A grammatical – morphological or syntactic – construction has a certain internal structure. At the level of syntax, its components contract syntactic relations to each other which generate syntagmas and a hierarchy among them. At the level of morphology, its components form syntagmas, too; and these may form a hierarchy, too. At any rate, such a construction contains boundaries at certain positions.

Reanalysis of a grammatical construction is a process which changes its syntagmatic composition. An internal boundary is either added or deleted or shifted. The shift variety may be represented by the following schema:

a | b c → a b | c

From a lingistic point of view, reanalysis amounts to a “rebracketing”. The important thing to note is that the sequence and form of the components a b c is not affected by the reanalysis. Consequently, the reanalysis itself is imperceptible, even for linguists. A reanalysis becomes perceptible and, thus, linguistically provable if a subsequent (perceptible) change happens which presupposes the reanalysis carried out.

A stock example of reanalysis at the morphological level concerns the Old English plural morph -es. The rows of the following schema show two different reanalyses involving this morph. The first example affects a loan word from Old French which ends in [z], which happens to be one of the allomorphs of the Old English plural morpheme. Since cherries are mostly found in the plural, a reinterpretation of this final segment as a plural morph suggests itself. In this case, the input to the process contains no morphological boundary, while the output inserts one.

The example of the last line concerns an Old English noun which does comprise a plural suffix. This is, however, a plurale tantum; so there is little motivation to preserve this plural feature. The invariable form is reanalyzed as a singular form. Thus, an existent morphological boundary is deleted.

The emergence and the absorption of the internal morphological boundary is, of course, imperceptible. It becomes unmistakable as soon as the subject-verb agreement of these nouns is adapted, i.e. when cherries conditions plural agreement and truce conditions singular agreement. In the latter case, the changed orthography also manifests the changed morphology.

Two aspects of these examples deserve being generalized:

• Both examples concern the plural form of a noun. The first displays emergence of an internal boundary; the second displays deletion of an internal boundary. The examples are, consequently, exactly converse. The conclusion is that reanalysis is not a directed process. If there are two alternative structural analyses of a given expression, either may be reanalyzed as the other.
• For both of these reanalyses there are models available in the language. The existent, phonologically and semantically similar, plural form berries motivates the reanalysis by which the form cherries contains the same suffix. The existent, phonologically and semantically similar, singular form peace motivates the reanalysis by which the form truce is monomorphemic. Without such analogical models, the two reanalyses would be inexplicable, the more so as they are exactly opposite. Thus, reanalysis presupposes an analogy.

A similar pair of examples, this time from the level of nominal syntax, concerns the indefinite article in front of a noun. The rows of the following schema again show two different reanalyses involving its allomorphs.

The example in the first line restructures the combination of the /ə/ allomorph with a noun starting with /n/, while the second example restructures the combination of the /ən/ allomorph with a vowel-initial noun, in the opposite sense. These examples illustrate the shift of an internal boundary. Both allomorphs of the indefinite article and the rule conditioning their distribution are well-established in the language system. These reanalyses would be inexplicable were there not the models of other, phonologically and semantically similar, nouns showing the target structure.

This second pair of examples confirms the two conclusions drawn on the basis of the first pair:

• Reanalysis is an undirected process.
• Reanalysis is motivated by a model.

Reference

Detges, Ulrich et al. 2021, ‘Positioning reanalysis and reanalysis research’. Journal of Historical Linguistics 5: 1-49.