Modules interfaces and some consequences thereof Tobias Scheer

Modules, interfaces and some consequences thereof Tobias Scheer Abralin ao Vivo 14 November 2020

Today's programme Chapter 1 Modularity Chapter 2 Translation: lexical or computational? Chapter 3 2. 1. In language 2. 2. Computational translation can't be right 2. 3. Arguments against lexical translation? Are there several translational mechanisms? Chapter 4 Consequence #1: Phonetic arbitrariness Chapter 5 Consequence #2: Substance-free phonology Chapter 6 Interlude #1: Phonology-phonetics mismatches and what they tell us about spell-out Consequence #3: No naturalness Chapter 7 Interlude #2: Is all of phonology arbitrary? Consequence #4: No diacritics (aliens in phonology) Chapter 8 Consequence #5: Hierarchy expressed as trees or lateral relations. Phonology is flat

Chomsky (1965) et passim, overview Boeckx & Uriagereka (2007) modules: the inverted T Lexicon 1 morpho-syntax (synt. items) Lexicon 2 (phonol. items) Lexicon 3 (phonet. items) phonology phonetics physiology concatenation (generative) semantics interpretative (no concatenation) (grammatical) computation

modules: psycho-linguistic literature Lexicon 1 lemmas Lexicon 2 lexemes Bock (1982), Levelt et al. (1999), Cohen -Goldberg (2013), Indefrey (2011) grammatical encoding phonological encoding phonetic encoding (grammatical) computation

Chapter 1 Modularity

modularity Modularity: Robbins (2017), Gerrans (2002), Carruthers (2006), Segal (1996) Phrenology: Sabbatini (1997), van Wyhe (2004) Connectionism: Smolensky (1988, 2003) • the mind/brain is made of a number of distinct functional units that carry out computation, each being competent for a specific task (Fodor 1983). • roots in 19 th century phrenology (F-J Gall), which already was about the mind and the brain • competitor: Connectionism (Rumelhart & Mc. Clelland 1986). Holds that computation is colourless, i. e. not specialized for specific tasks.

Fodor (1983: 58 -64), Coltheart (1999), Carruthers (2006: 3 -12), Collins (2017: 224), Jackendoff (1997: 41, 2002: 218 -227). modularity domain specificity • modules are only competent for their own domain • they work on their own proprietary vocabulary • this vocabulary defines their domain most prominent split in language: syntax, morphology, semantics phonology number labiality person stopness animacy voicing mood palatality voice sonorancy quantification tone

modularity granularity • how do we identify modules? • are they larger or more fine-grained units? as inquiry proceeds, relevant items are getting smaller (like elsewhere in science) • vision • Fodor was talking about vision as a module • vision is now thought to fall into a number of distinct computational systems are specialized in shape, colour, motion, form, face recognition or contrast (Marr 1982 and following, overview in Stevens 2012) • language • Fodor was talking about language as a module • language may fall into distinct computational systems specializing in syntax, morphology, phonetics, pragmatics. • and maybe phonology subdivides into distinct modules (sonority, place, laryngeal properties)

modularity Robbins (2017) encapsulation when carrying out their computation, modules have only access to a small subset of the total information that is present in the organism. classical illustration: optical illusions • even if you know that both horizontal lines have equal length, you will still perceive a difference in length. • conclusion: general conscious knowledge (Fodor's Central Systems) has no bearing on vision • which module exactly has access to what kind of information depends on its wiring: the Mc. Gurk effect shows for example that phonology has an input from vision, but not from general conscious knowledge. Müller-Lyer

modularity Mc. Gurk effect Hearing lips and seeing voices Mc. Gurk, Harry & John Mac. Donald 1976. Hearing Lips and Seeing Voices. Nature 264: 746 -748. 1936 -1998 audio signal b g d Mc. Gurk fusion perception visual signal audi-visual case of a dichotic effect

modularity encapsulation: common misunderstanding based on the fact that Fodor and most of the ensuing literature has left intermodular communication unexamined (Jackendoff 2002: 218 ff). encapsulation inhibits exchange between modules • which are insulated and cannot really talk to one another. • but we know that everything exchanges information with everything all the time, so modularity with its autistic modules can't be right.

modularity intuition? • this appears to be a rather intuitive reaction on encapsulation • rather, we may want to ask whether the evidence presented is right or wrong. encapsulation may be an opportunity • to find out how exactly computational systems are wired • that is: X feeds Y, but not Z, and A receives input from B, but not from C • to find out how exactly information exchange works when two modules communicate

modularity encapsulation #2 a different view (Jackendoff 2002: 219) [not incompatible with the other view] once computation has begun in a given module • nothing can leave the module before the modular computation is completed (no intermediate steps can break out) • no new information can join (nothing can break in) module input output this is called inclusiveness in syntax: Chomsky (1995: 228).

consequence #1 modularity enforces translation when two modules communicate, how exactly does information exchange work? • since different modules do not speak the same language, they cannot interpret the vocabulary of their neighbours • if they want to exchange information, the vocabulary of the sending module needs to be translated into the vocabulary of the receiving module syntax, morphology, semantics phonology number labiality person stopness animacy translation voicing mood palatality voice sonorancy quantification tone

Chapter 2 Translation: lexical or computational? 2. 1. In language

lexical vs. computational translation best-studied interface in language: morpho-syntax - phonology morphemic vs. non-morphemic ingredients in the input to phonological computation • morphemic • belongs to a morpheme • is stored in the lexicon • is not predictable from morpho-syntax • a root, an affix: learn, -ed, im-, etc. • non-morphemic • does not belong to a morpheme • is not stored in the lexicon • is predictable from morpho-syntax • represents morpho-syntactic information • successively: - juncture phonemes - SPE-type diacritics #, +, !, etc. - autosegmentalized versions thereof: ω, φ, etc. (Prosodic Hierarchy)

lexical vs. computational translation ils [z] ont une grande maison they have a big house ont-ils *[z] une grande maison ? do they have a big house? • morphemic information: ils, ont, une, etc. • non-morphemic information: the syntactic relationship of the words • ils ont liaison obligatory • ont-ils ## une liaison impossible

lexical vs. computational translation spell-out ↔ morpho-syntax table chat ils livre tu lexical translation • list-based (dictionary) • match • items are stored and learned phonology ont- une grande maison ? DP → ## computational translation • input-output • transforms X into Y through an algorithm • rules are stored and learned

lexical vs. computational translation SPE: Chomsky & Halle (1968: 12 f, 366 ff) Prosodic Phonology: Selkirk (1981, 1986, 2011), Nespor & Vogel (1986) all theories of the interface have this division of labour: • morphemic information comes to phonology through • non-morphemic information comes to phonology through lexical translation computational translation • SPE #-distributing algorithm (major categories N, V, A and projections thereof are flanked by #. • Prosodic Phonology • mapping rules: DP → ω • then mapping (or more recently match) constraints: Align (syntactic item, phonological item)

lexical vs. computational translation is theorized by Ray Jackendoff • first as translation rules (Jackendoff 1987), then as correspondence rules (Jackendoff (1997), then as interface processors (Jackendoff 2002). • Jackendoff has three types or modules, which he calls processors: • inferential processors Fodor's Central Systems: higher cognitive functions • integrative processorscompute content: vision, syntax, phonology etc. • interface processors translate information among the two other types • "Each module is strictly domain-specific in Fodor's sense: integrative and inferential pro-cessors deal with only one level of structure each; interface processors deal with two (we might therefore want to call them 'bi-domain-specific')" (Jackendoff 2002: 2002)

lexical vs. computational translation Jackendoff's (2002: 199) diagram representing the modular structure of language

2. 2. Computational translation can't be right

computational translation can't be right Scheer (2012: § 161 ff, 2020) computational translation is incompatible with modularity • it transforms the vocabulary of module X into the vocabulary of module Y • this transformation occurs within a module • this module thus needs to know and to be able to parse more than one vocabulary. • this is exactly what domain-specificity rules out • Jackendoff's bi-domain specificity merely puts a word on a contradiction in terms bi-domain specificity defies the purpose of interfaces • if modules were able to process two different types of vocabulary, there would be no need for specific translation devices in the first place • if modules were able to be bi-domain specific, they could do the interface job themselves. note on OT constraint-based mapping (Align) is worse than all other types of computational translation since it locates translation IN the phonology.

computational translation can't be right non-morphemic translation must also be lexical • morphemic information • non-morphemic information ↔ ils ↔ # all pieces of the string that is submitted to phonological computation • originate in the lexicon • have come into being through a dictionary-based match with the output of morpho-syntactic computation

2. 3. Arguments against lexical translation?

arguments against lexical translation? Lackner (1981), Lackner & Di. Zio (2000) issue #1 multisensory input when a given module receives input from two or more different sources. • Mc. Gurk (dichotic perception): phonology takes in information from sound and vision [this example puzzled Fodor 1983: 47 and was used to cast doubt on domain specificity, see the discussion on the Faculty of Language blog, post Dueling Fodor interpretations (permalink) by Bill Idsardi, 3 April 2019] • sense of body position and orientation: input from • stretch receptors in the muscles • tactile information (skin) • vestibular system in the inner ear: semicircular canals (rotational movements), otoliths (linear acceleration) • vision, audition • this does not mean that modules can parse several vocabularies (are bidomain specific). • it just means that multiple translation has occurred, from various different vocabularies into the one proprietary vocabulary of the receiving module

arguments against lexical translation? issue #2 infinitude of processing is not an argument against, but in favour of lexical translation infinitude of the possible (Gallistel & King 2010: xi) • modular computation may produce an infinite set of outputs • viz. language: Chomsky's (1988 et passim) discrete infinitude, i. e. the ability to produce an infinite number of sentences based on a finite number of stored vocabulary items. • this cannot be done by look-up tables or dictionaries: only an algorithm qualifies.

arguments against lexical translation? argument? Bill Idsardi argues that this undermines lexical translation: "spell-out via lookup table is literally the weakest possible architectural assumption about transduction": a last resort strategy (Post More on "arbitrary" (permalink) on the Language Faculty blog, 29 March 2019) no, this is a misunderstanding: • infinitude of the possible is about modular computation, not about transduction (= translation) • modular computation may produce infinite results, but translation may not • this follows from the fact that the output of translation is stored: it is some vocabulary item pertaining to the receiving module. support for a lexical mechanism • the fact that the number of items involved in translation is necessarily finite supports the idea that no algorithm is at play. • it indicates that translation is lexicon-based.

arguments against lexical translation? issue #3 is lexical translation the odd man out? Following Jackendoff (1997: 107), Bill Idsardi argues that the lexical translation we see in the morpho-syntax - phonology interface is the odd man out in the cognitive system (Post More on "arbitrary" (permalink) on the Language Faculty blog, 29 March 2019). "If we look at the rest of the brain, we do not immediately find anything with these same general properties. Thus the lexicon seems like a major evolutionary innovation, coming as if out of nowhere" Jackendoff (1997: 107) thus the empirical question • is there evidence beyond the morpho-syntax – phonology interface that lists of items are matched when two modules exchange information? • answer: yes of course, there are cases all around where intermodular communication descriptively implies finite lists whose members are matched.

arguments against lexical translation? cases where intermodular communication is based on lists whose members are matched • phonology-phonetics mapping (in language production) a list of phonological items (phonemes, or melodic primes: features) is matched with a list of phonetic categories (acoustic target values in Boersma & Hamann's 2008 Bi. Phon model).

arguments against lexical translation? 5 senses: real world contiunua mapped onto cognitive categories • categorical perception • color perception: a gradient spectrum of wave lengths is discretized into a finite number of distinct colors that people perceive. The slicing of the real-world spectrum into cognitive categories (colors) may be more or less finegrained according to culture, language (Sapir. Whorf) or expertise. But the number of cognitive categories is always finite, and they always correspond to a real-world item, i. e. a range of wave lengths. • the same goes for the other senses: odor, taste, sense of touch, audition. Cat. perception: Harnad (2003, 2005) Color perception: Grieve (1991), Choudhury (2014: 144 -184), Whittle (2003), Webster (2003), Hansen et al. (2006), Winawer et al. (2007), Athanasopoulos (2009) wave length (nm) [real world] 380 -450 450 -480 480 -510 510 -550 550 -570 570 -590 590 -630 630 -750 color associated [cognitive category] violet blue-green yellow-green yellow orange red Choudhury (2014: 7)

arguments against lexical translation? it thus seems that the question is not whethere is lexical translation elsewhere, but rather whethere is evidence for intermodular communication that is not lexical, i. e. does not match items pertaining to two lists of distinct vocabulary.

Chapter 3 Are there several translational mechanisms?

are there several translational mechanisms? summary • we have been looking at the interface in language that is best studied and best understood: morpho-syntax - phonology • nobody doubts that it involves lexical translation • there is reason to believe that this interface reduces to lexcial translation • lexical translation also exists elsewhere in language (phonology - phonetics), and elsewhere in the mind (5 senses discretizing real world continua)

are there several translational mechanisms? generality of lexical translation • there is only one type of module: input → algorithm → output • the zero hypothesis is that there is only one translational mechanism that allows modules to communicate. all translation among modules may be lexical in kind • are there other candidates known? • yes, computational translation, but we dismissed that. • other than that? Not as far as I can see.

are there several translational mechanisms? Scheer (2012, 2014, 2020) generalized lexical translation • each module is preceded by a lexicon • whose output are its own vocabulary items • whose input comes from different sending modules acousticphonetic vision (Mc. Gurk) phonological lexicon a ↔ p v ↔ p module A module B phonology module C lexicon of module D A ↔ D B ↔ D p ↔ D p ↔ E C ↔ E lexicon of module E module D module E

Chapter 4 Consequence #1: Phonetic arbitrariness

lexical translation is arbitrary arbitrariness • an inherent property of list-based mapping is that the relationship of the items matched is arbitrary. • in a language dictonary, there is no reason why a word appears as X in one language but as Y in another language (rather than as Z), and so on: table (English) ↔ stół (Polish) past tense (morpho-syntax) ↔ -ed (phonology) 450 -480 nm (wave length) ↔ blue rather than chair, or cat rather than 3 sg, or rather than -s, -a, etc rather than 450 -520 nm • the arbitrariness of Vocabulary Insertion (morpho-syntax - phonology interface) is so obvious that it is never argued for because: it goes without saying. • at the lower interface of phonology, though, arbitrariness is anything but obvious or intuitive.

Chomsky (1965) et passim, overview Boeckx & Uriagereka (2007) modules: the inverted T Lexicon 1 morpho-syntax (synt. items) Lexicon 2 (phonol. items) Lexicon 3 (phonet. items) phonology phonetics physiology semantics (grammatical) computation

phonology-phonetics interface post-phonological spell-out • if intermodular communication is lexical, there is a spell-out operation whereby phonological categories are matched with phonetic categories. • in Boersma & Hamann's (2008) Bi. Phon model, segments are matched with an ERB value, that is with an acoustic target s ↔ 20, 1 ERB (Equivalent Rectangular Bandwidth) "s is realized with a spectral mean of 20, 1 ERB" problem • the phonetic properties of s (20, 1 ERB) are not arbitrary • but the association of s with these properties must be arbitrary. therefore • s cannot be a phonological item • it must be a phonetic item

phonetic arbitrariness • the relationship between phonological and phonetic categories is arbitrary. • s, k, [labial], [continuant] etc. are phonetic categories. They do not exist in the phonology. • what phonologists call [labial] is not labial in any way in the phonology. • the use of [labial] is shorthand for a colourless phonological prime α whose phonetic correlate is labial, but which could have any other phonetic correlate as well. • phonology is made of primes α, β, γ whose purpose is • to express contrast • to allow (phonological) computation to be carried out • spell-out relates α, β, γ with phonetic categories α β γ ↔ X ERB ↔ Y ERB ↔ Z ERB

Chapter 5 Consequence #2: Substance-free phonology

substance-free phonology • there is no phonetics (s, labial, continuant etc. ) in the phonology. • phonetic categories come into being upon spellout. • substance-free phonology • was introduced by Hale & Reiss (2000) • today cuts through all theories (OT, GP, rule-based) • is arrived on a number of pathways Hale & Reiss (2000, 2008), Hale & Kissock (2007), Volenec & Reiss (2018, 2019) Boersma (1998: 461 ff, 2011), Boersma & Hamann (2008), Hamann (2011, 2014), Mielke (2008), Carvalho (2002), Odden (2006, 2019), Blaho (2008), Samuels (2011 a, b, 2012), Iosad (2012: 6 ff, 2017), Scheer (2014), Chabot (2019), Dresher (2014, 2018). Summary in Scheer (2019), forthcoming special issue of the Canadian Journal of Linguistics. • one pathway is through modularity: • post-phonological spell-out must be arbitrary • therefore phonological items must be devoid of phonetic properties: α, β, γ.

substance-free phonology acquisition • children are born with nothing but the ability to categorize and the predisposition to build categories that are phonological in kind. • phonological categories are thus "emerging", i. e. are shaped exclusively by the environment. • that is, the number of primes and their association to phonetic categories depend exclusively on the contrast and computation found in the target language. • [this view on acquisition is not endorsed by Hale-Reiss-Kissock-Volenec. ]

substance-free phonology • there is nothing wrong with using "labial" when talking about phonology: this is just shorthand (and mnemonically efficient) for "the phonological prime that is spelt out as labial". • black holes are neither black nor holes, but astronomers use the shorthand. • chemists talk about water when they mean H 2 O. • below is how phonologists talk about cat in increasing explicitness when a. IPA k æ t b. segmental c. infrasegmental O N O N | | | k æ t U A A | | | ʔ I ʔ | | H H d. primes and spell-out O | α | γ | ε N | β | δ O | β | γ | ε N α β γ δ ε ↔ u ERB ↔ w ERB ↔ x ERB ↔ y ERB ↔ z ERB

Interlude #1 Phonology-phonetics mismatches and what they tell us about spell-out

phonology-phonetics mismatches Hamann (2014) prediction • if the association of phonological and phonetic categories is arbitrary, the sound system should behave in such a way that items have a given ponological behaviour that does not correspond to how they are pronounced. • but this is strongly counter-intuitive: phonological [labial] is also pronounced as labial. • that may be true in many or 95% of the cases, but phonology-phonetics mismatches exist all over the place. Often they are so trivial that nobody talks about them.

phonology-phonetics mismatches Chabot (2019) Cyran & Nilsson (1998) Henton (1983), Harrington et al. (2008) examples Dressler (1981), Gussmann (1998) • r ↔ ʁ/χ, r ↔ ʃ/ʒ the phonological sonorant r appears as ʃ/ʒ in Polish, as ʁ/χ in French and German. We know it is a sonorant because it qualifies as a second member of a branching onset. • w ↔ v the phonological sonorant w has an obstruent realization v in a number of Slavic languages. We know it is a sonorant because preceding obstruents do not undergo voice assimilation. • uu ↔ ii uu-fronting in South-East British English: boot is pronounced [biit], but [ii] continues to trigger w-insertion in do it [dii w it], thereby attesting its phonological identity /uu/. • ŋg ↔ ŋ agma [ŋ] in English and German is a singleton on the surface but represents the partial geminate /ŋg/: diagnostics are the fact that it occurs only after short vowels and is absent word-initially.

phonology-phonetics mismatches Chabot (2019) a window on how spell-out works • the existence of phonology-phonetics mismatches, i. e. the arbitrary association of phonological and phonetic categories upon spell-out, is thus a documented fact. • the sound system is able to do that. • the only limit is acquisition: if children have no evidence for lexicalizing something different from the surface, they won't. In uu-fronting, if there were no w-insertion, children would not lexicalize uu ↔ ii, but ii ↔ii. • therefore cases of non-mismatch are accidentally non-mismatching. They could as well be mismatching. • why are 95% of the mappings faithful?

Chapter 6 Consequence #3: No naturalness

no naturalness • when you pour a bucket of water over someone's head, it is • natural that this person is wet • unnatural that this person is dry n → ŋ /__k k → ʦ / __i Old English natural rule n → ŋ / __p k → s / __i Modern English i → u / d__ Southern Pomoan r → ŋ / __i unnatural rule crazy rule

no naturalness NP: Stampe (1972), Donegan & Stampe (1979) NGP: Vennemann (1974), Hooper (1976) Naturalness in phonology: Chabot (2019, forth) • naturalness has played an important role in phonology, always in the sense that unnatural processes or workings in a theory should be disqualified because phonology is natural the major challengers of SPE in the 70 s were Natural Phonology and Natural Generative Phonology, followed by Anderson's (1981) article "Why phonology isn't natural" • naturalness is phonetic by definition it is not even possible to talk about what is natural and what is not without making reference to the surface

no naturalness • phonological naturalness is a contradiction in terms the illusion is prompted by the presence of phonetics ([labial] etc. ) in the phonology • phonology is not natural or unnatural, phonology is • if you take out the illusion, i. e. if you empty phonology from phonetic properties, phonological naturalness disappears • if phonology is made of α, β, γ etc. , it isn't even possible to talk of anything being natural or unnatural: α → β / __ γ isn't any more or less natural than β → γ / __ α because α, β, γ have no substance. • recall that phonology has no idea of how α, β, γ etc. will eventually be pronounced: substance only comes in upon spell-out.

no naturalness Hale & Reiss (2000, 2008), all of Substance-free Phonology since then. • phonology does not object against computing anything and its reverse: • anything can be transformed into any other thing and its reverse X → Y is just as good as Y → X • any process may occur in any environment and its reverse X → Y / __A is just as good as X → Y / __B • By the standards of phonology, all processes are equally probable or well-formed: n → ŋ /__k k → ʦ / __i Old English n → ŋ / __p r → ŋ / __i k → s / __i Modern English i → u / d__ Southern Pomoan

no naturalness summary double arbitrariness due to the absence of phonetics in phonology (α, β, γ) computation of phonological items pronunciation of phonological items α → β β ↔ [X] "language may be said to be a algebra" Saussure (1916: 168) arbitrary but processes observed are 95% natural but spell-out is 95% natural how come?

no naturalness Baudouin de Courtenay (1895), Vennemann (1972), Bermúdez-Otero (2007, 2014) • the source of naturalness lies outside of phonology • it originates in the area where it makes sense: in phonetics. • the life-cycle of phonological processes • phonetic variation (co-articulation etc. ) • phonetic precursor • phonologization of a phonetic precursor: phonological rule • morphological conditions on the rule appear • lexcial exceptions appear • the rule is levelled out and dies • rules are regular and natural when they are born in the phonetics • this regularity and naturalness is carried over into the phonology upon phonologization.

no naturalness Bach & Harms (1972), Scheer (2014), Chabot (2019) • phonology is fine with inherited naturalness, but it is just as happy when this naturalness is destroyed in diachronic evolution, following the life-cycle. • rules are not born crazy, they become crazy through aging • example from English: velar softening when inherited from French in the 11 th century: k → ʦ / __i loss of affricates k → s / __i unnatural • an event in the evolution of the language that is entirely unrelated to the rule, deaffrication, makes the rule unnatural. • but phonology cannot see the difference and does not object. • the diachronic evolution is in the spell-out, not in the phonology: • α+γ ↔ ʦ > α+γ ↔ s

Interlude #2 Is all of phonology arbitrary?

is all of phonology arbitrary? • no • arbitrariness originates in the colourlessness of α, β, γ • if phonology reduced to α, β, γ, it would be entirely arbitrary • but it doesn't: α, β, γ inhabit the lower part of a phonological representation, i. e. the area below the skeleton. • in the area at and above the skeleton, we find • syllable structure: nuclei, onsets, codas etc. • timing units: x-slots, morae, CV units • structure managing stress: feet, grid marks • structure importing morpho-syntactic information: the Prosodic Hierarchy • tone

Scheer (in press) is all of phonology arbitrary? • are items at and above the skeleton also colourless α, β, γ's ? ==> No σ σ O | x | α | γ | R | N | x | β C | x | γ O | x | δ | γ | R | N | x | β | α items • are phonologically meaningful • have no phonetic correlate • computation is not arbitrary items • are phonologically meaningless • have a phonetic correlate • computation is arbitrary • phonologically meaningful items • belong to a closed set: identical for all languages, innate (while α, β, γ emerge) • are not interchangeable: a nucleus cannot be an onset (but an α can be a γ) • have immutable properties: a coda bears on a nucleus in a way an onset does not

is all of phonology arbitrary? Bach & Harms (1972), Buckley (2000, 2003), Hyman (2001), Vennemann (1972), Chabot (to appear) Scheer (2015: 333 f) this shows in computation: • crazy rules are only ever segmentally crazy • in a crazy rule A → B / C, i → u / d__ Southern Pomoan A, B and C are only ever items that occur below the l → ʁ, ʕ / V__V Sardinian skeleton • inventory of crazy rules mentioned in the sparse literature on the subject • there do not appear to be crazy rules that manipulate items at and above the skeleton: • compensatory shortening • closed syllable lengthening • tonic shortening • anti-Latin stress (stress falls on the antipenultimate syllable except when the penultimate syllable is short, in which case this syllable is stressed) • of course the sample is incomplete and there may be syllabically or stress-wise crazy rules out there. • but there is no reason for all documented cases to accidentally concern only melodic properties.

Scheer (in press) is all of phonology arbitrary? two phonologies: • distinct vocabulary • hence distinct modules O | x | α | γ σ | R | N | x | β C | x | γ O | x | δ | γ σ | R | N | x | β | α phonologically meaningful items, therefore non-arbitrary computation phonologically meaningless items, therefore arbitrary computation

Chapter 7 Consequence #4: No diacritics (aliens in phonology)

aliens in phonology recall • one difference between lexical and computational translation is that non-morphemic information is the result of computation in the latter option, but originates in the lexicon in the former. • under lexical translation, all objects present in the input string to phonology are stored in the lexicon • therefore they must be members of the phonological vocabulary • only phonological vocabulary can be stored • only phonological vocabulary can be parsed in phonological computation

structuralism, SPE Selkirk (1981, 1984, 1986, 2011), Nespor & Vogel (1986) etc. aliens in phonology what is a diacritic? • #: an arbitrarily chosen typewriting symbol. Banana �� would be just as fine. • autosegmentalized version thereof: the Prosodic Hiararchy measure of alienness #1 • ω, φ, Cl, IP, U are the projection of nothing: they come from the outside. • ω | Ft | σ se this is unheard of not just in phonology, but in language as IP such ω | Ft | σ ti φ Cl ω Ft σ σ σ manderanno σ Cl | ω | Ft σ subito σ ω | Ft | σ le φ Cl ω | Ft σ σ σ ultime U Cl | ω Ft σ Ft σ σ cartoline, σ projection from nothing projection from lower items Nespor (1999: 123)

aliens in phonology Scheer (2008, 2011, 2012), Samuels (2011 a, b), Pak (2008) measure of alienness #2 • if they were not aliens, what kind of vocabulary would ω, φ, Cl, IP, U belong to? • certainly not the α, β, γ's of the lower part of phonology • they are not interchangeable • they occur above the skeleton • are they like onsets, nuclei, feet, grid marks etc. ? • no: all items above the skeleton are projections of some lower item, they are not. • no: they never play a role when phonology is by itself, i. e. in processes that do not involve morpho-syntactic information. Phonology is by itself: • in a palatalization k → ʧ / __i, e items involved: α, β, γ • in a compensatory lengthening items involved: onset, nucleus, coda • in stress assignment items involved: feet, grid marks • all other items above the skeleton are invovled when phonology is by itself

aliens in phonology Scheer (2009, 2012) measure of alienness #3 ω, φ etc. are sleepers • unlike all other vocabulary items either below or above the skeleton, their mere presence in a representation has no effect • [palatal] contributes palatality • a nucleus contributes vowelhood • but the existence of # or ω, φ in the phonological string does not influence the course of phonology in any way. • they only have an effect when they are accessed by some phonological rule/constraint: "process X applies within ω/ before #". • the reason for that is • that they are aliens • they are phonologically meaningless (which cannot exist above the skeleton) • their only purpose is to store morpho-syntactic information.

aliens in phonology Scheer (2009, 2012) measure of alienness #4 arbitrary effects: #, ω, φ etc. can produce any effect and its reverse • at the beginning of a word = after # or at the left edge of ω, • only TR (obstruent-liquid) can exist attested • only RT can exist does not exist in natural language • in #C__C, • insert a V • delete the V attested does not exist in natural language • why is the effect of #, ω, φ arbitrary? • because they are phonologically meaningless • how would you calculate the effect of banana �� on phonological computation? • language does not work like that • effects of the beginning of the word are not random • they are cross-linguistically stable

aliens in phonology Scheer (2009, 2012) cross-linguistically stable effects of the beginning of the word 1. word-initial clusters in some languages initial clusters are restricted to #TR. In others they have the same distribution as internal clusters. But there is no language where they are restricted to #RT (#TT, #RR). 2. strength of word-initial consonants in some languages word-initial consonants are especially strong. In others, they do not have any peculiar behaviour regarding strength. But there is no language where they are especially weak. 3. deletion of the first vowel of the word in some languages the first vowel of words is unable to alternate with zero. In others it does not show any peculiar behaviour when compared to other vowels. But there is no language where non-initial vowels are unable to alternate with zero, while initial vowels do.

aliens in phonology Zwicky & Pullum (1986), Scheer (2011) Lowenstamm (1999), Scheer (2009, 2012, 2014) carriers of morpho-syntactic information in phonology are not diacritics (#, ω, φ etc. ) but • truly phonological items • that are stored in the lexicon • and bear on phonological computation by there mere presence consensus morpho-syntax and the area below the skeleton are incommunicado in both directions: • [labial] etc. is never a carrier of morpho-syntactic information in phonology • syntactic movement is never triggered by [labial] etc. (Phonology-free syntax) the initial CV • the only carrier of morpho-syntactic information in phonology is syllabic space. • depending on your theoretical inclination: • empty CV units • x-slots • morae

aliens in phonology Scheer (2009, 2012, 2014) workings of the initial CV (in Strict CV) • beginning of the word = an empty CV unit • the nucleus of the initial CV must be governed • therefore the first nucleus of the root must be contentful: two empty nuclei in a row are ill-formed • is well-formed as it stands gvt • deletion will make the string ill-formed • is ill-formed as it stands • inserting a V makes it well-formed beginnning of the word C V - C V | | C V C | C V | V C C | C V | V V - C | C V V

aliens in phonology Scheer (2009, 2012, 2014), Ségéral & Scheer (2008) three birds with one stone recall the three cross-linguistically stable effects of the beginning of the word • word-initial clusters are restricted to TR or unrestricted • word-initial consonants are either strong or unmarked • the first vowel of the word may either be immune against deletion or be deleted the initial CV and the lateral relations of Strict CV • predict all three effects word-initial clusters word-initial consonants first vowel of the word initial CV present TR strong immune initial CV absent unrestricted unmarked deletable

Chapter 8 Consequence #5: Hierarchy expressed as trees or lateral relations Phonology is flat

hierarchy Linguistic structure is hierarchical • but is there only one implementation of hierarchy, trees? • this is the default assumption, which is ubiquitous across theories • there is reason to believe that hierarchy in morpho-syntax and hierarchy in phonology are not the same. • modularity allows for different hierarchies governing different modules

Kaye et al. (1990), Kaye (1990), Harris (1994), Lowenstamm (1996), Scheer (2004) hierarchy Scheer (2012: § 9) reason #1 • theory-internal • Government Phonology has progressively curtailed syllabic arborescence • replacing its function by lateral relations (government and licensing) • last step: Strict CV where phonology is flat illustration of this deforestation: Czech loket "elbow" gvt O N O N | | | l o k e t e Nom sg loket Gen sg lokt-e O | l σ | R | N | o O | k σ | R | N | e Gen sg lokt-e Nom sg loket O C | t σ | R | N | e • e present because ungoverned • e absent because governed • e present because followed by a coda • e absent because followed by an onset

hierarchy Scheer (2013) support: concatenation morpho-syntax: retrieves pieces from long term memory and concatenates them phonology: does not concatenate anything no concatenation, no trees • in morpho-syntax, trees are the consequence of concatenation, and of nothing else. • Merge is the essence of the universal hierarchy- and tree-creating mechanism • if phonology does not concatenate anything, there cannot be any tree-building device in this module.

hierarchy Neeleman & van de Koot (2006) evidence phonology does not exhibit any of the properties predicted by tree-type hierarchy • trees have certain formal properties that make predictions regarding the type of phenomena that should be found in a tree-bearing environment: • projection • long-distance dependencies • recursion • Neeleman & van de Koot (2006) demonstrate that phonological phenomena do not display any of these properties. • they therefore conclude that the presence of trees in phonology overgenerates: arboreal structure predicts things that are absent from the record.

hierarchy Scheer (2013) more support: recursion a long-standing observation: there is no recursion in phonology N. B. : there are no recursive phenomena, but there all kinds of recursive analyses of nonrecursive phenomena. • recursion is formally defined as a node that is dominated by another node of the same kind. • there can be no recursion in absence of a tree-building device. therefore • no concatenation, no trees • no trees, no recursion • discussion in van Oostendorp's (2010) Sigma strikes back

hierarchy Scheer (2013) input conditions to computation morpho-syntax: phonology: no linear order how could hierarchy be expressed in an environment • that has no trees • but linear order ? • by lateral relations • lateral relations are the kind of hierarchy prompted by linearity.

congratulations, you made it all through to see the last slide Thank you for your attention