An articulatory theory of sound change An articulatory theory of sound change Hypothesis: Most common initial motivation for sound change is the automation of production. Tokens reduced online, are perceived as reduced and represented in the exemplar cluster as reduced.
Therefore we expect sound changes to reflect a decrease in gestural magnitude and an increase in gestural overlap. What are some ways to test the articulatory model? The theory makes predictions about what is a possible sound change. These predictions could be tested on a cross-linguistic database.
Sound changes that take place in the languages of the world are very similar (Blevins 2004, Bateman 2000, Hajek 1997, Greenberg et al. 1978).
We should consider both common and rare changes and try to explain both. Common and rare changes might have different characteristics.
Among the properties we could look for are types of phonetic motivation, types of lexical diffusion, gradualness, conditioning environment and resulting segments. Common vs. rare sound change?
We need a database that allows us to test hypotheses concerning what types of changes are common and what types are not. A database of sound changes? Most sound changes have occurred in undocumented periods so that we have no record of them.
Even in cases with written records, the phonetic interpretation may be unclear.
Only a small number of languages have historic records.
So any sample of known sound changes would be biased towards those languages. A database of sound changes?
Sound changes are known only for some languages of the world:
Languages with written histories.
Sound changes can be reconstructed by comparing related languages. PIE *p > f in Germanic Also: Latin ped- Old English fōt Latin piscis Old English fisc Traditional reconstruction of PIE stops (Lehmann 1955)
*p *t *k *kw *d *g *gw *bh *dh *gh *ghw
The ‘glottalic’ reconstruction (Hopper 1973)
*p *t *k *kw *t’ *k’ *kw’ *b *d *g *gw A database of sound changes?
Reconstructed changes cannot be the basis of a theory of sound change.
Many languages have no recorded history.
What can we learn from synchronic descriptions?
Most phonological rules that describe sound alternations come about through sound change.
Allophonic ‘rules’ or distributions can be thought of as describing sound change. Phonetic conditioning only: Sound changes have histories, too.
A sound change starts as a small phonetic change, but then may continue until the affected segments are more distinct phonetically.
Palatalization of /k/ in Latin > Spanish k > c > tʃ > ts > s / θ before a front V or glide
Spanish: eléctri[k]o vs. electri[s]idad Rule: k > s / ___ front vowel (telescoping) Rule: k > s / ___ front vowel Telescoping (Hyman 1975)
Note that the rule is not productive. One can now have /k/ before a front vowel without a change occurring: Spanish: quince [kinse] ‘fifteen’ queso [keso] ‘cheese’ Sound changes have histories, too.
Inversion: (Vennemann 1972) If we think of the English plural marker as /z/, then it is necessary to add a vowel in words likes classes, wishes, churches.
Historically, however, the vowel was there and a deletion occurred in most contexts, but not where the preceding C was a sibilant. Sound changes have histories, too.
Today, the vowel-insertion is morphologically- conditioned, as it is restricted to applying where /z/ is a plural, possessive or 3rd sg. Verb suffix.
The k à s ‘rule’ in Spanish is lexically restricted. Phonetic conditioning
In order to be sure the synchronic processes we are looking at are phonetically conditioned, we have to exclude alternations that are morphologically or lexically conditioned and not phonetically productive. ALLOPHON Database
Shelece Easterday and I constructed a database of phonetically conditioned (allophonic) processes in 82 languages.
The languages were chosen to be maximally unrelated genealogically. The sample is based on the GRAMCATS sample described in The Evolution of Grammar (1994). ALLOPHON Database
The data was collected from reference grammars and coded as follows:
1. the segment undergoing change, 2. the resulting segment, and 3. the conditioning environment were all coded for phonological features according to the traditional classifications of consonant vs. vowel, voicing, place of articulation and manner of articulation. ALLOPHON Database
2. Multiple segments undergoing a change were described in a single process if
a. the input segments constituted a phonetically describable class b. the output segments constituted a phonetically describable class c. the features changed are the same d. the conditioning environment is the same. ALLOPHON Database
3. Whether the process occurs across word- boundaries or not. 4. Whether the process occurs in rapid or casual speech or other special sociolinguistic contexts.
26% of the 800+ processes coded were noted to occur in the conditions mentioned in (4), suggesting that the database represents a shallow time-depth of sound change. Assumptions
1. Allophonic processes and phonetically- conditioned processes are the first stage in the phonologization of sound change. 2. Such processes, like sound change, have phonetic explanations. 3. While phonetic processes are very similar across languages, we recognize that there may be language-specific differences among them. 4. However, we also assume in looking for phonetic explanations, that similar processes across languages will have similar explanations. Some general results
49% of processes are assimilatory and may be gestural retiming.
35% are lenitions or the reduction of gestural magnitude or duration.
3.5% are strengthening Glide strengthening 19 Fricative strengthening 9 Results on assimilation
Of 391 assimilation processes Anticipatory 56% (See also Javkin 1978) Carry over 27% Both 17% Place assimilation: C-to-V assimilation Anticipatory 59% C-to-C assimilation Anticipatory 67% (See also Blevins 2004) Strengthening or fortition
A phonological process (or sound change) that increases the magnitude or duration of a gesture.
NB: defined articulatorily, not acoustically. t > ts or p > pf is not a strengthening, it is a weakening.
Closure point is eroding. On a path towards further weakening: pf > f; ts > s. Apparent strengthening or fortition Excrescent Cs: OE þunrian ‘to thunder’; bræmle ‘bramble’ Old Spanish: salir + á ‘leave + FUT’ > salrá > saldrá venir + á ‘come + FUT’> venrá > vendrá Latin: homine ‘man’ > Spanish hombre
Excrescent C develops out of articulations that are already present, as shown by the points of articulation. It is not an increase in magnitude of a gesture. Some languages reported to have synchronic or historical palatal glide strengthening. Blust 1990, in a survey of known sound changes in Astronesian languages reports that fortitions are restricted to the semivowels *y and *w. These are detailed on page 50.
They include *w becoming [gw] and *y becoming a coronal affricate He even cites cases where the transitional glide between two vowels strengthens into [b] or [g]. Glide strengthening AlloPhon database
Most frequent type of C strengthening in AlloPhon database. 12 cases of a palatal glide strengthening
ex: Pech /j/ > [dj] (word- or morpheme-initial) (Holt 1999:16)
ex: Slavey (Diné) variation: [j] ~ [ʒ] [ʔaji̜lá] ~ [ʔaʒi̜lá] ‘3s did to 3s’ (Rice 1999)
Apinajé (Macro-Jê) (Oliveira 2005:58-59) • a. /ajet/ [aˈʑet] ‘to be suspended on a surface’ • b./atkaje/ [atkaˈʑe] ‘to crack; to fissure’ • c. /jar/ [ˈʑari] ‘that (one)!’ Glide strengthening
7 cases of a labiovelar glide strengthening
ex: Pech /w/ > [gw] (word- or morpheme-initial) (Holt 1999:16)
ex: Apinajé (Oliveira 2005:60a) • a. /kuwe/ [guˈvej] ‘bird’ b. /aw/ [aˈvəɾ] ‘towards you’
Glide strengthening occurs in syllable-initial position. Fricative > stop or affricate
3 cases of palato-alveolar to affricate or stop Pech (Chibchan) /ta-ʃùna/ > [ta-tʃùna] ‘my nail’ (Holt 1999:16 younger speakers use [tʃ]; possible influence of Spanish.)
Sheko (Omotic) (Hellenthal 2010: 86) Free variation bāʒà [ bāʒà ] ~ [ bādʒà ] ~ [ bāɟà ] ‘work’
Garífuna (Arawakan) (Taylor 1955: 235) /ʃ/ in unstressed syllables varies with [tʃ] in stressed syllables. Fricative > stop or affricate
3 cases of /f/, /ɸ/, /β/ to /p/.
Ningil (Sepik) [p] and [ɸ] fluctuate freely (Manning and Saggers 1977:57) Koiari (Papuan) (Dutton 1996:6) /f/ may have [p] as a free variant with [ɸ] word-initially preceding back vowels. [ɸuˈɸuri] ~ [puˈɸuri] ‘Fufuri (name of a rock) Oksapmin (Papuan) (Loughnane 2009:33) /ɸ/ has [p] as an allophone syllable-finally (no /p/ phoneme as historically, *p and *ɸ merged. Fricative to stop or affricate
Are these strengthening?
2 cases of a glottal fricative > palatal fricative before high front vowels
Maidu: j > ç / ___ high front V (Shipley 1964) Tinrin: j > c / ___ high front V (Osumi 1995) Fricative to stop or affricate
Most of the cases describe variation. Directionality unclear in most cases.
In AlloPhon
Weakening of /p/ is documented in 8 cases Weakening of a palato-alveolar affricate is found in 3 cases
Weakening of stop to fricative is found at every point of articulation, but reported strengthening only at the labial and palato-alveolar points of articulation. Strengthening
Evidence concerning fricative to stop / affricate is doubtful.
Overwhelmingly, best documented type of strengthening is glide strengthening. Strengthening
Evidence that a palatal (high, front) articulation plays a substantial role in strengthening.
A role, but a lesser one, is played by labial articulations.
No tongue-tip articulations involved in strengthening
In the following we compare the points of articulation found in strengthening with those found in assimilation. Strengthening
Given how specific ‘strengthening’ processes seem to be 1. Not just the converse of ‘weakening’ 2. Perhaps related to Palatalization and labialization assimilations 1. Both tend to occur in syllable-initial position 2. Both they tend to occur in the same languages ‘Place’ assimilation C-to-V and V-to-C Vowel and glide backness features affecting Cs Table 4. Vowel and glide height features that affect consonants When V features affect Cs, it is the high front vowel that causes assimilation most often. Palatalization
An assimilation of a C conditioned by a high and/or front vowel or glide.
A typical palatalization has as its outcome an articulation in the palatal or palato-alveolar region, but our study encompasses other outcomes as well.
Secondary palatalization can occur on any C but changes in place usually affect coronals and velars.
Data from Hyman and Moxley1996. The letter c stands for [tʃ]. Palatalization
Full palatalization affects coronals and dorsals.
(Alleged full palatalization of labials appears to be attributable to glide strengthening, Bateman 2010)
Secondary palatalization can affect any C (Bateman 2007; 2010). Palatalization
Though anticipatory palatalization of a C to a V is the most frequent place feature assimilation, it only occurs in about half the languages of the world.
Bateman’s 2007 sample of 117 lgs yields about 50%.
Allophone sample of 82 languages yields 55%. Palatalization
The most common palatalization processes are anticipatory, meaning it is a syllable-initial C that palatalizes in most cases.
Why does a high, front tongue position cause more assimilation than any other articulation?
Recasens’ 2014 theory of articulatory constraint or resistance: owing to the mass of the articulator involved, some articulatory gestures resist the influence of neighboring gestures more, and also influence other gestures more. Palatalization and strengthening
Evidence that place assimilation and strengthening are related.
1. The same points of articulation are favored for both processes.
2. Some processes involve both assimilation and strengthening.
3. As we will see later, glide strengthening tends to occur in languages that have developed palatalization. Palatalization and strengthening
Some processes involve strengthening in a palatalizing environment: AlloPhon: the following have glide strengthening before a front vowel: Tinrin (Austronesian) Carib (Ge-Pano-Carib) Apinayé (Ge-Pano-Carib) Selepet (Indo-Pacific) Lahu (Sino-Tibetan) Palatalization and strengthening
In the convenience sample, the following have glide strengthening before a front vowel: Atayal (Austronesian, Northern Formosan) Awa (Kainantu, Barbacoan) Fanti (Niger-Congo, Akan) German (Indo-European) Nupe (Niger-Congo)
Strengthening before a high vowel: Lower Grand Valley Dani (Trans-New Guinea) Do palatalization and strengthening occur in the same languages?
Some languages show a long history of palatalization processes, with glide strengthening developing after other processes (Romance languages).
Therefore, we want to check the presence of synchronic palatalization processes and
The presence of palatal and palato-alveolar Cs in the phoneme inventory, because palatalization processes produce new phonemes. Languages with glide strengthening, AlloPhon sample. All of these have current palatalization processes, or Cs in the palato-alveolar region in their C inventories, indicating either current or past palatalization processes. Languages with synchronic or diachronic glide strengthening and the presence of palatalization processes or palatal / palato-alveolar Cs in phoneme inventory. Diachronic case study: Romance The effect of a palatal vowel or glide in Romance languages
First century AD (Penny 1991) Vulgar Latin (Penny 1991)
Changes in C phoneme inventories
French adds [ʃ], [ʒ] and [ɲ]
Spanish adds [tʃ], [ɲ] and [lj]
Portuguese adds [ʃ], [ʒ], [lj] and [ɲ] Palatal glide strengthening in Romance
Latin: did not have a consonantal glide according to Kent 1945.
Portuguese Latin iustum > Port justo [ʒustu] ‘fair’ iam > Port já [ʒa] ‘now, already’
Argentine Spanish Mod Spanishya ‘now’ > [ʒa] yo ‘I’ > [ʒo] llamar > [ʒamar] Palatalization and strengthening in C clusters
This sequence starts as palatalization and ends up as strengthening (in Portuguese).
Latin Spanish Portuguese plicare ‘to arrive’llegar [ʎ] > [j] [tʃegar] > [ʃegar] chegar pluvia ‘rain’ lluvia [ʎ] [tʃuva] > [ʃuva] chuva clamare ‘to call’ llamar [ʎ] [tʃamar]>[ʃamar] chamar clave ‘key’ llave [ʎ] [tʃave] > [ʃave] chave flamma ‘flame’ llama [ʎ] [tʃama] > [ʃama] chama
Menéndez-Pidal: the [l] palatalizes and the initial C is lost. Palatalization and strengthening in C clusters
French strengthening of [j] in clusters with labial Cs: Is there a palatalizing type of language?
As we saw above, all the languages we have checked that have glide strengthening also have assimilation processes conditioned by a high and/or front V or glide, or
Cs in their phoneme inventory in the palatal or palato- alveolar region, evidence of such processes having occurred in the past.
Recall that only about half of the languages sampled by us and by Bateman have palatalization processes. Is there a palatalizing type of language?
Possible underlying causes: 1. Greater co-articulation of syllable-initial C and V. In that case other place assimilations would also occur.
2. Articulatory setting: perhaps these languages have a high front tongue position that is higher and more to the front than other languages. (Gick, Bryan, Ian Wilson, Karsten Koch & Clare Cook. (2004). Language-specific articulatory settings: evidence from inter-utterance rest position. Phonetica 61. 220-233.)
3. Articulatory constraint: perhaps the high front tongue position has greater articulatory constraint in some languages than in others. (Recasens, Daniel & Aina Espinosa. (2009). An articulatory investigation of lingual coarticulatory resistance and aggressiveness for consonants and vowels in Catalan. JASA 125. 2288–98.) Articulation vs. perception in the explanation for sound change Directionality
Lenition is very common Fortitions are very constrained, occurring only with specific articulations, which happen to be articulations commonly involved in assimilations. Assimilation is very common Dissimilation is rare and often not lexically regular Metathesis is rare. Crosslinguistic patterns
Strongly suggest phonetic explanations for sound change.
But, are these explanations grounded in articulation or perception?
My view is that sound change takes place by the automation of production. 1. high frequency words undergo sound change earlier than low frequency words 2. assimilation and lenition strongly resemble changes that occur in other highly-practiced neuromotor activities. Are sound changes motivated by articulation or perception? Ohala 2003
Variation: ‘essentially infinite’ (p. 672)
Present-day variation parallels sound change Examples in (2) are palatalizations of coronal before a high front vowel or glide. Ohala: misperception
Compensation in perception: normalization
Experiments have shown listeners compensate for the phonetic environment.
Beddor et al: /ɛ/ - /æ/ continuum under three conditions: /ɛd/ - /æd/ /ɛd̃ / - /æ̃ d/ (subjects heard more /æ/ vowels) /ɛñ / - /æ̃ n/
Listeners parse the nasalization with the nasal C. Misperception: failure to normalize
Applies to assimilative sound changes.
Presupposes prior articulatory variation. The ‘misperception’ phonologizes the variation. Can lenition be explained in the same way?
VdV > VðV > VV 1. Why would high frequency words change earlier than low frequency words? 2. How do misperceptions spread across the lexicon? 3. Given how finely-tuned the perceptual system is, why would listeners make mistakes? 4. Can misperception account for glide fortition? Why would listeners make mistakes?
Maybe they are not mistakes.
Maybe once the articulatory change has advanced to a certain point, there is a change in what the listener and speaker take to be the main feature or locus of the feature.
Beddor 2009 shows this for ṼN sequences in English. Are there any regular sound changes that have to be due to misperception? Theta to f: throw > frow
Other ‘long distance’ changes:
Czech dialects: p’ > t dissimilation
Latin Spanish gloss arbor árbol tree robre roble oak marmore mármol marble carcere cárcel prison locale lugar place dissimilation
Latin suffix –ālis > -āris When the stem ends in /l/ national popular marginal regular optimal modular Dissimilation
Ohala 2003: Dissimilation is perceptual hypercorrection
Listeners do a lot of normalization or correction when decoding speech. When a feature from one segment tends to spread over a large stretch of speech, listeners learn to disregard it.
If they disregard a feature that is not redundant, this might lead to dissimilation.
If the rhotic quality of the final vowel of arbor is attributed to the earlier /r/, then /l/ might be produced. dissimilation
Tongue-twister effect
Difficult to produce two /r/s or two /l/s in a row. Dissimilation
1. Features involved would be those whose acoustic- perceptual cues tend to spread over relatively long time intervals. Aspiration, glottalization, retroflexion, palatalization, pharyngealization, labialization, etc.
2. Conditioning environment is preserved.
3. Does not produce novel segments.
4. Occurs only within words.
1. Not usually lexically regular dissimilation
Lexically regular examples involve laryngeal features such as glottatalization or aspiration. Ancient Greek: only one aspirated stop per word Hausa (Chaddic) and Yucatec Mayan: cannot have two ejectives per word unless they are otherwise identical Quecha: only one glottalized C per word. dissimilation
Note that the result is always a ‘plain’ C.
Cases do not occur in which one of two plain Cs becomes aspirated or glottalized.
Could this also be a reduction? Metathesis
Metathesis: two segments appear to change position (Hume 2004) Latin riparia > [ribeira] > Sp [riβera] ‘river bank’ Latin casium > caiso >Sp queso ‘cheese’ [keso] sporadic English: axian ‘to ask’ > ask [aksian] Metathesis: Modern Hebrew
Morpheme surface form gloss hit-nakem hitnakem he took revenge hit-raxec hitraxec he washed himself hit-balet hidbalet he became prominent hit-darder hiddarderhe declined, rolled over hit-sader histader he got organized hit-zaken hidzaden he grew old hit-calem hictalem he took pictures of self hit-ʃamer hiʃtamer he preserved himself Metathesis: Hume’s account
1. The sequence involves acoustic features that can spread over more than one segment, making their assignment to a segment difficult. 2. Listeners (and speakers) tend to interpret ambiguous sequences in the order they are more familiar with—the sequence that is more common in the language. Other changes motivated by phonotactics
Figure 1: Pepsi add used in Argentina from 2010. The add says ‘drink Pecsi, save.’ Down below it says ‘drink Pepsi also’. Spanish syllable-final obstruents
Final labial C: concepto, concepción ‘concept’ ‘conception’ obtener ‘obtain’
Final velar C: acción [ks] ‘action’ exacto [ks] ‘exact’ técnico [kn] ‘technical’ fragmento [gm] proyecto [kt] ‘project’ signatura [gn] Brown 2006 pursued the hypothesis that the low frequency coda is replaced by a higher frequency one.
Brown’s point is that when phonotactic sequences are rare, speakers/listeners/learners may reinterpret them as sequences that are much more common. Other non-lenition changes
Vowel epenthesis happens in consonant clusters that (usually) involve sonorant consonants.
Vowel lengthening
Consonant gemination has two sources: CC assimilation rhythm Table 1: A typology of sound change based on six factors.