Phonological vs. Phonetic Representations

24.901 Phonological vs. Phonetic Representations 1. reprise • a speech sound is a particular combination of plus/minus choices for the distinctive features • the features define natural classes for phonological rules and determine the grammatically controlled properties of the articulation and acoustics of the utterance • most features are binary 2. languages make different choices among the distinctive features to encode their vocabularies • voicing ([±voice] is not contrastive in Finnish but it is in English: (i.e. sue vs. zoo; Lacey vs. lazy; face vs. phase; no such pairs in Finnish; vowel length is contrastive in Japanese (e: ‘yes’ vs. e ‘picture’) but not in Spanish • one goal of the phonological analysis of a language is to determine which features are contrastive • the sounds composed of contrastive features are called phonemes • pairs of lexical items that differ by a single feature value are called minimal pairs • they are good clues as to which features are contrastive in the language (but cf. writer vs. rider below) • the contrastive (nonpredictable) feature values that distinguish one lexical item from another encode the lexicon • predictable, redundant feature values are assigned by phonological rules when the lexical item is inserted into a sentence and enters the phonological component • the phonetic representation thus has plus/minus specified for all relevant features, telling the vocal apparatus how to articulate the sound • for most features there is an unmarked (default) value that represents the optimal state for that feature • when there is no lexical contrast for a feature, the unmarked value normally appears phonetically (unless there is assimilation to a neighboring sound): Finnish stops are voiceless [-voice], Spanish vowels are [-long]; sonorant consonants are voiced; voiceless stops and short vowels are more optimal: they are found in most/all languages • PGG Chapter 2 assumes that such noncontrastive features are [0F] in the lexicon and the unmarked value is assigned in the phonology by default rules • Recent research challenges this assumption and would represent unmarked optimal value in the lexicon • We will largely sidestep this question in this class • Typologically some features are very common sites of contrast while others are less so or rare: all languages contrast [±high] in vowels and almost all [±back]; [±nasal] is common while [±retroflex] (cf. English bird) is rare; almost all languages have a dental stop and a nasal; most have an /s/; many contrast [±voice]; while clicks are rare 1 [3]. Even if a feature is not lexically contrastive in a language, phonological rules may introduce the marked value in particular contexts; sounds that bear the marked value are allophones (predictable variants) of the underlying phoneme two examples from English Example 1 • In French [±nasal] encodes lexical contrasts in vowels: beau [bo] vs. bon [b!]" • English lacks such minimal pairs but it does have nasal vowels before a nasal consonant: cf. see [i] vs. seen [i]; " sew [o] vs. sewn [o]" • The French and English nasal vowels are phonetically comparable (identical) but have phonologically different status: nasality is assigned by rule in English but is contrastive in French French /bo/ English /b!"/ /so/ /so-n/ Lexical # Underlying Representation so"n [bo] • [b!]" [so] [so"n] [+syll] -> [+nasal] / ____ [+cons, +nasal] Phonetic Representation The nasal vowel in English [son] " is thus an allophone of the oral phoneme /o/ while in French the nasal vowel is phonemic • Psycholinguistic experiments show this difference is real: hearing part of a nasal vowel, English subjects expect a following nasal consonant while French (Bengali) subjects do not • In English the oral ([-nasal]) and nasal ([+nasal]) vowels are in complementary distribution: nasal vowels occur before a nasal consonant and nowhere else while oral vowels appear everywhere except before nasal consonants Example 2 • In Mandarin [±spread gl] encodes lexical contrasts: pha$ ‘to fear’ vs. pa$ ‘father’ • In English voiceless aspirated and unaspirated stops are in complementary distribution; no such minimal pairs exist: [ph]in vs. s[p]in; aspirated stops occur at beginning of the syllable while unaspirated stops appear elsewhere; initial medial final h [p ]in [b]in ra[p]id ra[b]id la[p] la[b] [t ]ot [d]ot a[t]om a[d]am ma[t] ma[d] [k ]ap [g]ap jac[k]et jag[g]ed pic[k] pi[g] h h • [±voice] is contrastive in initial, medial, and final position as seen in the above minimal pairs • text (and most literature) assumes that [−voice] is contrastive and aspiration is a redundant (enhancing) feature [−contin, −voice] -> [+spread gl] / . ___ (C) [+syll] (dot = syllable boundary) 2 “a voiceless stop becomes aspirated at the beginning of a syllable” cf. [ph]a'cific, [th]er'rific, [kh]or'rupt ra'[ph]idity, a['th]omic, a'[kh]oustic • [−contin, −voice] -> [−spread gl] / 'V__V • several rules of English phonology sensitive to intervocalic context V1___V2 where V2 is unstressed: cf. annual [nj] but annuity [n]; ve[ ]icle but ve[h]icular • sC clusters are site of no contrast in [voice] (and [spread gl]): s[p]in, s[t]em, s[k]in, *zbin, *bzin • unmarked default values emerge: [−voice] [−spread gl] • there is one qualification to the statement that no minimal pairs exist for aspiration; this is true at the level of the word; but at the level of the phrase one must know where the word-boundaries fall to assign aspiration: cf. ice cream [kh] vs. I scream [k] the distribution of such features are plausibly parsing cues that allow the listener to locate word • boundaries redundant features like aspiration and vowel nasality are commonly believed to enhance a • phonemic contrast to aid in the recovery of the underlying form that allows the listener to enter the lexicon to interpret the sentence 5. vowel is lengthened before voiced consonant/ shortened before voiceless lap vs. la:b hit vs. hi:d buck vs. bu:g fuss vs. fu:zz [æ] [æ:] [%] [&] [&:] [%:] [&] [&:] rope vs. ro:be heat vs. heed hake vs. Hai:g rice vs. rise [o] [o:] [i] [i:] [e] [e:] [aj] [a:j] 6. the phoneme /t/ has a variety of realizations (allophones) in English; [t] plain stem [t ] aspirated ten ['] retroflexed trip [( ] flapped atom [t ] glottalized hit, atlas [)] glottal stop bottle [ ] zero pants h ? [+coronal, −contin] -> [+apical, −anterior] / ___ [r] /t/ -> ['] [+coronal, −contin] -> [+constricted gl] / in coda of syllable 3 /t/ -> [t?] [+coronal, −contin, +constricted gl] -> [−coronal] in casual speech [t?] -> [)] [+coronal, −contin] -> [+sonorant, +voice] / [+syllabic] ___ [+syllabic] (flapping) −stress 7. writer vs. rider and Canadian Raising (Joos 1942) [&j] type tight tyke rice [aj] time tide tie rise [aj] -> [&j] / ___ [−voice] ( [+low] -> [−low] / __ [−syllabic] [−voice] ) −cons 8. write [*&jt] ride [*ajd] writer [*&j(+*] rider[*aj(+*] • if we define phonemes solely by minimal pairs then [aj] vs [&j] would be a phonemic contrast despite the predictable distribution of [&j] • analysis follows straightforwardly from two ordered rules that are needed anyway /*ajt/ /*ajt-+*/ /*ajd/ /*ajd-+*/ *&jt *&jt+* ------ -------- Canadian Raising ---- *&j(+* ----- *aj(+* flapping • which features are contrastive can only be determined for the grammar as a whole • opacity: the distribution of [aj] vs [&j] cannot be discerned from the surface phonetic representation: we must "undo" the flapping rule • expected result if the phonetic representation is derived from the underlying phonological representation by locally determined rules that apply without regard to their long-range, downstream consequences • implication: the underlying form can only be recovered by knowing the rules, raising nontrivial learnability problems 9. more opacity ten t[,:n] tent t[,nt] tend t[,:nd] ten-s t[,:nz] tent-s t[,ns] tend-s t[,:nz] vowel length and voicing of suffix depend on the /t/ vs. /d/ contrast in the cluster but the stop is normally deleted medially in the cluster 4 /#t,nt-z#/ /#t,nd-z#/ t,nts ------- suffix voicing assimilation ------ t,:ndz vowel lengthening t,ns t,:nz cluster simplification 10. typology of complementary distribution • one sound in restricted context, other is elsewhere case: e.g. flap in Am English • more balanced distribution: vowel length and voicing of following consonant: lengthening or shortening? • accidental complementary distribution: In English [h] is restricted to beginning of word or stressed syllable while [-] cannot begin a word or a stressed syllable • whether two sounds are analyzed as variants of the same underlying phoneme depends on the simplicity of the overall analysis and the plausibility and naturalness of the rules; since [h] and [-] share no features in common it is implausible to derive them from the same underlying phoneme More examples (from Odden 2006) Kenyang (Cameroon). The velar stop [k] and the uvular [q] are in complementary distribution. What are the contexts where each occurs? Are there reasons for deriving one stop from the other? en!q ‘tree’ enoq ‘drum’ eket ‘house’ nt.iku ‘I am buying’ nek ‘rope’ eyw/rek ‘sweet potato’ -g/q ‘knife’ ek/q ‘leg’ ayuk name ku ‘buy’ kebwep ‘stammering’ nk!q ‘chicken’ -k/p k! ‘walk’ ‘money’ Gen (Togo) The liquids [l] and [r] are in complementary distribution. Which occurs in in a more restrictive context? a-!li ‘ghost’ ago-glo ‘lizard’ al! ‘hand’ sra ‘strain’ avl! ‘bait’ edr! ‘dream’ exl! ‘friend’ hl, ‘read’ -l! ‘write’ 0ra" ‘be ugly’ kl! ‘wash’ tre ‘glue’ mla ‘pound drum’ zro ‘fly’ l! ‘like’ d ro ‘hint’ 1 5 Kikuyu infinitive prefix 2o-t,-,ra ‘to run’ 2o-kuua ‘to carry’ 2o-koora ‘to root out’ ko-ruya ‘to cook’ ko-oria ‘to ask’ ko-m,0a ‘to know’ ko-h!ta ‘to be able’ ko-ina ‘to dance’ ko-niina ‘to finish’ 2o-kaya ‘to cut’ 2o-t.uuka ‘to slander’ ko-2aya ‘to divide’ Palauan Part of the analysis of any language is decomposing complex words into their constituent morphemes: in the simplest cases roots, suffixes, and prefixes. Examine the following paradigms to isolate the roots. What must the underlying lexical representation be? present future-1 future-2 m+3da-+b d+-+3ball d+3-obl ‘cover’ m+3te)+b t+)+3ball t+3)ibl ‘pull out’ m+3-et+m -+t+3mall -+3toml ‘lick’ m+3tab+k t+b+3kall t+3bakl ‘patch’ m+3)ar+m )+r+3mall )+3roml ‘taste’ m+3ses+b s+s+3ball s+3sobl ‘burn’ Tibetan numerals. This is a more tricky case of the same morphological parsing problem. Look for recurring forms to isolate the morphemes composing these numbers. What is the phonological rule that derives the surface form? d1ig ‘1’ d1u ‘10’ d1ugd1ig ‘11’ .i ‘4’ d ub.i ‘14’ .ibd u ‘40’ gu ‘9’ d urgu ‘19’ gubd u ‘90’ -a ‘5’ d u-a ‘15’ -abd u ‘50’ 1 1 1 1 1 1 Interacting rules in two Bantu languages Lamba (Zambia) c4 = affricate [t.] Base Passive Neuter Applied Reciprocal c4ita citwa 4 c4itika citila 4 c4itana ‘do’ tula tulwa tulika tulila tulana ‘dig’ c4eta 4 cetwa c4eteka 4 cetela c4etana ‘spy’ 6 so-ka so-kwa so-keka so-kela so-kana ‘pay tax’ pata patwa patika patila patana ‘scold’ fisa fiswa fi.ika fi.ila fisana ‘hide’ c4esa 4 ceswa 4 ce.ika 4 ce.ila c4esana ‘cut’ kosa koswa koseka kosela kosana ‘be strong’ lasa laswa laseka lasela lasana ‘wound’ .ika .ikwa .ic4ika 4 .icila .ikana ‘bury’ seka sekwa sekeka sekela sekana ‘laught at’ poka pokwa pokeka pokela pokana ‘receive’ kaka kakwa kac4ika 4 kacila kakana ‘tie’ fuka fukwa fuc4ika fuc4ila fukana ‘creep’ Lomongo (Congo) Imper 1sg 2sg 3sg 1pl sa-ga nsa-ga osa-ga asa-ga tosa-ga losa-ga basa-ga kamba -kamba okamba akamba tokamba lokamba bakamba ‘work’ j4ila 4 njila oj4ila 4 ajila toj4ila loj4ila baj4ila ‘wait’ ,na 4 nj,na w,na ,na t w,na j4w,na b,na ‘see’ isa 4 njisa wisa isa t wisa j4wisa bisa ‘hide’ usa 4 njusa wusa usa t wusa j4wusa busa ‘throw’ ina 4 njina wina ina t wina j4wina bina ‘dance’ s s s s 2pl 7 3pl ‘say’