Clements (1985): the geometry of phonological features #
Clements replaces the SPE feature matrix by a tree of class nodes: a root node
linked to the CV tier, a laryngeal and a supralaryngeal node below it, a manner and a
place node below supralaryngeal, and each terminal feature hanging from one class node
(his (3) and Appendix; §4 gives the content of each node). Assimilation is the spreading
of one node ((5), after Mohanan), so the theory admits three types, total (the
root), partial (a class node) and single-feature, and a rule that must reach two
features with no common class node below supralaryngeal is total supralaryngeal
assimilation, which is why the SPE rule (14) with [αnasal] in place of [αanterior]
is unattested. The evidence is English coronal place assimilation ((10)–(13)), Icelandic
preaspiration ((6)–(7), after Thráinsson), Klamath lateral rules ((8)–(9), after
Barker), Sierra Popoluca nasalisation and release ((15)–(20), after Elson
and Steriade's Shared Features Convention) and the Kikuyu and palatalisation cases
of §5, and §4 splits the place features into a primary set P, present in consonants and
vowels, and a secondary set S, normally absent from plain consonants. The tree is an
instance of Phonology/FeatureGeometry.lean; its manner node, which the paper itself
flags as possibly superfluous, is dropped by the later geometries formalized here
(Studies/Sagey1986.lean, Studies/HalleVauxWolfe2000.lean).
Main definitions #
Node,Node.pred,classNode?— the five class nodes of (3), the tree, and the class node each terminal feature hangs from (§4); together theFeatureGeometryinstance.PlaceSet,placeSet?,PlaceSet.features— the primary/secondary split of the place features (§4).Spreading,Spreading.features,Spreading.apply— what one rule may spread, the root, a class node or one feature ((5)), and the assimilation it effects.rule12,rule14,rule14Nasal— the English rule (12) as a local rewrite rule copying the place class, its SPE statement (14) copying[anterior]and[distributed], and the[αnasal]variant the paper contrasts with it.
Main results #
supralaryngeal_subset_of_nasal_of_distributed,apply_eqOn_supralaryngeal_of_nasal_of_distributed— every spreading that reaches both[nasal]and[distributed]is total supralaryngeal assimilation, the argument against the[αnasal]variant of (14);eq_root_of_continuant_of_voiceis the Kikuyu case of §5.1 andsupralaryngeal_subset_of_anterior_of_continuantthe palatalisation case of §5.2.rule12_n_θ,rule12_t_θ,rule12_d_θ,rule12_n_esh— tenth, eighth, hundredth, insure from table (11): the target takes the trigger's place features and nothing else;rule12_n_pandrule12_s_θare the non-sites.rule14_n_θ,rule14_n_esh,no_spreading_nasal_distributed— (14) agrees with (12) on the English data, but its alpha-variables accept the[αnasal]variant that no single-node spreading can express.preaspiration— spreading the vowel's supralaryngeal node onto the delinked first half of an aspirated geminate keeps the stop's laryngeal features ((7)), by the substrate's disjointness of sister classes.mem_place_naturalClass_iff,primary_separates_stops— the place features are P ∪ S, and P alone separates the English stop places (table (22)).
Implementation notes #
classNode? is defined on the features the paper names and is none elsewhere, so the
root's natural class is the paper's inventory rather than all of Hayes's; [labial]
counts as primary by table (22). Spreading is Finset.piecewise on a node's natural class,
so the trigger's unspecified features under the node replace the target's, as the delinking
convention of (13) requires; single-feature spreading is Function.update
(Spreading.apply_feature). The English segments are the Fragment's; its /r/ is
[+anterior] where table (10) has [−anterior, −distributed], so the retroflex row of (11)
(tree, dream, enrol) is not derived. The SPE rule (14) is Effect.copyRight on the
two-feature set and (12) the same effect on the place class, so the comparison is between
the sets a rule may copy, not between formalisms.
TODO #
- The Sierra Popoluca release argument ((17)–(20)): the Shared Features Convention merges
the two
[−continuant]nodes of a homorganic cluster into one doubly linked node, so the insertion rule (18), which needs two such nodes, is blocked. This needs feature-level linking (Autosegmental/AR.lean), not the segment-wise spreading here. - §4's consonant–vowel asymmetries follow from plain consonants lacking S features and
vowels always carrying them; the Fragment's vowels do not specify
[back]and[round], so no instance is stated.
References #
- G. N. Clements, The Geometry of Phonological Features (1985)
- K. P. Mohanan, The Structure of the Melody (1983)
- H. Thráinsson, On the Phonology of Icelandic Preaspiration (1978)
- M. A. R. Barker, Klamath Grammar (1964)
- B. Elson, Sierra Popoluca Syllable Structure (1947)
- D. Steriade, Greek Prosodies and the Nature of Syllabification (1982)
- B. P. Hayes, Introductory Phonology (2009)
- N. Chomsky and M. Halle, The Sound Pattern of English (1968)
Class nodes #
Equations
- Clements1985.instDecidableEqNode x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- Clements1985.instReprNode = { reprPrec := Clements1985.instReprNode.repr }
Equations
- One or more equations did not get rendered due to their size.
- Clements1985.instReprNode.repr Clements1985.Node.root prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Clements1985.Node.root")).group prec✝
- Clements1985.instReprNode.repr Clements1985.Node.laryngeal prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Clements1985.Node.laryngeal")).group prec✝
- Clements1985.instReprNode.repr Clements1985.Node.manner prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Clements1985.Node.manner")).group prec✝
- Clements1985.instReprNode.repr Clements1985.Node.place prec✝ = Repr.addAppParen (Std.Format.nest (if prec✝ ≥ 1024 then 1 else 2) (Std.Format.text "Clements1985.Node.place")).group prec✝
Instances For
Equations
- Clements1985.instFintypeNode = { elems := { val := ↑Clements1985.Node.enumList, nodup := Clements1985.Node.enumList_nodup }, complete := Clements1985.instFintypeNode._proof_1 }
The node immediately dominating each node, the root fixed.
Equations
Instances For
a ≤ b when a lies on the walk from b up to the root, so the root is ⊥ and the parent
is Order.pred.
Equations
- One or more equations did not get rendered due to their size.
Equations
- a.instDecidableLE b = { decide := decide (List.elem a (List.iterate Clements1985.Node.pred b (Fintype.card Clements1985.Node)) = true), reflects_decide := ⋯ }
Equations
- Clements1985.Node.instDecidableLT = decidableLTOfDecidableLE
Equations
- Clements1985.Node.instOrderBot = { bot := Clements1985.Node.root, bot_le := Clements1985.Node.instOrderBot._proof_1 }
The parent as the predecessor.
Equations
- One or more equations did not get rendered due to their size.
Terminal features (§4) #
The class node a terminal feature hangs from. It is laryngeal for [spread],
[constricted] and [voiced], manner for [consonantal], [sonorant], [continuant],
[lateral], [strident] and, tentatively, [nasal], and place for the place features.
Features the paper does not place have none.
Equations
- Clements1985.classNode? Phonology.Feature.voice = some Clements1985.Node.laryngeal
- Clements1985.classNode? Phonology.Feature.spreadGlottis = some Clements1985.Node.laryngeal
- Clements1985.classNode? Phonology.Feature.constrGlottis = some Clements1985.Node.laryngeal
- Clements1985.classNode? Phonology.Feature.consonantal = some Clements1985.Node.manner
- Clements1985.classNode? Phonology.Feature.sonorant = some Clements1985.Node.manner
- Clements1985.classNode? Phonology.Feature.continuant = some Clements1985.Node.manner
- Clements1985.classNode? Phonology.Feature.lateral = some Clements1985.Node.manner
- Clements1985.classNode? Phonology.Feature.strident = some Clements1985.Node.manner
- Clements1985.classNode? Phonology.Feature.nasal = some Clements1985.Node.manner
- Clements1985.classNode? Phonology.Feature.labial = some Clements1985.Node.place
- Clements1985.classNode? Phonology.Feature.coronal = some Clements1985.Node.place
- Clements1985.classNode? Phonology.Feature.anterior = some Clements1985.Node.place
- Clements1985.classNode? Phonology.Feature.distributed = some Clements1985.Node.place
- Clements1985.classNode? Phonology.Feature.high = some Clements1985.Node.place
- Clements1985.classNode? Phonology.Feature.back = some Clements1985.Node.place
- Clements1985.classNode? Phonology.Feature.round = some Clements1985.Node.place
- Clements1985.classNode? x✝ = none
Instances For
The two sets of place features are P, distinguishing place in consonants, and S, distinguishing place in vowels (§4).
Instances For
Equations
- Clements1985.instDecidableEqPlaceSet x✝ y✝ = if h : x✝.ctorIdx = y✝.ctorIdx then isTrue ⋯ else isFalse ⋯
Equations
- Clements1985.instReprPlaceSet = { reprPrec := Clements1985.instReprPlaceSet.repr }
Equations
- One or more equations did not get rendered due to their size.
Instances For
[coronal], [anterior], [distributed] and, by table (22), [labial] are primary;
[high], [back] and [round] are secondary.
Equations
- Clements1985.placeSet? Phonology.Feature.labial = some Clements1985.PlaceSet.primary
- Clements1985.placeSet? Phonology.Feature.coronal = some Clements1985.PlaceSet.primary
- Clements1985.placeSet? Phonology.Feature.anterior = some Clements1985.PlaceSet.primary
- Clements1985.placeSet? Phonology.Feature.distributed = some Clements1985.PlaceSet.primary
- Clements1985.placeSet? Phonology.Feature.high = some Clements1985.PlaceSet.secondary
- Clements1985.placeSet? Phonology.Feature.back = some Clements1985.PlaceSet.secondary
- Clements1985.placeSet? Phonology.Feature.round = some Clements1985.PlaceSet.secondary
- Clements1985.placeSet? x✝ = none
Instances For
The features of a place set.
Equations
- π.features = {x : Phonology.Feature | Clements1985.placeSet? x = some π}
Instances For
The place features are exactly P ∪ S.
Assimilation as spreading ((5)) #
An assimilation rule spreads the root node (total assimilation), a class node (partial assimilation) or a single feature (single-feature assimilation), after Mohanan. A rule spreading more than one node costs more.
- node (a : Node) : Spreading
- feature (f : Phonology.Feature) : Spreading
Instances For
Equations
- Clements1985.instDecidableEqSpreading.decEq (Clements1985.Spreading.node a) (Clements1985.Spreading.node b) = if h : a = b then h ▸ isTrue ⋯ else isFalse ⋯
- Clements1985.instDecidableEqSpreading.decEq (Clements1985.Spreading.node a) (Clements1985.Spreading.feature f) = isFalse ⋯
- Clements1985.instDecidableEqSpreading.decEq (Clements1985.Spreading.feature f) (Clements1985.Spreading.node a) = isFalse ⋯
- Clements1985.instDecidableEqSpreading.decEq (Clements1985.Spreading.feature a) (Clements1985.Spreading.feature b) = if h : a = b then h ▸ isTrue ⋯ else isFalse ⋯
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Clements1985.instReprSpreading = { reprPrec := Clements1985.instReprSpreading.repr }
Equations
The features a spreading carries.
Equations
Instances For
Spreading from src onto tgt gives the carried features the values of src, specified
or not, the target's own being delinked ((13)), and leaves every other feature of tgt.
Instances For
Single-feature spreading overwrites the target's value at the feature with the source's.
Single-node spreading against the SPE rule (14) #
Place spreading carries [anterior] and [distributed] together and leaves [nasal]
alone, so (12) is a single-node rule while its [αnasal] variant of (14) is not.
No spreading reaches both [nasal], a manner feature, and [distributed], a place
feature, without carrying the whole supralaryngeal class (§3).
A rule assimilating [nasal] and [distributed] at once is total supralaryngeal
assimilation.
Kikuyu assigns [−continuant, +voiced] to postnasal obstruents (§5.1, (24)). Only the
root reaches a manner and a laryngeal feature at once, so the theory makes [voiced]
redundant there, filled in by a later rule.
English palatalisation assigns a glide's place and [+continuant] at once (§5.2, (25)):
one spreading doing both is supralaryngeal, so the paper decomposes it into spirantisation
and place assimilation.
English coronal place assimilation ((10)–(14)) #
In rule (12) as a local rewrite rule, a [−continuant, +coronal, +anterior] target, one
of /t d n/, takes the place node of a following [+consonantal, +coronal] segment.
Equations
- One or more equations did not get rendered due to their size.
Instances For
In the SPE statement (14), the same target before a [+coronal] segment copies that
segment's [anterior] and [distributed] by alpha-variables.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The variant of (14) with [αnasal] for [αanterior] is "no more and no less ordinary" as
an SPE rule, and unattested.
Equations
- One or more equations did not get rendered due to their size.
Instances For
In tenth, /n/ before /θ/ takes [+distributed] and nothing else, a dental nasal.
In eighth, /t/ before /θ/ does the same.
In hundredth, /d/ before /θ/ does the same.
In insure, /n/ before /ʃ/ is postalveolar, [−anterior, +distributed], and still a
non-strident nasal, [strident] and [nasal] being manner features.
A labial trigger is no site, and impossible falls to the separate nasal assimilation rule.
A fricative target is no site, since the rule affects the stops /t d n/ only.
On the English data (14) is not distinguishable from (12).
What separates them is that the [αnasal] variant is a rule of the same form and applies,
while no single-node spreading carries [nasal] and [distributed] without the whole
supralaryngeal class (supralaryngeal_subset_of_nasal_of_distributed).
Supralaryngeal spreading: Icelandic preaspiration ((6)–(7)) #
In (7) the first half of a geminate aspirated stop loses its supralaryngeal node and
takes the preceding vowel's, so the output shares every supralaryngeal feature with the
vowel and keeps only the stop's laryngeal features [+spread, −voiced], an [h]. Klamath
(9a) and Sierra Popoluca (16) spread the same node from a lateral and from a nasal.
Primary and secondary place features (§4) #
As table (22) shows, the primary features alone separate the English stop places, so
[high] and [back] are not needed for consonant place.