Chapter Four: Analogical change
As well as regular phonemic change, some changes take place by analogy to other forms.
Analogical levelling and analogical extension
'Weak' English verbs are regular verbs, and new verbs follow their pattern - so I google, I googled, I have googled. 'Strong' verbs are irregular ones which use vowel changes to indicate past tense e.g. I sing, I sang, I have sung. Analogical levelling reduces the number of allomorphs a form has, making it more regular. So 'live' is now a weak verb; its alterations have been levelled. Analogical extension, slightly rarer, extends the irregular pattern to previously regular verbs. So 'wear' now has past tense 'wore'.Sturtevant's paradox
Sturtevant's paradox is that "sound change is regular and causes irregularity; analogy is irregular and causes regularity" (Antilla 1989: 94) In other words, when a sound change applies (regularly) to a given environment, it can cause singular and plural forms, or present and past tenses, or various other paradigms to become more different from one another, and unpredictable in their formation. Given these irregular paradigms, analogy is applied to make them more similar again.Immediate and non-immediate models
Immediate models involve words that are frequently said together - like January and Febuary (February), male and female (femelle), four (whour) and five.
Non-immediate models are more common, and include things like the classes of verbs we've already seen.
Miscellaneous types of analogical changes
Other changes that are sometimes loosely categorised as analogy:- Hypercorrection
Where a dialect has undergone a merger and the prestige (or 'standard') dialect has not, and attempts to reverse the merger cause sounds that were already correct to change - Folk etymology
'Hamburger' means 'of Hamburg', but is analysed as 'ham-burger', giving also 'cheeseburger', 'lamb burger' etc. - Back formation
A type of folk etymology, assuming things like plurality where it doesn't exist - e.g. French singular 'cheris' was assumed to be 'cheri-s', so the English singular is 'cherry'. - Metanalysis
Reanalysing which parts of a phrase belong to which word - e.g. 'a napron' became 'an apron'. - Blending
E.g. smoke + fog = smog. Blends may be deliberate or accidental.
Chapter Five: The comparative method
The comparative method is used for reconstructing the ancestor language ("proto-language") of a group of languages by comparing cognates across those languages.Latin
Latin is a useful tool for this, because it is a well-documented language in its own right, so we can verify the results of using the comparative method on Romance languages (French, Spanish, Italian etc - those descended from the Roman language). This gives us confidence in applying it to languages whose ancestor is not documented, e.g. Germanic languages (English, German, Swedish). The success of the comparative method depends on how closely related a proto-language is to its descendents; results can vary depending on the data available. For example, whilst a great deal of Latin is recoverable, "the former noun cases and tense-aspect verbal morphology which Latin had" is mostly not recoverable from the modern Romance languages.Assumptions
The Comparative Method relies on a few assumptions:- The proto-language is only a single dialect, with no variation. (We have no way of recovering more than this.)
- Language splits are sudden, and there is no further contact between related languages. (Again, this is not true, but the comparative method does not have tools for dealing with borrowings.)
- Sound change is regular
The steps involved
The steps of the comparative method - not necessarily in this order - are:Assemble cognates
These will usually be 'basic vocabulary' which is unlikely to be borrowed, and is therefore inherited from the proto-language.- Establish sound correspondences
E.g. Spanish /k/ (spelt 'c') corresponds to French /ʃ/ (spelt 'ch')
e.g. caro = cher (dear), cabra = chèvre (goat), cosa = chose (thing) etc.
This correspondence has to hold frequently; it cannot just be one or two words that happen to have similar sounds. (So we probably cannot say that Spanish /k/ corresponds to English /g/, because in other places English has /d/ or /θ/ (th) or... but we need far more words than these few!)
Reconstruct the proto-sound
Once we have a sound correspondence, we want to find out what the proto-sound was (i.e. which sound the various ones descended from).- Directionality Some sound changes tend to go in one direction, but not the other
- Majority wins Whichever sound most languages have for a particular sound correspondence is likely to be the original sound. It is more likely that a minority of languages underwent a change than that most languages underwent the same change.
- Factoring in common features The descendents of the proto-sound are called "relexes", because they reflect the original sound. Phonetic features that are shared among reflexes are likely to reflect features of the original e.g. b,p are both labial sounds (articulated with the lips), and so the proto-sound is likely to also have been labial.
- Economy The fewer changes, the better.
Partially overlapping correspondence sets
If two correspondence sets differ only partially, are they descended from one proto-sound which split, or two seperate proto-sounds which merged?
If there is a predictable environment for the split - i.e. if the change is conditioned - then it is assumed there was one proto-sound. If the difference is unpredictable, it is assumed there are separate proto-sounds.
This is the same rationale that underlines synchronic phonological theory: if the pronunciation of a sound is predictable from its environment, it is an allophone of an underlying phoneme; if it is unpredictable, that is its underlying form. (See 'Introduction to phonology'.)Phonological fit & typological fit
Certain phonological inventories are more common than others. For example, languages tend to have symmetry: if a language has (p, t, k) and (b, d), it will tend to also have g.
Also, the existence of certain classes of sounds mandates the existence of others: e.g. no languages have only glottalised consonants and no non-glottalised ones.
Given the reconstructed proto-sounds as a set, we can check whether their distribution is a more or less common one, which can help decide between competing proto-sounds.
Not only are certain distibutions of sounds more common, but individual sounds are also found with different frequencies across the world's languages.
e.g. s > h, but h does not tend to change to s
e.g. voiceless stops (p,t,k) become voiced between vowels, but (b,d,g) tend not to become voiceless there. This is a phonetically motivated change (it is easier to continue voicing than to stop and start again).
However, it is possible that a subsection of languages have an intermediate ancestor which underwent a single change; that all the languages have changed; or that the sound change is so common that they did all undergo the same change.
Finally, we can reconstruct individual morphemes by combining proto-sounds.
Indo-European sound changes
There are three famous laws from the study of Indo-European: Grimm's Law, Grassmann's Law, and Verner's Law.Grimm's Law (published 1822) is a series of changes of Proto-Indo-European to Proto-Germanic: voiceless stops > voiceless fricatives; voiced stops > voiceless stops; and voiced aspirated stops > voiced plain stops. With these changes, we can establish sound correspondences between Germanic and other Indo-European languages
e.g. pater / father, tres / three; decuem / ten, duo / two, genus / kin, bhratar (Sanskrit) / brother.
Grassman's Law (published 1862) "regularly dissimilated the first of two aspirated stops within a word so that the first lost its aspiration".
Verner's Law (published 1877, also called grammatical alternation) is that medial voiceless stops and fricatives before an accent became voiced in Germanic.
Together, these laws provide evidence that sound change is regular and exceptionless.
Chapter Six: Linguistic classification
Dialects are mutually intelligible varieties of languages. Dialects become languages when they are no longer mutually intelligible. These languages are called sister languages, all daughters of the same proto-language. (I don't know why languages and words are female. They just are.) Related languages form language families. The more closely related parts of a language family are called subgroups or branches.The criterion for deciding subgrouping is "shared innovation". This is a change from the proto-language to the current languages which they both undergo. As mentioned previously, this is more likely to happen in one descendant which later splits than in two separate descendants. Obviously, the accuracy of this analysis is dependent on correctly identifying the original form: two languages may retain a feature despite sharing only the most distant proto-language in common.
There is no way of dating language splits at present. The Swadesh 100-word list of core vocabulary, claimed to decay at a uniform rate across languages, in fact does no such thing. Furthermore, the words it contains are not resistant to borrowing, and are not core in many languages (being either more or less specific than their English equivalents).
Reference
Quote, via Historical Linguistics, from Raimo Antilla An introduction to Historical and Comparative Linguistics (New York: Macmillan, 1972 [2nd edn: Current issues in Linguitic Theory, 4, Amsterdam: John Benjamins, 1989]