TETU and Non­reduplicative Correspondence

24.962 Advanced phonology 2 Mar, 2005 TETU and Non­reduplicative correspondence (1) Today’s agenda • Finish up discussion of reduplication with a brief discussion of fixed segmentism • Move on to correspondence relations beyond those found in morphological reduplication TETU and fixed segmentism (2) Overview of what we have seen so far • Reduplication is specified in the input, as a RED morpheme • The source, amount, and placement of copied material can be determined by constraints on the RED morpheme (A LIGN, L OCALITY, etc.) • The exact segmental makeup is determined by the interplay of faithfulness and markedness – Here, F­IR/F­BR) and M • When F IB � M � F {I,B}R we get a pattern where marked structures are preserved in the base (F IB � M) but eliminated in the reduplicant (M � F {I,B}R ) – The Emergence of the Unmarked (TETU) (3) Example: reduplication of loanwords Tagalog (Zuraw 1996) Pattern for at least some speakers: Base TeNkju fambUl SapiN • Redup mag­te­TeNkju mag­pa­fambUl mag­sa­SapiN Gloss ‘say thank you’ ‘to fumble’ ‘to shop’ I DENT­IO[±contin],[±anterior],etc. � *T, *f, *S� I DENT­BR[±contin],[±anterior],etc. (4) This raises the possibility that all I DENTBR constraints can be outranked; the segment is copied, but can’t retain any of its feature specifications (unless they happen to be completely unmarked) Hypothetical example, assuming [P] = least marked C, [@] = least marked V, CV = least marked σ Base b1 l2 u3 p4 i5 m1 o2 f3 o4 g1 a2 b3 u4 Redup P1 @2 ­b1 l2 u3 p4 i5 P1 @2 ­m1 o2 f3 o4 P1 @2 ­g1 a2 b3 u4 • Technically, the theory provides for a reduplication analysis of this pattern. Should we be worried? (5) Yoruba (Marantz 1982, Pulleyblank, Alderete et al, etc.) Verb lo ` d˙un gb´oná jE rı́ Noun lı́lo ` dı́ḋun gbı́gb´oná jı́jE rı́rı́ Gloss ‘go/act of going’ ‘be tasty/tastiness’ ‘be warm/warmth’ ‘eat/act of eating’ ‘see/act of seeing’ • Reduplicant always has [ı́], no matter what the input vowel • What are three possible analyses of the [ı́]? 24.962—2 Mar, 2005 p. 2 (6) Alderete et al: phonological fixed segmentism by defaults • [i] is default: *non­i � *i /d/ ☞ a. b. c. d da, de, do, du di N UC *! D EP * NON ­ I * * *! *i * • Prevented from copying non­[i] vowels (TETU configuration) – I DENTIB (vowel features) � *non­i � MAXBR , *[i] • Result: epenthetic [i] rather than copied vowel /RED­da/ a. d1 a2 d1 a2 b. d1 i2 d1 i2 ☞ c. d1 id1 a2 I DENTIB * NON ­ I **! M AXBR *i * * ** * *! • Alternatively: imperfectly copied [a] rather than perfectly copied [a] /RED­da/ a. d1 a2 d1 a2 b. d1 i2 d1 i2 ☞ c. d1 i2 d1 a2 I DENTIB * NON ­ I **! I DENTBR *i * * ** * *! Either way, claim is that the quality of the [i] is predictable. • A desirable type of result: in reduplication, you either get a perfect copy of the original seg­ ment, or one that deviates from the original in a predictable way (by being less marked) (7) Unfortunately, not true of all reduplication cases • English schm­ reduplication (table, schmable); [Sm] is almost certainly not epenthetic or less marked than [t] • Still need “morphological fixed segmentism”: /RED+schm+table/ (8) Claimed distinctions • Phonological fixed segmentism: segment is a “default” (least marked) • Morphological fixed segmentism: segments must display properties of an affix – Appear near edge of a word (prefix/suffix) – Other “morpheme”­like properties (irregularity, contextually conditioned allomorphy, etc. (9) How can we tell the difference? E.g., Igbo a. High vowels copied exactly ti­ti ‘cracking’ nu­nu ‘pushing’ ji­ji ‘snapping’ ju­ju ‘being full’ mI­mI ‘drying’ mu­mu ‘learning’ b. Non­high: i before alveopalatal, u before labial (with ATR harmony) cI­cO ñi­ño yI­yO bu­be gbu­gbe ku­kwe ‘seeking’ ‘shadow’ ‘begging’ ‘cutting’ ‘crawling’ ‘agreeing’ 24.962—2 Mar, 2005 p. 3 c. Otherwise i, harmonizing in ATR and rounding with stem vowel ki­ke nI­na kU­kO nu­no ‘sharing’ ‘going home’ ‘telling’ ‘swallowing’ • Alderete & al claim that the vowel is a copy in (a), epenthetic in (b, c) • But what precludes an underlying /i/? or [+high] feature? (A prefix, forced to infix into redu­ plicant to satisfy O NSET) (10) The upshot • The idea of phonologically­determined fixed segmentism has the potential to yield a more restrictive theory of possible fixed segments; unfortunately, it’s too restrictive (we still need the more powerful mechanism of overwriting in reduplication) • The choice seems arbitrary in many cases – In fact, it’s not even clear that a purely phonological analysis is ever truly necessary • This is not to say it’s harmful (the theory predicts it, and it does not generate monsters) Non­reduplicative correspondence (11) Tagalog vowel alternations and pseudoreduplication [see solution on separate handout] (12) What creates a correspondence relation between simlar syllables? • Possibility 1: at some point, when the similarity between consecutive syllables (or feet, etc.) is too great to be a coindence, it is given reduplicative structure in the lexicon • Possibility 2: rankable constraint remanding that all words have reduplicative structure (13) The R EDUP constraint (Zuraw 2002) • Words must have reduplicative structure (i.e., substrings that stand in correspondence with one another) (14) Interaction of R EDUP with I DENT constraints • New dimension of correspondence: base to pseudo­reduplicant (I’ll call this BP, to distin­ guish from BR; Zuraw calls it F κκ (κ = “coupled”) /dodo/ ☞ a. b. /dado/ a. ☞ b. I DENT­BP[±low] [d1 o2 ][d1 o2 ] [d1 o2 d3 o4 ] [d1 a2 ][d1 o2 ] [d1 a2 d3 o4 ] R EDUP *! I DENT­BP[±low] *! R EDUP * (15) Allowing BP correspondence in spite of mismatches: R EDUP � I DENT­BP /dado/ ☞ a. b. /dedo/ ☞ a. b. [d1 a2 ][d1 o2 ] [d1 a2 d3 o4 ] I DENT­BP[±low] *! I DENT­BP[±mid] * I DENT­BP[±low] [d1 e2 ][d1 o2 ] [d1 e2 d3 o4 ] R EDUP R EDUP *! I DENT­BP[±mid] * 24.962—2 Mar, 2005 p. 4 • But a problem: what candidate would really win here? (16) Preventing total agreement with I DENT­IO /d1 e2 d3 o4 / a. [d1 1 e2 2 ][d3 1 o4 2 ] b. [d1 1 e2 2 d3 3 o4 4 ] c. [d1 1 o2 2 ][d3 1 o4 2 ] d. [d1 1 e2 2 ][d3 1 e4 2 ] I DENT­IO[±bk] I DENT­BP[±low] R EDUP ☞ I DENT­BP[±mid] * *! * * *! *! • Subscript = IO relations, superscription = surface correspondence (BP) relations • Crucially, IO relation holds over entire string (a la Inkelas & Zoll) • The basic scheme: pseudo­reduplicative correspondence is established only when the mis­ matches are small enough to be tolerable (that is, violate lower­ranked I DENT­BP constraints) • If I DENT­IO[f ] � I DENT­BP[f ], then pseudo­reduplicative correspondence will be undetec­ tive (doesn’t enforce any additional identity beyond what is necessary to establish correspon­ dence in the first place) (17) Using BP correspondence to block raising of nonfinal mids /todo­in/ ☞ a. b. c. d. e. f. I D ­IO[voi] [to][do]hin [to][du]hin [todu]hin [tu][du]hin [to][to]hin [do][do]hin /gaston­an/ ☞ a. b. c. d. I D ­BP[hi], I D ­BP[lo] I D ­IO[hi] /non­stem­final R EDUP *N ON F INAL M ID * *! *! *! *! *! I D ­IO[voi] [gas][tos]an [gas][tus]an [gastos]an [gastus]an I D ­BP[voi] * * * * * * I D ­BP[hi], I D ­BP[lo] *! *! I D ­IO[hi] /non­stem­final R EDUP * * *N ON F INAL M ID * *! I D ­BP[voi] * * * * • When I DENT­BP outranks some M, regular phonological process can be blocked (but only when conditions for BP correspondence are met) (18) An extremely important point • The blocking of raising when the word looks pseudo­reduplicated is only a statistical ten­ dency in Tagalog, not a hard and fast rule – Really R EDUP­ish words tend not to undergo (yoyo, todo; also many pseudoreduplicated native words, like dede and totoPo) – Moderately R EDUP­ish words go one way or the other, or vary – Other words tend to raise (though extremely recent or nonce loans may also resist it) • Lots of exceptions (esp. pseudo­reduplicated words that do raise); we’ll talk more about the analysis of exceptions later in the semester (19) What are we to make of this R EDUP constraint? • Why would words want to have otherwise unsupported reduplicative structure? • Plausible alternative: don’t care about having reduplicative structure per se, but when sylla­ bles are similar enough, speakers are driven to conclude that similarity is too great to be due to chance, and must involve correspondence (20) • How much does it take to conclude the similarity is “too great to be accidental”? 24.962—2 Mar, 2005 p. 5 • Tagalog provides abundant evidence for what reduplication should really look like. These aren’t right at all – E.g., tonto: would have to be toton or tonton – Similarly piloto: repeated syllable at wrong end of word, has otherwise unlicensed seg­ mental alternation – Words are also mostly too short to be reduplicated (native roots are minimally disyllabic) (21) English words “fixed” to enforce surface identity (Zuraw 2000) Standard (etymological) sherbet pompon orangutan hara kiri smorgasbord sacrosanct Inuktitut Abu Dhabi asterisk Google hits 62,900 17,700 55,600 11,100 71,500 39,500 23,500 135,000 613,300 “Fixed” sherbert pompom orangutang hari kari smorgasborg sancrosanct Inuktituk Abu Dhabu/Abi Dhabi asterist/askerisk Google hits 12,000 15,500 6,130 8,430 1,740 201 751 126/67 57/100 • OED etymology of orangutang: “In many of these languages, the second element utan ‘woods’ has been corrupted to jingle with the first” (22) Some other historical changes: • hotchpot > hotchpotch • maamajumbo (or something like this) > mumbojumbo • catapillar > caterpillar > callerpillar And lots and lots (and lots) of errors, of varying frequencies frustrum farmiliar aprostrophe perservere proprogate porpcorn cornucorpia crocrodile (also in French) heroric Figuerora sumblime frangrance (55,400 hits!!!) rhondodendron parenthensis/parenthenses gongonzola gonblins denfense/denfend fan(s)tan(s)tic nunchunks pumpkim axphyxiate Mitshubishi/Mitshubitshi andpersand/ampersamp andaconda arkvark/ardvard snapshop tollboth/toolbooth Monongahola Tuscaloosca Abbis Ababa • Ralph on The Simpsons: “I’m bembarrassed for you!” • Seemingly purely orthography examples: – mnemnomic, rhubharb, whirlwhind, Philiphines/Philliphines (23) Reduplication as a strategy to “fill up time” when you don’t really know the word very well • Kyracyra [kir@sir@] – http://pdaphonehome.com/forums/showthread/t­9761.html I actually removed the Palm and Kyracera software twice. Still can’t get a conntection. • gazpazo, okinomiyaki, • Yasfat Arafat/Yasfak Arafak (by a guy trying to remember Yasser Arafat’s name); Afrafat and Arafrat more common (24) What we see from English • Low­level tendency to enforce identity across remote parts of the word 24.962—2 Mar, 2005 p. 6 • Tendency is strongest in case the two halves of the word are very similar (orangutan, smor­ gasbord) • Also shows up when word not fully known/accurately retrieved The issue: • If speakers simply assign (pseudo)­reduplicative structure when they have observed similar­ ity that’s too great to be coincidental, why would we get the same effect with these “partially known words”? (25) Some conjectures • Zuraw’s R EDUP constraint is part of a more general bias that speakers have to establish cor­ respondences between elements—even if they are not extremely similar • Calculating these correspondence relations permits learning morphology, phonological al­ ternations, etc. – Start by assuming that A and B are the same item, and see whether the relation between them is a general one that holds of other pairs, too • The bias to assume a R EDUP relation may be stronger than others; reduplicated forms are often ambiguous (they could be sequences of accidentally similar syllables), and the learner needs to be predisposed to try to analyze them as reduplication, or else nothing would ever be learned as reduplication • R EDUP also acts as a form of data compression: – Learner wants to reduce unpredictability as much as possible; for each word, attempts to represent word as economically as possible – Of course, the segmental makeup of words is by and large unpredictable (it just has to be listed). But two things are predictable: ∗ Epenthetic segments ∗ Copies of segments that are listed elsewhere – Epenthesis can only get you so far in a vocabulary of realistic size, but R EDUP could greatly reduce the amount you need to list: ∗ [tono] = /t1 o2 [+nas]1 X2 / (schematic; not meant to be a concrete proposal) � � ∗ But how far does this extend? Is aardvark underlyingly /X2 X3 +cor −dors +voi 4 v1 a2 r3 k4 /? ☞ It is an open question whether these factors need to be encoded as a grammatical constraint that is rankable (as Zuraw does), or whether they could exert a force as learning biases, that determine how words are represented lexically. Agreement by correspondence Rose and Walker (2004) A typology of consonant agreement as correspondence. Language 80, 475­531 (26) Aari (data cited by Rose & Walker, p. 482) Verb giP duuk sug naS tS’aaq Saan Zaag Causative giP­sis duuk­sis sug­zis naS­SiS tS’aaq­SiS Saan­SiS Zaag­ZiS Gloss ‘hit’ ‘bury’ ‘push’ ‘like’ ‘curse’ ‘urinate’ ‘sew’ 24.962—2 Mar, 2005 p. 7 • Standard analysis: spreading • The problem: how do segments affect one another at a distance? (27) The observation: consonant harmony typically targets a few particular classes of segments • liquids/nasals • sibilants • anterior segments These classes of segments seem to involve sets of very similar consonants. (28) Building in the intuition that similarity drives harmony: • C ORR­C↔C: consonants should stand in a correspondence relation to one another • Perhaps a universal ranking based on similarity : C ORR­T↔T (identical stops) � C ORR­T↔D (same place) � C ORR­T↔K (same voicing), etc. . . (29) This idea is very similar to suggestion in (19) above: if you’re already similar, you get put into a correspondence relation (which might make you get even more similar) (30) A parallel between harmony and reduplication: locality • Reduplicants tend to take material from neighboring part of the base • Consonant harmony is stronger when the relevant consonants are in adjacent syllables “It’s good for adjacent things to be identical” (***if they are in a correspondence relation) (31) How can we interpret this correspondence relation? (compare possible interpretations of R EDUP above.) Can it explain the English “error” data?