1. Levels in American Structuralist phonology

The phonemic level generalizes across forms that are phonetically different but not meaningfully distinct (e.g. allophones):
 

(1)	[ph]it,	si[pt],	s[p]it		Allophones
	\	|	/
		/p/			Phonemes

The morphophonemic level generalizes across forms that are phonemically different forms of the same morpheme (i.e. allomorphs):
 

(2) a.	s/aI/n	s/Ig/nature		b.	electri[k]	electri[s]ity		Phonemes
	\	/			\	/
	{sIgn}				{elektrIk}			Morphophonemes

Although similar transcriptions are used on all three levels, American structuralists regarded them as different in kind. For example, phones are actual speech events, whereas phonemes are abstract algebraic symbols. Both kinds of relations can be formalized using the rewriting rule formalism of generative phonology, which describe how to convert a string of symbols at one level to a different string at another level. However, in structuralist phonology such mappings are regarded as realization rules, rather than rewriting as such.

(3) [+consonantal, +tense] -> [+spread glottis, -constricted glottis] /  __ V
     [+consonantal, +tense] -> [-spread glottis, +constricted glottis] / V __
     [+consonantal, +tense] -> [-spread glottis, -constricted glottis] / s __ V

G-deletion (e.g. sign, paradigm etc.): g -> Ø / __ C #  (Ø = zero/empty segment)
Velar softening: k -> s / __ i

2. Multiple levels in generative phonology

Halle, M. (1959) The Sound Pattern of Russian: A Linguistic and Acoustical Investigation. Mouton.

In Russian, all the voiceless obstruents except /ts/, /t/ and /x/ have distinctively voiced phoneme counterparts.  /ts/, /t/ and /x/ have voiced allophones, which occur when they are followed by a voiced obstruent. e.g. [^jet l^ji]  `should one burn?' ([l<sup>j</sup>], although voiced, is not an obstruent) vs. [<sup>j</sup>ed b] `were one to burn'. An allophonic rule of voicing assimilation seems to be required:

(4)  /ts/, /t/, /x/  -> [+voice] / __ C[-sonorant, +voice]

In structuralist terms, this is a phonemic realization rule, relating the phonemic level to the allophonic level. However, almost the same kind of rule is required for the other voiceless obstruents, which also undergo voicing assimilation e.g. [m^jok l^ji]  `was (he) getting wet?' vs. [m<sup>j</sup>og b] `were (he) getting wet'. Here, however, the voicing distinctions are phonemic, so the relevant rule would be a morphophonemic realization rule, e.g.:

(5)     C{-sonorant, -voice}  -> /+voice/ / __ C/-sonorant, +voice/

Halle observed that  rules like (4) and (5) are extremely similar in form and function, and that (4) would not be necessary were it not for the fact that they are regarded as applying on different levels, as (5) applies to all voiceless obstruents. It appears from examples such as this that rules can operate at different levels, with the consequence that there is not a neat separation between a fixed, small number of levels. Halle and other generative phonologists observed that in many cases, grammars are simpler if several rules are permitted to apply in sequence, implicitly defining numerous intermediate levels of representation. E.g. in the derivation of sign:

(6) sIgn => sIXn     (g­deletion; X is an empty segment slot)
             => sI:n  (compensatory lengthening)
             => saIn  (vowel shift)
             => ...

Generative phonology only attributes a special status to the input level (the level at which lexical entries are expressed), called the systematic phonemic level, and the output or surface phonetic level, called the systematic phonetic level.

3. What is a process?

Not all rules represent (putative) phonological processes. For instance, the redundancy rule

(7)     [+sonorant] -> [+continuant]

is a feature­filling rule which simply refines the specification of sonorant: it does not change sonorants into something else. The rule gives the value "+" to the feature [continuant] in all [+sonorant] segments. This is rather different from the effects of a rule such as

(8)     [+sonorant] -> [+spread glottis] / [+spread glottis] __

the rule which devoices the sonorants in initial /pr/, /kl/ etc.

Processes, therefore, are typically feature­changing. In rules that refer to phonological structure, processes are usually structure­changing (e.g. resyllabification), but sometimes structure­building operations (such as syllabification) are also regarded by many phonologists as processes.

4. Phonetics or phonology?

In a phenomenon such as consonant­vowel coarticulation (e.g. keep/cart/cool), is a phonological rule at work (i.e. in the grammar of the language), or is it an involuntary side­effect of the motions of the articulators for the consonant and vowel targets?

There are not always clear answers to this question. Some guiding principles:
 

(9)	Phonological rules		Phonetic phenomena
	· Vary between languages		· Language­independent
	·  Involve categorial distinctions     e.g. electri/k/ vs. electri/s/ity		· May be gradient      e.g. ran/ quickly
	· Typically morphophonological		· Blind to morphology

5. Assimilation: a context­dependent change in the value of a feature in order to make it more similar or even identical to another feature or segment in the neighbourhood.

5.1. Progressive or perseverative assimilation

(10) e.g. open: [oUpm], seven: [sev]

5.2. Regressive or anticipatory assimilation

(11) e.g.    I'm coming: [aIkmI]

Phonological or phonetic? The following examples are more clearly phonological:

(12)  in+legal        = illegal
        in+relevant   = irrelevant
        in+possible   = impossible
        in+credible   = icredible

cf. French voicing assimilation (Prosodic Domains and Prosodic Structure lecture §2.2), Place assimilation in English nasal + obstruent clusters (Prosodic Domains and Prosodic Structure lecture §2.3)

5.3. Vowel harmony

cf. Kirgiz (Prosodic Domains and Prosodic Structure lecture §4.2)

5.4. Fusional assimilation

(13)    e.g. did you => [dIdu]
(14)    Sanskrit vowel sandhi:  Maha+Indra => Mahendra
(15)    Japanese fast speech/men's vulgar speech: /itai/ => [itee] `painful'; /semai/ => [semee] `narrow, small'
(16)    KiHaya slow speech:  [emwá ikambóna] `a dog saw me' vs. normal rate speech: [emwé:kambóna]

6. Dissimilation

(17) Latin /arbor/ > Modern Spanish /arbol/ (dissimilation of [±lateral] in /r/ and /l/)
(18) English dialectal `chimney' => 'chimley' => `chimbley' (dissimilation of [±nasal] in /m/ and /n/; the [b] is epenthetic; see below)
(19) Sanskrit: Grassman's Law, e.g. /bhudh + am/ => budham (dissimilation of aspiration)
cf. unsuffixed form [bhut] <= /bhudh/, with final devoicing.
(20) English cognate < con+nate.

7. Lenition (softening or weakening)

(21) Strength hierarchy:

a) Stop > Fricative > Approximant > Zero
b) Voiceless > Voiced

7.2. Phonological e.g. North Welsh
 

(22)	N	his N
	/pen/	/i ben/	`head'
	/braud/	/i vraud/	`ship'

7.3. Morphophonological

(23)  English electri[k] vs. electri[]an vs. electri[s]ity (perhaps [k] -> [c] -> [t] -> [] -> [s])

7.4. Historical

(24)  French []eval < Latin [k]aballus
(25)  Germanic /f/ < Proto­Indo­European /p/; English fish cf. Latin piscis
         Germanic // < Proto­Indo­European /t/; cf. brother vs. frater, tooth vs. dent­(al)

7.5. Vowel reduction e.g. phot[]graph vs. photógrapher (unstressed /o/ surfaces as schwa). The usual analysis is that all the features of /o/ are deleted, leaving an empty V­slot. The realization of an empty V position defaults to schwa in English.

8. Fortition (strengthening)

Fortition is rarer than lenition, and some phonologists deny the existence of fortition, analysing all cases of historical fortition as e.g. borrowing, and all synchronic cases as e.g. suppletion. E.g. Fula `suudu' house vs. `tuudi' houses, beside lenition e.g. `pullo' Fula person vs. `fulbe' Fula people.  Many languages without /f/ borrow [f]­initial words with [p] e.g. `pilipino' = the name for the Tagalog language (in Tagalog), the main language of the Philippines. American Hispanic pronunciation of English [j] as in `you' with an initial palatal plosive could be regarded as fortition. Faroese geminate approximates strengthen e.g. /jj/ -> [], /ww/ -> [gv]. Are these instances of articulatory overshoot?

One case of fortition is common, however: devoicing. E.g. Final obstruent devoicing in German, Polish, Catalan, English. E.g.

(26)    German [ba:t] `bath' vs. Bä[d]er `baths'; [to:t] `death' vs. To[d]es `of death'

There has been a protracted debate in the phonetics literature about these cases, between the view that final devoicing is purely phonetic and does not result in the identity of derived [t] and lexical /t/, and the opposite view, that the process is neutralizing and therefore an instance of phonological rewriting.

9. Insertion

9.1. Prothesis e.g. Fr. esprit < Latin spiritus.
9.2. Anaptyxis: Insertion of a vowel to break up a cluster e.g. athVlete. /l/ is neither a well­formed syllable onset nor a well­formed syllable coda in English. Insertion of V's tends to make the syllabification more like the simpler CVCVCV... pattern, which is universally unmarked (allegedly).

Prothesis and anaptyxis are often called simply `vowel­' or `schwa epenthesis'. The usual analysis is that an empty V defaults to a vowel such as schwa or [e].

9.3. Breaking of vowels (the opposite of fusion).

9.4. Consonant epenthesis e.g.

(27) Ø -> [t] / [n] __ [s]       printce vs. prints
       Ø -> [p] / [m] __ []       warmth
       Ø -> [k] / [] __ []       length

Since a phonetic distinction is maintained between `prin<sup>t</sup>ce' and `prints', this kind of epenthesis is usually regarded as a fairly low­level, phonetic phenomenon. (How else can we distinguish between epenthetic [t] and lexical [t]?) One view is that this is not a case of segment insertion at all, but of temporal coordination i.e. the nasality and voicing of [n] are `turned off' in preparation for the following [s] before the coronal closure is released, producing, as a side­effect, a momentary period of voiceless coronal closure. But this is just a detail of the transition from [n] to [s], not a `real segment' at all.

9.5. Linking or liaison consonants
 

(28) e.g. English	`intrusive r' and `linking r'		e.g. saw[r] only = sore only
	`linking w'		e.g. go[w] off
	a/an allomorphy		e.g. a cart vs. a[n] apple.

NB no linker in Massachusetts English shoulda eaten, the Beqaa in Lebanon, suggesting that maybe all instances of intrusive r are really linking r.

10. Deletion

10.1. Phonetic (?) e.g. diction'ry

10.2. Phonological e.g. Welsh gardd `garden' => yr ardd `the garden'; English /damn/ => [dam] cf. dam[n]ation; /paradigm/ => [paradaIm] cf. paradi[g]matic.

The Welsh case might alternatively be regarded as a lenition of /g/ to Vowel (perhaps via [g]), or as a purely paradigmatic alternation (i.e. with gardd and ardd not in a derivational relationship). The English cases are so unproductive that the justification for using a rule may be questioned.

11. Reordering (permutation)

A not­uncommon historical process, but what synchronically much rarer.

11.1  Re­parsing e.g. Coda /l/ in `belittle' => Onset /l/ in `belittling' is commoner. But maybe this is just different parsing of different strings i.e. we do not parse `belittle' with a coda /l/ and then add ­ing and re­parse. Hebrew: /hit + sadr + ud/ => [histadrut]. But since V and C patterns are separate morphemes in Semitic languages, /hit/ is really {i, h<t}. Segmental order is only partially specified, and the consonant orders h<t and s<d<r are not violated in [histadrut].

11.2. Movement of single features:

(29) Sanskrit: Bartholomae's law e.g.  /labh + ta/ => [labdha] (permutation of aspiration feature of /b/ and /d/).

(30) English: historically, /iw/ > /ju/ in e.g. `new'. This could be analyzed as a transfer of syllabicity from /i/ to /u/ in /iu/. But that would only be a movement if a syllabicity feature is employed. Otherwise, it can be regarded simply as reparsing (or different parsing).

Further reading: Lass Phonology ch. 8, Kenstowicz ch. 2-3.