3. Internal Reconstruction.pdf

3. Internal Reconstruction : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:29 PM

3. Internal Reconstruction

DON RINGE

Subject

Linguistics » Historical Linguistics

DOI:

10.1111/b.9781405127479.2004.00005.x

“Internal reconstruction” (IR) is the exploitation of patterns in the synchronic grammar of a single language

or dialect to recover information about its prehistory. The methods of IR are generally less reliable than the

standard methods of comparative reconstruction (CR; see Rankin, this volume) for the following reasons.

Many of the changes that occur naturally in languages over time eliminate language structures in

unrecoverable ways. These include the replacement of lexemes by completely different words (e.g., the

replacement of Old English (OE) sinwealt by Middle English round ); the syntactic merger or loss of

grammatical categories (e.g., the merger of the dative and instrumental cases within the OE period, and the

subsequent loss of the dative); the leveling of morpho-phonemic alternations (on which see further below);

the unconditioned merger or loss of phonemes; and other, less common processes (see Hoenigswald 1960:

28–37, 90–1). CR circumvents the effects of these changes by adducing evidence from related languages or

dialects in which the same changes have not occurred; IR has no comparably straightforward means of

“undoing” the changes. In the absence of comparative evidence, IR must make use of several assumptions

about which types of changes are most likely to have given rise to the synchronic patterns observed. Many of

those assumptions are not problematic, but the only one that is completely reliable in every case is the

fundamental observation on which CR is also based – namely, that sound change is overwhelmingly regular.

IR is therefore of limited use in historical linguistics; CR is so much more reliable that it is preferred

whenever possible. But there are situations in which the linguist is not offered a choice, either because a

language is not demon-strably related to any other, or because it has been developing in isolation from its

nearest kin for so long that comparative work encounters massive practical difficulties. A firm grasp of the

principles of IR is therefore an essential part of the historical linguist's professional knowledge.

Like all methods of linguistic reconstruction, IR proceeds by making inferences about unobservable stages of

a language's development on the basis of what is known from the observed history of languages. Therefore

one can best gain an understanding of IR by studying, in light of the known principles of language change,

linguistic patterns whose origin and development is already well understood. Most of this chapter will

accordingly be devoted to discussion of relevant examples. Since the standard theoretical treatment of

Hoenigswald (1960: 68–9, 99–111) can scarcely be bettered, I will concentrate on the practical problems

that IR involves.

The structural patterns that are most useful for IR are alternations between (lexical) phonemes in

morphological contexts. I shall first discuss and exemplify the exploitation of individual alternations, then

consider other types of patterns that can be used in IR. For further discussion see now Fox (1995: 145–

216). 1

1 Alternations Resulting from Conditioned Merger

IR most often exploits alternations resulting from the conditioned merger of phonemes, which is necessarily

accompanied by split of one of the original phonemes (Hoenigswald 1960: 91–3); in Hoenigswald's

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274795

Page 1 of 14

3. Internal Reconstruction : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:29 PM

accompanied by split of one of the original phonemes (Hoenigswald 1960: 91–3); in Hoenigswald's

maximally concise formulation:

phonemic SPLIT in several of its varieties leads to morphophonemic alternation, provided that

morph boundaries fall between the conditioning and the conditioned phoneme or phonemes

and provided that the same phoneme in the same morph thus comes within the range

sometimes of one, sometimes of the other, type of conditioning phoneme or phonemes.

(p. 100)

The type of conditioned merger that presents us with patterns of data most favorable to IR involves the

neutralization of phonemic contrasts in easily recognized environments which occur often enough to provide

numerous examples (cf. Hoenigswald 1960: 100–2). A straightforward case is the devoicing of word-final

obstruents observable in (Standard) German. 2 Especially numerous are examples involving stem-final

alveolar stops, of which the following partial noun paradigms are typical: 3

Singular Plural

Meaning

/ta:t/

/ta:tәn/

‘deed’

/p f a:t/ /p f a:dә/

‘path’

/gra:t/ /gra:dә/

‘degree, rank’

/gra:t/ /gra:tә/

‘edge, ridge’

/špa:t/ /špa:tә/ d /

špe:tә/

‘spar' [mineral]

/ra:t/

/re:tә/

‘council, councilor’

/ra:t/

/re:dәr/

‘wheeL’

It is clear that the shape of the plural cannot be predicted from the shape of the singular; and one of the

unpredictable details is whether the final /t/ of the singular reappears in the plural or /d/ appears in its

place. The same phenomenon occurs in the inflection of other classes of words which have endingless forms.

For example, among adjectives one finds /bunt/ ‘mottled' with inflected forms /buntә/, etc., but /gәzunt/

‘healthy,’ /gәzundә/, etc.; among verbs one finds narrative preterite /zi: ri:t/ ‘she advised' and /zi: ri:tәn/

‘they advised,’ but /zi: fermi:t/ ‘she avoided' and /zi: fermi:dәn/ ‘they avoided.’ Nor is the phenomenon

restricted to these particular consonants; one also finds the alternation /-k/ d /-g-/ contrasting with

invariant /k/(/verk/ ‘work,’ pl. /verkә/ but /t s verk/ ‘dwarf,’ pl. /t s vergә/, etc.), the alternation /-s/ d /- z -/

contrasting with invariant /s/, and so on.

Because this phenomenon occurs in the inflection of words of different morphological classes, its origin

cannot plausibly be attributed to morphological change; after all, it is most unlikely that three or more

different morphological changes would give precisely the same result. Because a large proportion of the

language's basic vocabulary is involved, any explanation involving borrowing from another language 4 is

likewise implausible. Only sound change could reasonably have given rise to so pervasive a pattern, and the

suspicion that sound change is responsible is confirmed by the details of the pattern: it involves a natural

class of sounds, namely obstruents, in a clearly definable phonotactic position, namely at the ends of

phonological words.

Once it is clear that sound change is responsible for the observed pattern, we can exploit the fact that sound

change is overwhelmingly regular – that is, that the conditions which govern sound change are strictly

phonological. If the stem-final consonant had originally been *t in all the forms adduced above and had

become /-d-/ between vowels, we would be unable to explain why it had become /-d-/ in /p f a:dә/ ‘paths'

but not in /gra:tә/ ‘ridges,’ and so on, since no phonological conditioning for the difference can be stated.

Therefore we must conclude that the paradigms in question were originally 5 *ta:t, *ta:tәn; *p f a:d, *p f a:dә;

*gra:d, *gra:dә (‘degree’); *gra:t, *gra:tә (‘ridge’); and so on, and that the alternation between word-final /-

t/ and non-final /-d-/ was created by a regular sound change which devoiced word-final obstruents

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274795

Page 2 of 14

3. Internal Reconstruction : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:29 PM

t/ and non-final /-d-/ was created by a regular sound change which devoiced word-final obstruents

(affecting also *-g, *-z, etc.; see above). The exceptionlessness of sound change is reflected in the

exceptionlessness of the alternation, which is completely automatic: one simply does not find word-final /-

d/, etc., in this variety of German.

Since this is a maximally simple example with which further examples will be compared, it is worth noting a

number of additional facts about it. Because the alternation between voiced and voiceless obstruents is fully

automatic, it remains fully transparent to the native speaker: a theory of phonology which permits any

abstraction from surface contrasts at all will analyze the alternation /-t/ d /-d-/ simply as /d/, and in fact

that is the analysis reflected in Standard German spelling ( Pfad, Pfade , etc.). In such a simple case IR

replicates phonological analysis point for point, and the reconstruction of the earlier state of affairs is

achieved simply by deleting a single phonological rule from the grammar.

Yet even in such a straightforward example, not every detail of our reconstruction is historically accurate.

For instance, IR fails to tell us that Grad ‘degree,’ unlike the other nouns listed above, was borrowed into

German well after the devoicing of word-final obstruents occurred. 6 That Grad should nevertheless exhibit

the alternation is scarcely surprising: once devoicing of word-final obstruents had become an exceptionless,

“surfacey” phonological rule, every new loanword ending in a voiced obstruent would have become subject

to it automatically. But it should be clear that the chronological relationship between the acquisition of new

lexemes and the acquisition of postlexical phonological rules will not, in general, be recoverable by IR, since

it is the nature of such rules to apply to any and all lexemes regardless of their origins.

Another detail which IR cannot recover is the original identity of non-alternating phonemes in the position of

neutralization; for example, IR will not tell us whether the final stop of the invariant particle /unt/ ‘and' was

originally *t or *d. A related problem involves lexemes that ought to exhibit the alternation but are seldom

used in the form(s) in which the neutralization failed to occur. For example, the Old High German noun

‘value, worth’ and adjective ‘valuable, worth’ were both werd , and one might expect that the modern words

would be “/vert/ d /verd-/,” with underlying //d//; but in fact we find invariant /vert/, /vert-/, with

underlying // t //. Apparently the unsuffixed form, which was affected by the regular devoicing of word-final

obstruents, was so much commoner (or more salient) than all the inflected forms together that its surface

/-t/ was reanalyzed as underlying /t/ by some past generation of German speakers. Since the result is a

non-alternating paradigm, IR cannot recover this sequence of events; instead we are led to reconstruct a

historically inaccurate *vert, *vert-.

This last type of development is among those traditionally called “analogical changes” – in effect, changes

that depend (at least in part) on morphological structure, 7 as opposed to sound changes, which are strictly

phonetic or (low-level) phonological. Both types of change, occurring subsequently to a given sound change

that gave rise to a given alternation, can increase the difficulty of IR from that alternation in a variety of

ways. Two cases that illustrate these processes are the fronting of *a: in the Attic dialect of Ancient Greek

and the rhotacism of intervocalic *s in early Latin.

The historical changes that affected *a: in Attic Greek were not simple (see Szemerényi 1968; Gates 1976),

but the net result of the changes was a straightforward alternation: original *a: appears as /ⓘ:/ (merging

with original /ⓘ:/) except when preceded by /i/, / e /, or /r/, in which positions it remains as /a:/. The

distribution of /a:/ and /ⓘ:/ is quite clear, and the alternation between them is pervasive in Attic Greek

morphology; it appears in the singular endings of hundreds of “first declension” nouns and adjectives, in the

sigmatic aorists of verbs with roots ending in resonants (Smyth 1956: 173), in a small class of very common

verb stems (“mi-verbs”; Smyth 1956: 134–9), and so on. This wide distribution makes it clear that the

alternation is the result of a sound change. Both /ⓘ:/ and /a:/ appear without restriction after front vowels

and /r/, but examples of /a:/ not after a front vowel or /r/ are usually explainable as more recent

developments (see below); the principle that sound changes are regular therefore leads us to reconstruct this

change as “*a: > /ⓘ:/ except after /i, e, r/.” 8 The merger of *a: and *ⓘ: occurred in so many phonological

environments that it is sometimes not clear from internal evidence which older sound a given instance of /

ⓘ:/ reflects simply because it does not happen to occur after /i/, /e/, or /r/. That *a: was the original sound

is sometimes shown by the fact that /ⓘ:/ alternates with short /a/ (the latter appearing, for example, in the

plural endings of first declension nouns and adjectives), while original /ⓘ:/ alternates with short /e/.

Compare the following partial paradigms of some Attic Greek verbs:

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274795

Page 3 of 14

3. Internal Reconstruction : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:29 PM

/apédra:/

/éstⓘ:/

/ésbⓘ:/

‘(s)he ran away’

‘(s)he stood up’

‘it [the fire] went out’

/apodrâ:nai/

/st :nai/

/sb :nai/

‘to run away’

‘to stand up’

‘to be extinguished’

/apodraíⓘ:/

/staíⓘ:/

/sbeíⓘ:/

let him/her run away’

‘let him/her stand up’

‘let it go out’

/apodrántes/

/stântes/

/sbéntes/ ‘(upon)

‘(upon) running away

(nom.pl.masc.)’

‘(upon) standing up

(nom.pl.masc.)’

being extinguished

(nom.pl.masc.)’

Note that in the third and fourth form given for each verb the vowel after the /r/, /t/, or /b/ respectively

appears shortened, and ‘stand up’ shows /a/, like ‘run away’ but unlike ‘be extinguished’ – showing that the

/ⓘ:/ of /éstⓘ:/ and /st :nai/ was originally *a:.

This clear pattern has been complicated by a considerable number of subsequent changes, but not all have

been equally disruptive. Many new /a:/'s have arisen by two later sound changes, vowel contraction and the

“second compensatory lengthening” (2CL); however, those changes also gave rise to alternations from which

the original state of affairs is still recoverable, and for that reason they do not seriously obscure the /a:/ d /

ⓘ:/ alternation. For example, compare forms of the present tense of /tolmâ:n/ ‘dare’ with the corresponding

forms of /p h ére:n/ ‘be carrying’ :

/tolmô:men/ ‘we dare’ /péromen/ ‘we are carrying’

/tolmâ:te/ ‘you (pl.) dare’ /p h érete/ ‘you (pl.) are carrying’

The paradigms differ in two ways. In the first place, if we consider the vowels between the invariant root-

syllables /tolm-/, /p h er-/ and the invariant endings /-men/, /-te/, it is clear that the vowels found in

‘dare’ are both longer and lower than the corresponding vowels of ‘be carrying’, but that they resemble the

latter to some extent (‘we …’ always shows a back round vowel, for example). Second, the accent falls on

the third syllable from the end of the word in ‘be carrying,’ but on the second syllable from the end in ‘dare.’

When we consider also the fact that there is a noun /tólma/ ‘courage’ obviously related to ‘dare,’ it becomes

clear that the most plausible and economical way to account for all these phenomena is to posit earlier

forms *tolmáomen for /tolm :men/ and *tolmáete for /tolmâ:te/. The /a:/ of the latter form, then, resulted

from contraction of the sequence *ae; and because we can explain its appearance by such a development, it

does not seriously obscure the pattern according to which we expect /a:/ after /i, e, r/ but /ⓘ:/ elsewhere –

a pattern which, we now see, applies only to those older (“original”) instances of *a: which existed before

vowel contraction had occurred. In fact, the larger pattern now permits us to reconstruct the relative

chronology of the sound changes involved: the change “*a: > /ⓘ:/ except after /i, e, r/” must have run its

course before the change “*ae > /a:/” produced new /a:/s, since those new /a:/'s did not undergo the

former change. 9

The 2CL can likewise be recovered from the patterns of alternation to which it gave rise. Consider the

following partial noun and adjective paradigms:

‘guard’

‘serf’

‘black’

‘(upon) standing up’

nom.sg.masc.

/p h ülaks/ /t h

:s/

/méla:s/ /stá:s/

nom.pl.masc.

/p h ülakes/ /t h

:tes/

/mélanes/ /stántes/

The invariant endings are nom.sg. /-s/ and nom.pl. /-es/, and the stem of ‘guard’ is likewise invariant

/p h ülak-/; but the other stems participate in various alternations. The stem of ‘serf’ appears as / t h

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274795

Page 4 of 14

3. Internal Reconstruction : The Handbook of Historical Linguistics : Blackwell Reference Online

12/11/2007 03:29 PM

/p h ülak-/; but the other stems participate in various alternations. The stem of ‘serf’ appears as / t h : t -/ 10

when a vowel follows, but as / t h :-/ when followed by /- s /; and since fuller study of the grammar shows

that /- t -/ was not normally inserted between vowels in Greek, we must conclude that the nom.sg. was

originally *t h :ts, and that stem-final *-t- was lost when *-s followed immediately. By a similar line of

reasoning we conclude that the nom.sg.masc. of ‘black’ was originally *mélans, and that the sequence *ans

became /a:s/ by the 2CL. Finally, in the nom.sg.masc. of the participle of ‘stand up’ both changes have

occurred – first the loss of *-t-, then the 2CL – and the development can be reconstructed as *stánts >

*stáns > /stá:s/. This accounts for numerous additional cases of unexpected /a:/; and of course the fronting

of *a: must likewise have run its course before the 2CL occurred.

One would expect greater disruption to have resulted from changes that tended to obscure the /a:/ d /ⓘ:/

alternation and are not reconstructible . For example, already in the sixth century BCE, Attic possessed two

noun stems which obviously belong in the first declension 11 but show stem-final /ⓘ:/ < *a: after /r/,

/kórⓘ:/ ‘girL' and /dérⓘ:/ ‘necklace.’ Comparative evidence from other dialects shows that these words

originally had a *w before the stem-final vowel (cf. Arkadian <korwa> 12 ‘Persephone’, <derwa> ‘ridge, spur

(of a hill)’), and it is reasonable to infer that the *w had not yet been lost in Attic when the fronting of *a:

occurred (so that the *a: was not then immediately preceded by *r); but since *w was subsequently lost

without a trace in Attic, IR cannot recover those events. A similar case is /kórrⓘ:/ ‘temple (of the head)’ :

comparative evidence shows that /kórrⓘ:/ < *kórsⓘ: (preserved unchanged in East Ionic) < *kórsa: (preserved

unchanged in East Aiolic), but that development could not be recovered from Attic evidence alone. Still other

cases of the same sort are Attic /póa:/ ‘grass’ < *poía: and /stoâ:/ ‘colonnade’ < *stoiá: (both preserved

unchanged in Doric dialects). At least one analogical change contributes further examples. A coherent class

of pairs of present and aorist stems show the expected pattern pres. /-aíne:n/, aor. /-ǎ:nai/ d /- :nai/, the

vowel alternation in the aorist stem depending on the preceding sound:

Present

Aorist

/hügiaíne:n/ ‘be welL’ /hügiâ:nai/ ‘get welL’

/ksⓘ:raíne:n/ ‘be drying (it) out’ /ksⓘ:râ:nai/ ‘dry (it) out’

/p h aíne:n/ ‘show (continually)’ /p h :nai/ ‘show’

/sⓘ:maíne:n/ ‘indicate (continually)’ /sⓘ:m :nai/ ‘indicate’

/ksaíne:n/ ‘scratch (repeatedly)’

/ksⓘ:nai/ ‘scratch (once)’

/k h alepaíne:n/ ‘be offended’

/k h alep :nai/ ‘take offense’

and so on (the list could be extended considerably). But toward the end of the fifth century BCE we find a

few aorists with /a:/ not after /i, e, r/:

Present

Aorist

/koilaíne:n/ ‘be hollowing

out’

/koilâ:nai/

‘hollow out' (Thucydides 100.4.2)

/kerdaíne:n/ ‘gain’

/kerdâ:nai/ ‘make a profit’ (Andocides 1.134; 13 Xenophon, Apology of

Socrates 9)

The /a:/ of these aorists can only be the result of analogy with other aorists in which /a:/ is etymologically

justified (though exactly which verbs provided the model for the analogical change is not clear).

But all these exceptions together do not suffice to obscure the pattern from which the fronting of original

long *a: can be reconstructed internally, for a simple reason: there are hundreds of forms which show the

expected alternation, and very few which fail to show it. In most cases IR can only identify these exceptional

forms, not explain why they fail to behave as expected; and it should also be obvious that IR cannot tell us

http://www.blackwellreference.com/subscriber/uid=532/tocnode?id=g9781405127479_chunk_g97814051274795

Page 5 of 14

Plik z chomika:

Inne pliki z tego folderu:

Inne foldery tego chomika: