Phonology Summary

Fall 2012 Phonology Summary • Speakers of English learned something from the data they were presented with as babies that caused them to internalize (learn) the rule exemplified in (1) — just as Tojolabal speakers learned from the data that they heard as babies, and ended up with the rule exemplified in (2). • The English rule is real and active "on-line", governing creative linguistic behavior. A native speaker of English will apply it to new words they have never heard. The /t/ in tib will be aspirated, and the /t/ in stib (both nonsense words, I hope) will not be. Probably Tojolabal speakers will show similar behavior with respect to their rule. (contains all examples from class slides, and more!) 1. Phonology vs. phonetics • • The path from memory (lexical access) to speech is mediated by phonology. Phonology = system of rules that apply when speech sounds are put together to form morphemes and words. (1) stop consonant aspiration in English: initial within a stressed syllable ASPIRATED initial within a stressed syllable UNASPIRATED after s or initial within an unstressed syllable word-final (therefore syllablefinal) pan tone kin span stone skin nap note nick upon attack akin supping attic hacking • What English speakers know is a rule that relates a form stored in memory (in the sound field of the lexicon) to pronunciation. The rule concerns speech sounds, and therefore involves phonetics. But aspiration at the beginning of stressed syllables is not an inevitable part of a human language — you can have a human vocal tract and not use the rule exemplified in (1). This makes it a rule of phonology rather than a fact about phonetics. • Speakers of Spanish, for example, do not have the English rule, which is why the syllable-initial stop consonants of a native speaker of Spanish sound "funny" to the English ear (and vice-versa). More strikingly yet, speakers of Tojolabal (a Mayan language of Mexico) have a rule that is almost the exact opposite of the English rule, aspirating word (syllable?)-final [t]. but not [t] elsewhere: (2) aspiration of [t] in Tojolabal: word (syllable?)-finally1 1. tʃit˭am 'pig' 5. tʃat˭atʰ kind of plant 2. makt˭on 'a patch' 6. mutʰ 'chicken' 3. pot˭otʼ kind of plant 7. nahatʰ 'long' 'upside down' 8. ʔinatʰ 'seed' 4. t˭inan (t˭ = unaspirated /t/) 1 Source: Gleason, Henry A. (1955) Workbook in Descriptive Linguistics. New York:, Holt, Rinehart and Winston. Note: I only have information about [t], and don't know if the rule extends to the other stops. 2. What phonetic distinctions are made in lexical entries? Part 1: phonological rules that eliminate distinctions from the lexicon • English: lexicon does not need to distinguish aspirated from unaspirated stops. There is no reason to suppose that information about aspiration forms part of the sound field of lexical entries of English words, since it is entirely predictable. Though pan is pronounced /pʰæn/ and span /spæn/, there is no reason to distinguish the aspirated and unaspirated bilabial stops in the lexical entries of the two words. Instead, we may simply use the unaspirated version in the lexicon for both words (pæn, spæn) and let the phonological rule take care of the pronunciation difference. • To make the same point differently, we could not even conceive of two distinct words in English that differ only in the aspiration of a stop consonant. If you invent a product and call it "pʰæb", I could not invent a competing product and call it "pæb" to distinguish it. The second form is simply the first form with a foreign accent! The English lexicon does not even allow the aspirated and the unaspirated variants to be distinguished. It hands the job off completely to the phonology. • This too is an English-specific fact! In some languages the lexicon does make the relevant distinction. For example, Hindi, where words are often distinguished only by a contrast in aspiration for a stop consonant: (3) Hindi: lexicon distinguishes aspirated from unaspirated stops in the lexicon note the minimal pairs pal 'take care of' pʰal 'knife blade' tal 'beat' tʰal 'plate' ʈal 'postpone' ʈʰal 'wood shop' tʃal 'turn' tʃʰal 'bark' kal 'era' kʰal 'skin' -2• The same points can be made with other phonological rules. For example: (4) vowel length in English: V→ [+long] / __ [C, +voiced] (short and long vowels in complementary distribution, i.e. not distinguished in the lexical entries for English morphemes, but derived by a phonological rule) short vowel meet back pot loss half [mit] [bæk] [pɑt] [lɔs] [hæf] long vowel mead bag pod laws have made-up words wuck [wʌk] wug nis [nɪs] niz • • • [mi:d] [bæ:g] [pɑ:d] [lɔ:z] [hæ:v] [wʌ:g] [nɪ:z] Nonsense words and the pronunciation of non-English words like wuck and wug show that this too is a rule. Vowel length is not part of lexical entries in English! You could not imagine, for example, /nik/ and /ni:k/ being distinguishable names for new inventions in English. On the other hand, many languages do distinguish vowel length in the lexicon. Latin, for example, and many other languages, e.g. Finnish (a brilliantly demonstrated for us in class): (5) vowel length in Finnish: (short and long vowels distinguished in the lexical entries for Finnish morphemes; note the minimal pairs!) tuli 'fire' tu:li 'wind' ehto 'condition' ehto: 'evening' sika 'pig' si:ka 'whitefish' (6) [l] vs. [r] in Luganda (Bantu, Uganda) [source: workbook by Gleason] [l]→[r] / [V, -back]2 [l] and [r] in complementary distribution, i.e. not distinguished in the lexical entries for Luganda morphemes, but derived by a phonological rule 1. òkúdʸdʸùlà 'to be full of' 11. éfírímbí 'a whistle' 2. òkúdʸdʸúlá 'to dish up' 12. éñdʸálíírò 'beam' 3. òkúgúlá 'to buy' 13. èríñá 'name' 4. òkúggúlá 'to open' 14. éttúúndíró 'shop' 5. òkùléétà 'to bring' 15. myéréèrè 'only' 6. òkútábáálà 'to attack' 16. ñdʸírí 'Gospel' 7. òmùlálà 'insane person' 17. òkùlíírà 'to eat with' 8. òmùlémà 'lame person' 18. ólúgéró 'story' 9. náálwááná 'fought' 19. ókúmíírá 'to swallow' 10. túlèègà 'we tighten' 20. pírípírí 'chili pepper' (7) [l] vs. [r] in Korean [l] → r / V __ V 1. kal 2. kɯnɯl 3. mul 4. pal 5. pʰal 6. səәul 7. tatɯl 8. ilkop 9. ipalsa 10. onɯlppam [source: workbook by Gleason] 'that'll go' 11. pulpʰyəәn 'discomfort' 'shade' 12. silkwa 'kind of fruit' 'water' 13. əәlmana 'how much' 'foot' 14. irɯm 'name' 'arm' 15. kiri 'length' 'Seoul' 16. kɯrəәm 'then' 'all of them' 17. kəәriro 'to the street' 'seven' 18. saram 'person' 'barber' 19. uri 'we' 'tonight' 20. yəәrɯm 'summer' ([ɯ] is a high back unrounded vowel) (8) [l] vs. [ɹ] in English distinguished in the lexical entries for English morphemes: note the minimal pairs late rate lot rot lamb ram belly berry hail hair pool poor 2 The data are insufficient to tell us whether lexical entries contain [l] and a rule derives [r], or whether it is the other way around. The same is true for Korean below. -3Where the zero-ending form has a voiceless consonant and the -a form has a voiced counterpart, the lexicon contains a voiced consonant that the rule of final devoicing turns into its voiceless counterpart. 3. What phonetic distinctions are made in lexical entries? Part 2: phonological rules that do not eliminate distinctions from the lexicon Russian Consider the following pairs of forms from Russian. I call the forms in the left column "zero-ending forms" and the forms in the left column "-a forms", though the meaning and use of the two forms actually varies widely among the examples. Assume that the -a form contains two morphemes: a root and a suffix -a (brought together into onew word by the rule Merge). (11) a. b. c. d. e. f. g. h. i. j. k. (9) Final devoicing in Russian a. b. c. d. e. f. g. h. i. j. k. • • zero-ending form -a form grɔp lip ʃɔpəәt gɔrəәt ruk rɔk rɔk kəәrəәndaʃ ekipaʃ ras ras grɔba lipa ʃɔpəәta gɔrəәda ruka rɔɡa rɔka kəәrəәndaʃa ekipaʒa rasa raza 'coffin' 'linden tree' 'whisper' 'city' 'hand' 'horn' 'fate' 'pencil' 'crew' 'race' 'time' • step 1: Comparing the two columns, we note that sometimes the -a form has a voiced consonant (stop or fricative) where the zero-ending form has the corresponding voiceless consonant. (Example: the /p/ vs. /b/ in the first example.) This already suggests that Russian has either a rule that makes voiced consonants voiceless in a certain environment, or else a rule that makes voiceless consonants voiced. Which is it? step 2: Next we notice that sometimes both columns have the voiceless alternant. (Example: the /p/ in the second example.) But there are no examples in which the zero-ending form has a voiced consonant while the -a form has a voiceless consonant. • Now let's add an additional fact: No Russian word is ever pronounced so as to end with a voiced consonant (with some law-governed exceptions coming up). A Russian accent in English has this property too: an English word like [prawd] 'proud' comes out as [prawt]. • Solution: Russian has a rule of final devoicing that devoices a voiced consonant at the end of a word. The rule leaves consonants that are already voiceless alone. (10) Final devoicing (first version, to be revised)3 [+consonant] ! [-voiced] / ___ # 3 The symbol # indicates the boundary of the word. See the textbook for the format of phonological rules used here. Final devoicing in Russian zero-ending form -a form grɔp lip ʃɔpəәt gɔrəәt ruk rɔk rɔk kəәrəәndaʃ ekipaʃ ras ras grɔba lipa ʃɔpəәta gɔrəәda ruka rɔɡa rɔka kəәrəәndaʃa ekipaʒa rasa raza lexical entry for the root4 'coffin' 'linden tree' 'whisper' 'city' 'hand' 'horn' 'fate' 'pencil' 'crew' 'race' 'time' grɔb lip ʃɔpəәt gɔrəәd ruk rɔg rɔk kəәrəәndaʃ ekipaʒ ras raz Important points: 1. Listing the voiced alternant in the sound field of the lexical entries for the roots of these words makes it possible to predict the pronunciation of both the zero-ending form and the -a form. If we were to list the voiceless alternant, we could not predict when the -a form shows a voiced consonant and when it shows a voiceless consonant. 2. Forms with a voiced final consonant are never pronounced that way! So this is a case in which the lexicon makes more distinctions than those made in the spoken form of the language (a voicing contrast in word-final position) — a contrast to the previous section, in which we saw the lexicon making fewer distinctions than those made in the spoken form (e.g. with respect to aspiration in English). • Finally, the details of Russian motivate a restriction of the rule in (10) to exclude liquids (/l/ and /r/) and nasals (12) No devoicing of liquids and nasals dɔm dɔma 'house' plan plana 'plan' skəәzal skəәzala 'said' vɔr vɔra 'thief' (13) Final devoicing (final version) +consonant -sonorant → -voiced / ___ # 4 Our lexical entries ignore some irrelevant additional details of Russian phonology — for example, the fact that the schwa vowels listed here are not actually part of the lexical entries either, but are the result of rules that apply to other vowels when they are unstressed. -4English nasal assimilation: Could there be a word chinp (vs. chimp)? What about /þink/ (vs. /þiŋk/ )? emd (vs. end)? No, because English has a rule of nasal assimilation, identical to the rule also illustrated in (17) below: 4. More phonology • Phonology vs. phonetics once more: Rules like (13) are very common in the world's languages. But not universal. English, for example, does not have a rule of final devoicing. (17) Itzá6 (Mayan, Guatemala): nasal assimilation tsʔon 'gun' ʔintsʔon 'my gun' kʔab 'hand' ʔiŋkʔab 'my hand' bat 'ax' ʔimbat 'my ax' pal 'son' ʔimpal 'my son' kol 'field' ʔiŋkol 'my field' siʔ 'wood' ʔinsiʔ 'my wood' lak 'bowl' ʔinlak 'my bowl' lets 'trap' ʔinlets 'my trap' (18) Nasal Assimilation Rule A nasal assimilates in place of articulation to the following consonant Part of having a "Russian accent" when speaking English is to erroneously apply this rule when speaking English. Another very common (but not universal) kind of rule is assimilation. Voicing assimilation (14) Voicing assimilation in English: the plural suffix a. plural morpheme as /s/: [lɪp-s, pɑt-s, nɛk-s, læf-s ...] b. (15) Voicing assimilation in English past tense: invented words a. plural morpheme as /s/: [zɑp-s, wʌk-s, ...] b.  plural morpheme as /z/: [kæb-z, pɑd-z, ɛg-z, hajv-z ...] past morpheme as /d/: [zɑb-z, wʌg-z, ...] The point (yet again!): the past tense rule is not about a list of phones turning into another list. Rather it is about an abstract property of phones in general (voiced vs. voiceless):5 • Here is an approximation at a formal statement of the rule, using the terms introduced in class for the phonetic properties of consonants: [+nasal] → [a bilabial, b alveolar, γ velar]/ ___ [α bilabial, β alveolar, γ velar] — where α, β, γ are variables that range over plus and minus values. Using the features discussed in the book, but not in class, that would be: [+nasal] → [α labial, β coronal, γ dorsal]/ ___ [α labial, β coronal, γ dorsal] Important note: From the data given, we can't tell what the underlying form of 'my' is in Mayan. It might be ʔin, ʔiŋ or ʔim. Vowel harmony (16) Voicing assimilation z → [-voiced] / [-voiced] __ When assimilation affects a vowel, the process is called vowel harmony. For example: Nasal assimilation • As noted, rules like (16) — assimilation rules — are very common in the world's languages. One of the most common phonological processes across the world's languages is nasal assimilation. A nasal consonant takes on the place of articulation of a consonant that follows it. 5 As came up in answer to a question, a schwa is inserted after sibilants: [kɪsəәz, bʌzəәz, edʒəәz] etc. 6 Data once again from the Gleason worksbook cited above. Gleason calls the language "Mayan", which is actually a language family. Thanks to Robert Henderson for identifying the actual language. -5(19) Turkish vowel harmony singular plural diʃ diʃler tʃotsuk tʃotsuklar asker askerler kedi kediler son sonlar masa masalar getse getseler baba babalar kuʃ kuʃlar køj køjler jyz jyzler kɯz kɯzlar • 'tooth' 'child' 'soldier' 'cat' 'end' 'table' 'night' 'father' 'bird' 'village' 'face' 'girl' Tojolabal and English: predictable alternation of +aspirated and -aspirated (different rules) Spanish: always unaspirated Hindi: unpredictable whether a consonant is aspirated or not Conclusions: Tojolabal, English: aspiration is not phonemeic (for example, /t/ and /tʰ/ are allophones of the same phoneme) [y], [ø] = high and mid front rounded vowels (Turkish spelling: ü and ö) [ɯ] = high back unrounded vowel (Turkish spelling: dotless ı) • In the plural suffix, we find a back vowel ([a]) when the root ends in a back vowel ([u], [a], [o], [ɯ;) and a front vowel ([e]) when the root ends in a front vowel ([[i], [e], [y], [ø]). This is the result of a rule of front-back vowel harmony. Hindi: aspiration is phonemic (for example, /t/ and /tʰ/ are distinct phonemes) • 5. Phonemes • • • • We can immediately establish that two phones are not in complementary distribution in a language if we can find a minimal pair of words that differ by only that phone. The minimal pair is the gold standard of phonological analysis. but near-mininal pairs will often do almost as well, i.e. two words containing XaY and XbY where X and Y are the same. At least we know that a vs. b is not predictable on the basis of X vs. Y.] Question: What is the inventory of the sounds that are found in the morphemes of a language as they are stored in the sound field of the lexicon? The list of these sounds is the list of the phonemes of the language. 6. Phonemic inventory The list of actual sounds used in the language (the phonemes, as modified by the phonological rules of the language), is the list of its phones. • Why did our brains in infancy go to the trouble of working out how to parse the speech stream into morphemes? Probable answer: to store information about these units in the database that we call the lexicon. When the infant acquiring language notes that two or more phones are in complementary distribution, he/she uses this fact to economize on the distinctions made in the lexicon. Two phones in complementary distribution are taken to correspond to only one sound stored in memory, a phoneme — with the variant pronunciations related to the single phoneme by phonological rule. The variant forms of phoneme that are related by rule are called its allophones. The set of phonemes of a language are called its phonemic inventory. record low: maybe Rotokas (S. Pacific), 11 phonemes record high: !Xoo (Khoisan group, Botswana, Namibia), 141~160. 80 of them are clicks! These concepts in context • Review, in this connection, our discussion of aspiration: English: about 37-41, depending on dialect and analysis • Alphabetic writing systems often choose represent phonemes rather than their allophones. • Speakers of languages tend to be minimally aware or unaware of allophonic distinctions. (Were you aware of the rule governing stop aspiration in English?)  Categorical perception for non-phonemic distinctions seems to be suppressed in the process of language acquisition. Work done at Haskins laboratories in the 1970s and 80s (by Virginia Mann, among others) seems to show that adult speakers of English -6make a categorical distinction between [l] and [r], but adult speakers of Japanese do not. • Example 1: Tunica Other experiments show that infants -- both Japanese and American -- make the categorical distinction. So language acquisition from this perspective looks more like a process of losing distinctions than gaining them — unlearning rather than learning. 7. Distinctive features. • Phonological processes tend to be "natural", in that they apply to natural classes of sounds. The environment of a rule is also generally a natural class of sounds, and the change effected by the rule involves a natural class of sounds. What is a natural class of sounds? A group of sounds that have a common phonetic property. For example, for consonants: point of articulation, manner of articulation and voicing. Other (rather sophisticated) ways of classifying sounds are discussed at length in the textbook, but were not discussed in class. Take 24.901 next Fall to learn more. 8. Rule ordering • • Lexical entries may undergo more than one phonological rule. The various rules do not seem to apply equally or all at once, but are prioritized. There is controversy within the field of phonology about the precise nature of this prioritization. (One currently popular approach, called Optimality Theory is not the approach taught here, but is stressed in other classes in linguistics.) For our purposes, the prioritization can be understood as an ordering of the rules: they apply in a serial fashion, each rule applying to the output of the preceding rule. Answer: Vowel harmony in the 3sg fem form precedes unstressed vowel deletion — which if it had applied first, would have eliminated the environment for harmony [source: Roca, Iggy and Wyn Johnson. 1999. A Course in Phonology] Example 2: Shona The sequence of ordered rules is called a derivation. In class, we examined three problems that show how an ordered derivation works. Answer: The output of nasal assimilation (see below) reveals the place of articulation of a following voiceless stop that has turned into [h]. So nasal assimilation must precede the stop→[h] rule, so it can see the difference between a [p], [k] or [t] that later changes to [h]. Roca, Iggy, and Wyn Johnson. A Course in Phonology. Blackwell Publishers, 1999. © Blackwell Publishers. All rights reserved. -7as the [r] can be syllabified into the coda of a preceding syllable. Thus: airtent is a possbile word, since it is syllabified [σ ejr][ σ tɛnt]. 9. Syllable structure This is a topic that I presented very briefly, which really deserved a full class or two to itself — mainly because the next topic was writing systems, and I wanted to discuss syllables so we could later discuss syllabaries. Here's what we talked about (and there's more material in the textbook). • • Phonological segments are not arranged like beads on a string (any more than morphemes within a word are, or words within a sentence) — but are organized in a hierarchical fashion into syllables. Though there is some disagreement among experts (and some interesting recent work revising the model), a standard view holds that segments are grouped as in (20), where the Greek letter σ (a lowercase sigma) indicates "syllable".7 Note that in general, sonority increases as you move from either syllable-edge towards the nucleus.8 (20) • Both onset and rime are optional, but the nucleus (containing one or more vowels, syllabic liquids or nasals) is obligatory. Evidence for the existence of onset and coda as structural units includes: • The existence of phonotactic constraints (constraints on co-occurrence) within the onset and within the rime, vs. the near-absence of comparable constrants that cross the coda-onset divide. For example, in English trent is a possible (but not an actual) word, but *rtent is an impossible word, since [rt] is not a possible onset (violating the principle of increasing sonority). On the other hand, [r] may immediately precede [t] even in English, so long 7 Linguists' spelling of rime is an homage to Coleridge's poem "The Rime of the Ancient Mariner" (http://www.gutenberg.org/files/151/151-h/151-h.htm -- in case you don't know it). 8 Vowels are more sonorous than consonants; nasals and liquids are more sonorous than stops; voiced consonants are more sonorous than voiceless consonants. Crosslinguistically, the behavior of sibilants is interestingly anomolous: a word like [trɪp] obeys the sonority generalization, but the initial consonant of [strɪp] violates it. • Branching rimes and branching codas attract stress in many languages, while branching onsets generally do not (i.e. there is a generalization that requires reference to these constituents within the syllable). (continued on next page) -8step 2. Link each consonant that immediately precedes a vowel to Onset. Add as many other consonants as possible to Onset ("maximize Onset") — so long as phonotactics (sonority sequencing) is not violated. Example: the Latin stress rule (21) Latin stress: data short penult9 mí-se-rum dú-ci-mus ré-fi-cit dó-mi-nus prin-cí-pi-bus closed10 penult a-mán-tis re-féc-tus im-pe-di-mén-ta con-sér-vat 'poor' (NEU) 'we lead' '(s)he repairs' 'master' 'chiefs' (ABL) long penult a-má:-vi: a-má:-bo: du-cé:-mus hor-tó:-rum stop+liquid 'I have loved' cé-le-bris 'I will love' vó-lu-cris 'we will lead' 'gardens' (GEN) 'loving' (GEN) 'repaired' 'baggage' '(s)he preserves' σ σ σ σ σ σ σ σ R R R R R R R R ONONON ON N O NON O N i m p e d i m e n t a 'crowded' (GEN) 'winged' (GEN) v o l u c r i s The underlined segments are not added to onsets in step 2 because of sonority. [mp] and [nt] are not possible onsets in Latin. step 3. Link remaining (unsyllabified) consonants to Coda two-syllable words té-git 'covers' mó-rem 'sea' (ACC) (Latin spelling: c = [k], v = [w]) σ σ σ σ σ σ σ R R R R R R R σ R Latin syllabification rules: step 1. Link all vowels to Nucleus, project Rime and Syllable: N CO NONONCON O NON v o l u c r i s σ σ σ σ σ σ σ σ i m p e d i m e n t a R R R R R R R R Latin stress rule: N N N N N i m p e d i m e n t a N N O N C 1. If the penult (second-to-last syllable) has branching structure in its rime (is heavy), stress the penult. a-mán-tis re-féc-tus a-má:-vi: du-cé:-mus N v o l u cri s im-pe-di-mén-ta (see syllable structures above) 2. Otherwise, if the word has 3 or more syllables, stress the antepenult [third syllable from end] prin-cí-pi-bus dú-ci-mus cé-le-bris vó-lu-cris 3. 9 Penult = second syllable from the end of a word. Antepenult = third syllable from the end of the word. 10 A syllable is closed if it contains material in its coda (e.g. [d 3. mí-se-rum (see syllable structures above) Otherwise, stress the first syllable té-git mó-rem