be ween canonically oiceless and pho- ne ically oiced plosi es, and be ween canonically oiced and de oiced ones). The esul s o he acous ic analyses a e summa ised in Table 5. acous ic pa ame e manne o ealisa ion phone ic oicing de oicing du a ion can_ d > s oc can_ l > d+ a e, no e i ied in ensi y di e ence can_ d > s oc can_ l > d+ can_ d < l– ha monici y can_ l < s oc (can_ d < s oc) can_ l < d+ can_ d > l– oicing p o ile can_ d < s oc can_ l < d+ can_ d > l– Table 5:Compa ison o acous ic pa ame e alues o canonical (can) and semi- ocalised (s oc) e- alisa ions and o oicing changes ( l = phonologically oiceless, d = phonologically oiced; d+ pho- ne ically oiced, l– phone ically de oiced). I is wo h poin ing ou ha ica i ised a ian s did no show any s a is ically sig- ni ican di e ences in any o he examined acous ic pa ame e s. This may seem su - p ising; we hypo hesised he in ensi y ange wi hin ica i ised plosi es o be lowe han wi hin canonical ones. As can be seen in Figu e 7, howe e , such a endency o lowe alues o in ensi y anges can in ac be obse ed in /ɟ/, / / and /c/; o e all, he di e ence as compa ed o he canonical plosi es does no each signi icance. 6 CONCLUSIONS The aim o his s udy was o con ibu e o he desc ip ion o he ela i e s abili y o phone ic ea u es o phones, in pa icula h ough he s udy o he s abili y o phone ic ea u es o Czech plosi es in spon aneous speech. While desc ibing he ac ual eali- sa ions o phones, he compa isons in he s udy a e based on he “ideal phone”, which se es as a heo e ical cons uc including all phone ic ea u es p esen in i s ull (ca- nonical) ealisa ion, ha is he inhe en phone ic ea u es. In he ac ual ealisa ion o he phone, hese ea u es end o be weakened o comple ely absen and simul ane- OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 33 ously eplaced by ex insic phone ic ea u es which end o be he inhe en ea u es o su ounding phones. 6.1 STABILITY OF PHONETIC FEATURES OF INTERVOCALIC PLOSIVES The s abili y o an inhe en phone ic ea u e o agi en ype o plosi es may be quan- i ied by de e mining he a io o he numbe o plosi es in which his ea u e was ealised and he o al numbe o plosi es analysed. Fo ins ance, i 300 ou o 500 oiced plosi es a e p onounced wi h closu e, he s abili y o ha ea u e is 0.6. Ta- ble 6 p esen s alues o he ela i e s abili y o he phone ic ea u es o plosi es. I is in e es ing o compa e hese alues wi h hose conce ning he s abili y o oicing in Czech phonologically oiced ica i es in he in e ocalic posi ion: [z] 0.67; [ʒ] 0.73; [ ] 0.42; [ ] 0.99 (based on Machač, 2008: Table 2). Unlike in plosi es, phone ic oic- ing in he phonologically oiced ica i es is a he an uns able ea u e, especially wi h he ica i e ill /  / (0.42). inhe en PFs oiced plosi es oiceless plosi es closu e 0.62 0.81 o ali y 0.95 0.99 oicing/ oicelessness 0.99 0.93 place o a icula ion 1.00 1.00 Table 6:Rela i e s abili y o phone ic ea u es exp essing he a io o Czech in e ocalic plosi es wi h he ea u e ac ually ealised (see ex ). The alues ob ained o in e ocalic plosi es imply he ollowing hie a chy o he s a- bili y o phone ic ea u es: place o a icula ion > oicing and oicelessness = o ali y > clo- su e. The same hie a chy can be ound o a ica es and oiceless ica i es in Czech. 6.2 GENERAL DISCUSSION The issue o speech educ ions is app oached in his pape om he pe spec i e o phone ic ea u es. We belie e ha i we know hei ela i e s abili y in speci ic (classes o ) phones in agi en segmen al con ex , we may be able, in modelling speech educ ions, o dis inguish mo e easily be ween educ ions o igina ing om he ins abili y o aphone ic ea u e as an inhe en ai o agi en phone (such e- duc ions a e hus mo e easily p edic able), and hose o igina ing om ajoin in- luence o b oade condi ions. On he basis o such adis inc ion, i migh be possi- ble o be mo e success ul in disco e ing he p inciples and limi a ions o educ ion p ocesses in ela ion o he con ex o speaking and he si ua ion. The p esence o such ac o s in e e yday communica ion enables he lis ene o iden i y he mean- ing o he message, in spi e o apossibly signi ican accumula ion o e y s ongly educed o ms (see In oduc ion). The ela ionships wi hin he sys em o speech educ ions and hei limi a ions can be assessed, o example, acco ding o he spe- ci ic o m o minimal phone ic in o ma ion, which ep esen s o he lis ene he OPEN ACCESS 34 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 lowes accep able le el o he use o inhe en phone ic ea u es and he maximum le el o he use o ex insic ea u es. The ollowing sen ence may se e as an example o ap obable minimal phone ic in o ma ion: já ne ím, jes li jsem o už nepsala “Idon’ know whe he Iha en’ w i en i al eady” [jaːˈnɛ iːmˈjɛs lɪsɛm ouʃˈnɛpsala] > [aˈəɪˈɛssən əuʃˈəpsaa]. In e p e ing he educ ions by means o phone ic ea u es also p o ides o a a he ap desc ip ion o ea u es ha a e some imes in e p e ed as elision. In ull p onuncia ion, he speake o he s a ed sen ence would ealise 27 segmen s, while only 17 segmen s a e iden i- ied in he educed o m. Tha ne e heless does no mean ha 10 segmen s ha e been elided. The numbe o segmen s which le no ace wha soe e is only 4: [ ] in [nɛ iːm] > [ə ɪ]; [ l] in [jɛs lɪ] > [ɛ s ]; [l] in [nɛpsala] > [ə psaa]. In he o he 6 cases, he seemingly missing segmen is ep esen ed by some o i s phone ic ea u es ans- e ed o he neighbou ing phone (pa allel a icula ion), o i is indica ed by apho- ne ic de ail: [j] in [jaː] > [a] ( esidual so oice onse a he beginning o he owel); [n m] in [nɛ iːm] > [ə ɪ] (nasalisa ion o owels); [j] in [jɛs lɪ] > [ɛ s ] (pala alisa ion o he owel); [ɪ] in [jɛs lɪ] > [ɛ s ] (du a ion and syllabici y o he oiceless obs u- en ); [n] in [nɛpsala] > [ə psaa] (nasalisa ion o he owel). Machač & Zíko á (2015) show ha inhe en phone ic ea u es o “pseudo-elided” segmen s, ans e ed on o neighbou ing phones whe e hey ac as ex insic phone ic ea u es, a e in many cases able o dis inguish he meaning o he message and a e hus essen ial o adequa e unde s anding. Based on he analyses o s ongly educed speech uni s ha ep esen ap obable minimal phone ic in o ma ion o a e close o i , and wi h he knowledge o he inhe en s abili y o phone ic ea u es, i is possible o in e he s abili y o phone ic ea u es om he pe spec i e o wide seman ic uni s; in hese uni s, ea u e s abili y would be dependen on he s uc u e, meaning and equency o he lexical and sen en ial uni s, as well as on o he ac o s. The knowledge o inhe en and “con ex ual” s a- bili y should make i easie o model he minimal phone ic in o ma ion in agi en language, and can con ibu e o he unde s anding o he ela ionship be ween p o- duc ion and pe cep ion in o dina y speech. Based on he exis ing obse a ions o phone educ ions (including elisions) in Czech, i can be o ins ance concluded ha he mos s able phones wi hin wo ds a e hose ha a e dis inguished mos by sono i y. Speci ically, should awo d o ag oup o wo ds con ain owels, sono an s, oiced obs uen s and oiceless obs uen s, hen i is p obable ha owels and oiceless obs uen s (i.e., segmen s ha a e ecip ocally mos con as ing) will app oxima e ull p onuncia ion mos , while segmen s which suppo he acous ic con as o asmalle deg ee (sono an s, oiced obs uen s) will mo e likely be suscep ible o leni ion, o aloss o some o he phone ic ea u es, o o elision. I mus be s essed ha he me hod o phone ic ea u es is ega ded as apho- ne ic app oach and no necessa ily aphonological one. The p ima y objec i e is no o s udy he dis inc i e ea u es wi hin agi en language sys em, bu he s udy o phone ic o phonological ules in eal speech. In his app oach, we hus do no s i e o c ea e he smalles possible se o ea u es; qui e he opposi e, all phone ic de ail is aken in o conside a ion. OPEN ACCESS PAVEL mACHAč — RADEK SKARNITZL 35 REFERENCES Blache, P., & Meunie , C. (2004). Language as acomplex sys em: The case o phone ic a iabili y. P oc. o VI Cong eso de Lingüís ica Gene al. Sain -Jacques de Compos elle. Bladon, R.A.W., & Al-Bame ni, A. (1976). Coa icula ion esis ance in English /l/. Jou nal o Phone ics, 4, 137–150. Boe sma, P. (1993). Accu a e sho - e m analysis o he undamen al equency and he ha monics- o-noise a io o asampled sound. IFA P oceedings, 17, 97–110. Boe sma, P., & Weenink, D.J.M. (2011). P aa : doing phone ics by compu e (Ve sion 5.2.30). A ailable om h p://www.p aa .o g. B owman, C.P., & Golds ein, L.M. (1992). A icula o y phonology: An o e iew. Phone ica, 49, 155–180. Bybee, J. (2003). Phonology and Language Use. Camb idge: Camb idge Uni e si y P ess. Cangemi, F., & Niebuh , O. (2018). Re hinking educ ion and canonical o ms. In F.Cangemi, M.Claya ds, O.Niebuh , B.Schupple & M.Zelle (Eds.), Re hinking educ ion (pp.277–302). Be lin: Wal e de G uy e . Če mák, F., Křen, M. e al. (2004). F ek enční slo ník češ iny. P aha: Naklada els í Lido é no iny. E nes us, M., & Wa ne , N. (2011). An in oduc ion o educed p onuncia ion a ian s. Jou nal o Phone ics, 39, 253–260. Fowle , C.A., & Sal zman, E. (1993). Coo dina ion and coa icula ion in speech p oduc ion. Language and Speech, 36, 171–195. G eenbe g, S. (1999). Speaking in sho hand— Asyllable-cen ic pe spec i e o unde s anding p onuncia ion a ia ion. Speech Communica ion, 29, 159–176. Hamann, S. (2009). The lea ne o ape cep ion g amma as asou ce o sound change. In P.Boe sma & S.Hamann (Eds.), Phonology in pe cep ion (pp. 111–149). Be lin: Mou on de G uy e . Johnson, K. (2004). Massi e educ ion in con e sa ional Ame ican English. In K.Yoneyama & K.Maekawa (Eds.), Spon aneous speech: Da a and analysis. P oceedings o he 1s Session o he 10 h In e na ional Symposium (pp. 29–54). Tokyo: The Na ional In e na ional Ins i u e o Japanese Language. Kea ing, P.A. (1984). Physiological e ec s on s op consonan oicing. UCLA Wo king Pape s in Phone ics, 59, 29–34. Keune, K., E nes us, M., Van Hou , R., & Baayen, R.H. (2005). Social, geog aphical, and egis e a ia ion in Du ch: F om w i en MOGELIJK o spoken MOK. Co pus Linguis ics and Linguis ic Theo y, 1, 183–223. Kohle , K.J. (1999). A icula o y p osodies in Ge man educed speech. P oc. o he 14 h ICPhS, 1, San F ancisco, 89–92. Lade oged, P., & Maddieson, I. (1996). The Sounds o he Wo ld’sLanguages. Ox o d: Blackwell Publishe s. Lahi i, A., & Ma slen-Wilson, W.D. (1991). The men al ep esen a ion o lexical o m: Aphonological app oach o he ecogni ion lexicon. Cogni ion, 38, 245–294. Lindblom, B. (1990). Explaining phone ic a ia ion: Aske ch o he H&H Theo y. In W.J.Ha dcas le & A.Ma chal (Eds.), Speech P oduc ion and Speech Modelling (pp. 403–439). Do d ech : Kluwe . Machač, P. (2006). Tempo ální aspek ální s uk u a českých explozí . Ph.D. disse a ion, Ins i u e o Phone ics, Facul y o A s, Cha les Uni e si y, P ague. Machač, P. (2008). Desono izace českých in e okalických ika i . In J.Volín & J.Janouško á (Eds.), AUC Philologica 2/2007, Phone ica P agensia XI (pp. 105–116). P aha: Ka olinum. Machač, P., & Ska ni zl, R. (2009). P inciples o Phone ic Segmen a ion. P aha: Epocha. Machač, P., & Zíko á, M. (2015). Pa allel a icula ion: The phone ic base and he phonological po en iali y. Slo o aslo esnos , 76(1), 3–21. Mi e e , H., & Blome , R. (2003). Coping wi h phonological assimila ion in speech pe cep ion: E idence o ea ly compensa ion. Pe cep ion & Psychophysics, 65, 956–969. OPEN ACCESS 36 STUDIE ZAPLIKOVANÉ LINGVISTIKY 2/2020 Möbius, B. (2004). Co pus-based in es iga ions on he phone ics o consonan oicing. Folia Linguis ica, 38, 5–26. Niebuh , O., & Kohle , K.J. (2011). Pe cep ion o phone ic de ail in he iden i ica ion o highly educed wo ds. Jou nal o Phone ics, 39, 319–329. Ohala, J.J. (1981). The lis ene as asou ce o sound change. In C.S.Masek, R.A.Hend ick & M.F.Mille (Eds.), Pape s om he Pa asession on Language and Beha io (pp.178–203). Chicago, IL: Chicago Linguis ic Socie y. Ohala, J.J. (1983). The o igin o sound pa e ns in ocal ac cons ain s. In P.F.MacNeilage (Ed.), The P oduc ion o Speech (pp. 189–216). New Yo k, NY: Sp inge . Ohala, J.J. (1989). Sound change is d awn om apool o synch onic a ia ion. In L.E.B ei ik & E.H.Jah (Eds.), Language Change: Con ibu ions o he S udy o I s Causes (pp.173–198). Be lin: Mou on de G uy e . Pie ehumbe , J.B. (2003). P obabilis ic phonology: Disc imina ion and obus ness. In R.Bod, J.Hay & S.Jannedy (Eds.), P obabilis ic Linguis ics (pp. 177–228). Camb idge, MA: MIT P ess. Pluymaeke s, M., E nes us, M., & Baayen, R.H. (2005). Lexical equency and acous ic educ ion in spoken Du ch. Jou nal o he Acous ical Socie y o Ame ica, 118, 2561–2569. Pollák, P., Volín, J., & Ska ni zl, R. (2007). HMM-based phone ic segmen a ion in P aa en i onmen . P oc. o XII h SPECOM 2007 (pp.537–541). Moscow. Ranbom, L.J., & Connine, C.M. (2007). Lexical ep esen a ion o phonological a ia ion in spoken wo d ecogni ion. Jou nal o Memo y and Language, 57, 273–298. Schupple , B., E nes us, M., Scha enbo g, O., & Bo es, L. (2011). Acous ic educ ion in con e sa ional Du ch: Aquan i a i e analysis based on au oma ically gene a ed segmen al ansc ip ions. Jou nal o Phone ics, 39, 96–109. Ska ni zl, R. (2011). Znělos ní kon as nejen češ ině. P aha: Epocha. Ska ni zl, R. (2018). Fone ická ealizace slo ního příz uku udelších slo češ ině. Slo o aslo esnos , 79, 199–216. Ska ni zl, R., Š u m, P., & Machač, P. (2013). The phonological oicing con as in Czech: An EPG s udy o phona ed and whispe ed ica i es. P oc. o In e speech 2013 (pp.3191–3195). Lyon. S i sky, M.A., S e ens, K.N., Ma hies, M.L., Manzella, J.P., Pe kell, J.S., & Wilhelms- T ica ico, R. (1997). Tongue su ace displacemen du ing bilabial s ops. Jou nal o he Acous ical Socie y o Ame ica, 102, 562–571. Wes bu y, J.R. (1983). Enla gemen o he sup aglo al ca i y and i s ela ion o s op consonan oicing. Jou nal o he Acous ical Socie y o Ame ica, 73, 1322–1336. Pa el Machač | Ús a obecné ling is iky FF UK <pa el.machac@ .cuni.cz> Radek Ska ni zl | Fone ický ús a FF UK < adek.ska ni zl@ .cuni.cz> This wo k was suppo ed by he Eu opean Regional De elopmen Fund p ojec “C ea i i y and Adap abili y as Condi ions o he Success o Eu ope in an In e ela ed Wo ld” ( eg. no.: CZ.02.1.01 /0.0/0.0/16_019/0000734) and by Cha les Uni e si y p ojec P og es Q10, “Language in he shi ings o ime, space, and cul u e”. OPEN ACCESS