Word change in morphology

3.1. Morphology-Phonology

In this subsection we will take a closer look at changes at the morphology–phonology interface—so first we should define this kind of change. We can say that such a change occurs when a phonological rule becomes opaque because of a phonological change and speakers can no longer identify the source of this change and therefore reinterpret the rule as a morphological property. This process is also called dephonologization. One of the most cited instances of this change are the irregular plural forms in the Germanic languages that developed due to i-mutation.

Table 1 shows the singular and plural forms of the set of nouns in pre–Old English (OE) times (marked with an asterisk) that in present-day English show irregular plural forms built by the process of i-mutation. I-mutation is a regressive assimilation whereby the final [i] (or [j]) in words like *manni marking the plural raised the stem vowel to / e /, which thus partially assimilated to / i / (another option was fronting as can for example be seen with mice). In pre-OE we have an alternation between a singular form *mann and a plural form *manni. In the course of time, the i-ending, i.e., the trigger for the change, was first reduced to schwa and finally to zero, and only the product of the assimilation process was retained: two different morphological forms, man and men. As a result, the vowel alternation obtained the function of marking plural. In other words, the once phonologically conditioned forms were reanalyzed as conditioned by the categories defining the morphological identity, in our case PLURAL. In German, this process even led to the rise of new phonemes and graphemes, which are called “Umlaut(e)”: <ä>, / ɛ / as in Männer ‘men,’ <ö>, / ø / as in Wölfe ‘wolves,’ and <ü>, / y / as in Füße ‘feet.’ These three examples show that the umlauted stem vowel is not the only marker for PLURAL present: rather it is the “Umlaut” + the suffix —er or —e. However, there are other cases like Mutter-Mütter ‘mother-mothers’ or Vater-Väter ‘father-fathers’ that mark PLURAL only with the “Umlaut.” So, often original phonological processes are undergoing the process of morphologization and add to what is known under the term “non-concatenative” morphology. Another example of this sort can be seen in the preterite and past participle forms of irregular verbs that are built by ablaut like sang, and sung, etc. (again note a difference from German, where in addition to the ablaut, the circumfix ge-sung-en marks the past participle).

Whereas in PDE only residues of the process of i-mutation are left, including the irregular plural forms discussed above, in German the “Umlaut” also marks the comparison as in arm-ärmer ‘poor-poorer,’ the subjunctive with würden, the formation of the diminutive with the suffixes -lein and -lich (Bruder ‘brother,’ Brüderlein/Brüderchen ‘little brother’), and denominal adjectives with the suffix -lich ‘-ly’ (based on nouns that mark the plural forms with “Umlaut”) as in väterlich ‘fatherly.’ We will see in Section 4 that a reason why PDE hardly exhibits any traces of i-mutation could be that some of these irregular plural forms like pre OE bēc ‘books’ were made regular in the course of time by processes of analogy.²

The example of i-mutation illustrated above is an internal change, but the morphology of a language may also change due to foreign influences, i.e., language contact. One quite interesting example is the loss of prefixes in the history of English conditioned by changes in the prosody, as reported by Molineaux (2012). One of the main and hotly debated issues in historical linguistics is the loss of a remarkable number of native prefixes during the time spanning Old English and Middle English (ca. 800–1500). For example, the OE prefixes or— (oreald ‘very old’), ed- (edbyrdan ‘regenerate’), tō— (tōberstan ‘to burst apart), ymb- (ymbgang ‘circuit’), and others were lost in ME or no longer productive. All kinds of explanations have been put forward, for example, changes in the syntax and semantics as well as general processes of grammaticalization and changes triggered by contact. Kastovsky (1992) and others saw a causal link between this loss and the linguistic consequences of the Norman Conquest and claimed that the massive borrowing of material from Anglo-Norman led to the weakening of native word-formation as such. What cannot be explained according to Molineaux, however, is the fact that only some of the prefixes were lost while others were preserved. This is why he takes another stance and proposes a prosodic approach where language-internal factors as well as language contact play a role.

Based on a comparative corpus study of prefixation at the relevant stages, i.e., Old English and Middle English, Molineaux pursued the following questions (to exclude semantic and syntactic triggers): Are the OE prefixes still productive in ME? Is the category of prefixed words relevant in the loss? Is the separability of prefixes crucial for their survival? First, he found a correlation between the preservation of prefixes and their syllable weight. More precisely, disyllabic and light monosyllabic prefixes survived because they were compatible with the prosodic properties of ME roots. However, stronger heavy, stressable monosyllabic prefixes at the left edge of a word were lost, because these properties made them incompatible with the prosodic properties of ME roots.

In OE the major word categories allow monosyllables under the condition that they have a long vowel and/or a coda consonant (e.g., wā ‘woe’). In polysyllabic words, primary stress usually falls on the first syllable, and in the case of prefixed words we find the following general rules: nominal and adjectival prefixes and separable verbal prefixes take primary stress (e.g., ‘ūtgān ‘go out’), and inseparable verbal and adverbial prefixes are unstressed (e.g., unbindan ‘loosen’; for details and exceptions see Molineaux, 2012). Secondary stress is found on the second element in compounds or on bound suffixes deriving from free morphemes (e.g., cýne-dṑm ‘dominion’). (For details, see Molineaux, 2012; Dresher & Lahiri, 1991).

In ME the prosodic conditions had changed, which especially affected the interaction of primary and secondary stress at the prefix-root boundary. Due to (internal) changes in weight in OE and ME times, adjacent stress on prefix and root caused a clash that was no longer acceptable in ME. This conflict had to be resolved, and that was accomplished by eliminating the heavy monosyllabic prefixes responsible for this clash. In OE prefixed words like ófþànc ‘envy’ did exist despite the clash, but they were lost in ME because of the development of a constraint that did not allow clashes at the prefix-root boundary. What led to the rise of this constraint was, among others, the adoption of Romance loanwords which displayed a regular alternation between stress and non-stress in polysyllabic lexemes. Thus, a loanword from OF like còuntenánce would have been adapted to fit the Germanic stress pattern by reversing primary and secondary stress: cóuntenànce. This pattern would have been in line with many native polysyllabic words (e.g., éaldor-mànn ‘elderman,’ óferfèng ‘buckle’). The constraint against the succession of stressed syllables, however, strongly dispreferred heavy native prefixes, and as a result they either dropped out of the language or were preserved in lexicalized lexemes. The merit of Molineaux’s analysis is that it can explain why some prefixes were preserved and others lost as opposed to previous analyses. It also shows quite nicely that changes in morphology can be prosodically conditioned and that internal as well as external factors may play a role in language change.

The example of the “Umlaut” has shown that there are phonological phenomena that develop into morphological phenomena, thus crossing component boundaries. In the case of the loss of prefixes, we have seen that changes in the prosody of English led to changes in its morphological inventory, so conditions part of the phonological component affected morphology. Both types of changes illustrate the interaction between phonology and morphology.

3.2. Morphology-Syntax

In this section we will deal with changes at the morphology–syntax interface. Generally we can say that new morphology may arise from the reinterpretation or reanalysis of once syntactic structures. This process is also called desyntactization in the literature (especially in the context of grammaticalization; see Booij, 2012). Standard examples of this type of morphological change are the rise of inflectional endings in Romance languages (see, e.g., Roberts, 1993). First, we will take a closer look at the development of inflectional future marking in French from the Latin analytic construction with habere ‘have’ + infinitive.

In Latin, future tense was expressed by inflectional endings, for example, laudabimus meant “we will praise” marking 1st person plural. The syntactic construction that came to compete with it was a combination of the verb habere ‘have’ + infinitive, so habemus laudare meant something like “we have to praise,” which had a flavor of obligation and future orientation. In Late Latin the order of infinitive + habere started to occur (laudare habemus), which was the basis for the development of French louerons (louer + avoir). At this point of the development, fusion across morpheme boundaries became possible, and the once-free verb avoir was reanalyzed as an inflectional ending on the preceding verb. The paradigm of the French future form of louer (Table 2) shows the connection between the future endings and the forms of avoir.

Another example that illustrates that morphological structures can arise from syntactic structures is the case of nominal compounding in the Germanic languages. In Old High German (OHG, ca. 700–1050) two types exist, one where the nominal non-head exhibits a compositional vowel, as, for example, in tag-a-sterro ‘morning star,’ and one where the first element bears the genitive case and shows a syntactic relation between the nominal non-head and head as in senefes corn ‘mustard seed’ (see Demske, 2001; Trips, 2006). In the following we will focus on the latter type.

It is not entirely clear that senefes corn is a compound—it could also be a syntactic phrase where the prenominal genitive modifies the nominal head. Evidence for this ambiguity comes from the semantics of these “phrases”: at this time both a non-referential and a referential reading of the first constituent is possible. Following Demske (2001), in Early New High German (ENHG, ca. 1350–1650) the nominal system underwent a number of changes that led to the postposing of genitival complements. This in turn led to the reanalysis of the existing structures as morphological pattern and the concomitant rise of a “new” word formation pattern. For Demske this reanalysis can be described as the loss of a word boundary, but it also includes a structural reinterpretation, i.e., the speaker has clear evidence that entities of the type N_GEN + N have properties like non-referentiality and compound stress that are attributed to morphological structure. Most importantly, the former suffix marking genitive is reinterpreted as a linking element (in German, Fugenmorphem), which is evidenced by data where this suffix now occurs in genitive compounds where it is non-paradigmatic as in des Bawers-man ‘the farmer man’ (we would expect to find des Bauern-mann). A further piece of evidence for the reanalysis is the rise of the linking element in copulative compounds in ENHG like alle Manspersonen ‘all man’s persons,’ where we have a coordination relation between the first and second noun that cannot be expressed syntactically (see Demske, 2001, p. 311). Interestingly, the linking-s has been increasing since ENHG, alternating with and gradually replacing the zero element (Seminar(s)arbeit ‘term paper’). As Nübling and Szczepaniak (2011) note, today it is the most frequent and productive linking element in German, depending directly on the phonological quality of the preceding noun. It seems that in addition to a process of desyntactizization, we may witness even a further step in the direction of phonologization. I will briefly comment on this observation.

Recently, it has been observed that compounds with the non-paradigmatic linking-s outnumber those with a paradigmatic linking element. This trend is especially evident with feminine nouns like Liebesbeweis ‘proof of love,’ Hilfsangebot ‘offer for help,’ and Arbeitsplatz ‘work place’ (see above). This can be explained by the special status of this element: among the set of linking elements in German it is the one that has developed farthest from its origin as an inflectional ending. In doing so, it has gained a number of new functions that determine its propagation. Generally, the occurrence of —s depends on the phonological properties of the first part of the compound. More precisely, this linking element is sensitive to the phonological quality of the first noun and improves the phonological structure of words that deviate from the ideal phonological structure, which is either a monosyllable (e.g., Wort [‘vɔrt] ‘word’) or a trochee (e.g., Wolke [‘vɔl.kə] ‘cloud’) in German. For example, a noun with an unstressed prefix like Verkáuf ‘sale’ deviates from the ideal phonological structure. Compounds with these prefixed nouns as first part (e.g., Verkáufsgespräch ‘sales talk’) therefore regularly occur with the linking-s because it improves the structure of the compound in that it strengthens the right edge of these words (Verkaufgespräch is much less preferred). Similarly, compounds that contain non-native nouns of Romance or Greek as the first element frequently display the linking-s (Abitursfeier, Religionsunterricht) since these nouns deviate from the ideal phonological structure: they show word-final stress (Abitúr) or full stress on the suffix (Rèligíon ‘religion’). Overall, the development of the linking-s in German illustrates a process that first affected the syntax–morphology interface and then came to affect the morphology–phonology interface (for studies with a focus on statistical and psychoacoustic aspects, see Krott, Schreuder, Baayen, and Dressler (2007) on German compounds and Kuperman, Pluymaekers, Ernestus, and Baayen (2006) on Dutch).

Concerning the development of nominal compounds in English, we find a similar picture: as in German, both types of compounding discussed above occur—the type with a compositional vowel (hilderinc ‘battle man’) and the type with a genitive marker on the first noun (cinnesmen ‘kinsmen’). In the latter case we find evidence for structural ambiguity because of the contrast between a referential and non-referential reading. What distinguishes the development of the compounds with genitive markers in the two languages is the loss of inflectional morphology in English between OE and ME. Allen (1998, 1995) has shown that in the 13th century agreement inflection had become optional and, among other things, led to case syncretism of vocalic genitive endings, and ultimately to an unclear status of these endings. Accordingly, different forms of the “prenominal genitives” coexisted:

(1)

Clearly, at this time genitive endings were no longer governed by transparent syntactic rules; examples like these are lexicalized, i.e., compounds. As a result of the reanalysis of the genitive marker as a linking element, forms incompatible with their nominal paradigms start to occur, for example, OE land-man now occurs as ME londes-men. In addition, in ME copulative compounds like burgess tun ‘castle town’ start to occur as a productive word-formation pattern (which parallels the development in German). Thus, the development in English is another example for a reanalysis of syntactic structures as morphological structures. The difference from German is that the linking element that existed in OE and ME times fell out of use, probably due to the loss of inflection, and therefore never had the potential to become part of productively generating compounds, as, for example, the German linking —s did.

The discussion of the rise of inflectional endings in French supports Givón’s (1971, p. 413) observation that “today’s morphology is yesterday’s syntax.” It is a case that shows how syntactic structure may develop into morphological structure acquiring morphosyntactic function, i.e., becoming part of inflection. The development of N+N compounds with the linking element —s has revealed that syntactic structure may also become part of word-formation processes and acquire even prosodic function. Although both phenomena can be analyzed as the crossing of the boundary from syntax to morphology, the outcome is clearly different.

3.3. Morphology-Semantics

In this section we will take a closer look at the interaction between morphology and semantics. The examples given so far have made it clear that morphological phenomena are often tightly linked to phonology and syntax. Now what do we expect to find when it comes to morphology and semantics? Let us begin with an example:

(2)

The word shrimpburger is a primary example of reanalysis within the word. Reanalysis is a development that alters the structure of a form because this structure is or becomes ambiguous. In our case, the basis of the reanalysis was the word hamburger, which originally denotes a person from the city of Hamburg. But at the beginning of the 20th century in the United States a type of food called Hamburger steak became popular, which was often shortened to hamburger. According to the Oxford English Dictionary (OED), burger is a terminal element denoting a roll or a sandwich that contains the foodstuff specified in the first element. So evidently at some point in time (or more precisely in the 1940s in the United States as stated by the OED), speakers reanalyzed the structure of [hamburger] as the new structure [ham][burger], which then made way for many new formations, for example, krautburger, kosherburger, soyburger, veggieburger, beefburger, (bacon-)cheeseburger (all found in the Corpus of Contemporary American English, COCA), and shrimpburger from the example above. What the reanalysis of hamburger shows is that this change on the level of word structure was triggered by a new semantic analysis of innovative speakers, so morphology and semantics interacted and brought about the change.

The development of the suffix —gate is a similar but even more interesting case (Bauer, Lieber, & Plag, 2013 call elements like —gate and —burger splinters). The suffix originated in the famous Watergate scandal in the early 1970s. It refers to the political scandal in which members of U.S. President Richard Nixon’s Republican administration were caught breaking into the headquarters of the Democratic Party at the Watergate Building in Washington, D.C. From this time on, the word Watergate has been associated with (this) scandal, and when new scandals comparable to Watergate became public, the form —gate has been used: Irangate, Koreagate, and many others. The interesting thing is that speakers associated the second part of the word Watergate with a scandal, and so they built other words with —gate on the basis of the rule X-gate with the meaning “scandal involving X and X being similar to Watergate” (see Joseph, 2001, p. 360). Not long ago, Fox commentator Glenn Beck announced that —gate is on its way of dying out:

And tonight, I want to mark the passing of someone you know very well, only 37 years old. Perhaps the most overused suffix in all of media history. Tonight, let me be the first to announce the death of “gate.” Rest in peace. Since the Watergate scandal of 1972, everything—everything that’s gone on in this country when it revolves around Washington, has been dubbed “gate.” There was Iran-gate, Monica-gate, hooker-gate, Jersey-gate, trooper-gate, one, two and three. But we have turned the corner, America. The paradigm has changed. Forget the word “gate.”

(COCA, Fox Beck, August 14, 2009)

But headlines like “Trump-Gate continues” or “Hilary’s email-gate” ) commenting on current political events clearly belie this development heralded by the self-proclaimed language expert.

The discussion of the examples of burger and —gate highlighted the development of new word-formation patterns on semantic grounds. In both cases reanalysis of morphological complexes occurred that led to the rise of new combination forms or even suffixes. However, it has been observed that there are more general processes whereby derivational suffixes develop from the second part of compounds that used to be heads of syntactic phrases. Here we take a brief look at the rise of some such suffixes building abstract nouns in the history of English.

In Trips (2009) the rise of the suffixes —hood, —dom, and —ship are investigated. Some of the questions she pursues are: What’s the origin of derivational suffixes? Are there general developmental stages that can be found for all derivational items? What is the trigger for the development of derivational items?

Marchand (1969) noted that suffixes come into existence via three stages:

(3)

The following examples give evidence for the independent character of —hood, —dom, and —ship in OE times:

(4)

Back then hād denoted ‘status, office, rank,’ the noun dōm denoted ‘authority, judgement,’ and scipe denoted ‘(resultant) state, condition.’ These nouns were often found as heads of compounds, and because they were relational nouns, they required an argument which in these cases was of the stage-level predicate type (see the examples in (6)). Trips assumes that it was exactly this property that was the prior condition for the nouns to develop into suffixes.

In the course of time, the three elements developed a number of meanings through metonymic shifts on the basis of their salient meanings, leading to their polysemous character. For example, —hood acquired the meanings ‘state’ (boyhood) and ‘time’ (childhood), to —dom’s salient meaning were added the meanings ‘territory, realm’ (kingdom), and —ship acquired further meanings by metonymic shifts like ‘forms of address’ (lordship) and ‘skill/art’ (horsemanship). All aspects of meaning of the three elements derive from their salient meanings and their semantic development, including similarities and differences between them. In more abstract terms, the suffix —hood can be associated with the feature [state], —dom with the feature [process], and —ship with the feature [achievement]. These features can be seen as a diachronic imprint since they represent the development of the suffixes and define the differences between the suffixes. Crucially, at the stage when their salient meanings had been extended to further meanings and they had generally acquired a more abstract meaning, they became suffixes. Under this approach, the semantic development of the suffixes led to their development from free morpheme to bound morpheme. The semantic differences between the three suffixes become most evident in derivations with a base like dog as in doghood, dogdom, and dogship. According to the OED, doghood denotes “the status of a dog” (“Bo reverts to a wholesome dogginess. It is agreed between us that I won’t call him out of his doghood”; Christian Sci. Monitor (Nexis) 19, July 21, 1982), whereas dogdom denotes “the realm/world of dogs” (“It was clearly … an infinitely more sophisticated representative of dogdom than the elusive yellow dog”; K. Atkinson Emotionally Weird, 2001, p. 266). From what was said above, derivations with —ship include the meaning of “title of N,” and dogship exhibits exactly this meaning (“His dogship was … becomingly thankful, but … not overly demonstrative in protesting the old underlying affection for his … master”; J. Fredston Snowstruck iii. 2005, p. 83).³ (For a different view, see Bauer, Lieber, & Plag, 2013.)

The three case studies presented here have shown an interaction between morphology and semantics. More precisely, in all the cases discussed, the semantic development, involving structural reanalysis or not, has triggered the rise of new morphological forms. Actually, it is quite hard to see how semantics could not be involved since lexical entities are, after all, form-meaning relations. It seems that due to this tight link morphology and semantics stand in a more direct relation than syntax and morphology and phonology and morphology do. If this assumption turns out to be correct, a multimodular approach would have to reflect this.

4.1. Analogy

In broad terms, analogy is a “synchronic or diachronic process by which conceptually related linguistic units are made similar (or identical) in form, especially where previous phonetic change had created a variety of forms. Analogy is often regarded as the result of the move towards economy of form or as a way to facilitate the acquisition of the morphological forms of a language” (Bußmann, 1996, p. 21). Further, it is generally assumed that analogy is mainly concerned with the link between form and meaning, which combine to express meaningful units. Analogical processes maintain this link by keeping sound structure, grammatical structure, and semantic structure in line, i.e., keeping this relation transparent, especially if sound changes might have led to opacity (for other definitions, see Anttila,1972, 1977; Hermann, 1931; Mańczak, 1958; Postal, 1968).

At least since Neogrammarian times scholars interested in the workings of language have dealt with analogy. Stating that only historical linguistics could be truly scientific, the Neogrammarians investigated sound changes and postulated the Regularity Hypothesis, which assumes that sound change is totally regular and inviolable. Since they also believed that changes operate with “blind necessity,” they saw an interaction between (regular) sound change and analogy expressed in Sturtevant’s Paradox (Sturtevant, 1947, p. 109):

Phonetic laws are regular but produce irregularities. Analogic creation is irregular but produces regularity.

So the task of analogy is to clear up irregularities that were produced by sound change. A good example of their way of thinking can be seen in how the Neogrammarians treated the irregular forms of nouns built by i-mutation discussed in Section 3.1. They did not treat forms like feet or men as synchronic irregularities but as diachronic regularities. By looking back at the pre–Old English forms they showed that plural marking with final —i (or —j) was regular and that in the course of time irregularity was created by sound changes. Examples like these stressed the primacy of historical linguistics, on the one hand, but also the importance of analogy, on the other hand. Other irregular plural forms like pre–Old English *bēc (sg.*bōc), which we would expect to find today as beek (or beech) do not exist. Instead, the form books occurs due to analogy:

(5)

This is an example of four-part analogy which is based on a proportional model indicated in (5a). Proportions generally encode a relationship between four entities and generalize a morphological pattern of given forms (stone: stones) to other forms (book: books) that previously did not show this pattern. Felicitous conditions for the success of analogical extension are that the forms that change are synchronically derived forms (e.g., ‘plural’ vs. ‘singular’) and that the pattern is productive—see Kuryłowicz (1949), where he proposes a number of “laws,” and comments in Hock (2005) and Anderson (2015). Forms that were not regularized are, for example, the irregular plural forms oxen or sheep, which is evidence for the fact that analogy works in a sporadic fashion.

The second systematic type of analogy is analogical leveling. Whereas analogical extension relates to patterns, analogical leveling relates to paradigms (therefore also the term paradigmatic leveling).

This process eliminates irregularities across paradigms as can be seen in Table 3. The vowel alternations of the past tense forms are eliminated, but the distinction of vowels between present tense and past tense forms is retained. Analogical leveling is most successful if the morphological representation of important distinctions (like singular and plural) is not affected.

A much less frequent type of analogy is folk etymology, one example of which was discussed in the introduction. PDE crayfish or sand-blind (from OE sām-blind ‘half blind’) show that speakers aim at assigning transparent compound structure to words. Other less frequent cases are blending (breakfast + lunch = brunch), contamination (OF male, femelle = ME male, female), and univerbation (OE dægesēage > ME dais ei(e) > PDE daisy) (see Campbell, 2004).

We have seen that the Neogrammarians postulated a clear-cut distinction between sound change and analogy (Sturtevant’s Paradox). Crucially, they believed that analogical processes should be seen as a response phenomenon to the effects of sound change. There are, however, examples that speak against this view, one being the regular plural formation campuses instead of the irregular Latin campi (among many others). Clearly new, regular morphological forms are constantly built, especially in the context of borrowing, that cannot be seen as reactions to sound changes. Many scholars have disagreed with this claim and favored either the “grammatical conditioning” or the “sound change as analogy” approach (see Hock, 2005). According to the first approach, analogical processes may block or condition sound change (see, e.g., Anttila, 1972; Postal, 1968), whereas according to the second approach sound change is a subtype of analogy (see, e.g., Schuchardt, 1885; Sturtevant, 1917). Hock (2005) notes that the long-term debate in the literature has shown that neither approach can account for all the relevant empirical facts. Therefore, he proposes a definition of analogy in a broader sense—as an extension of linguistic patterns. Moreover, he assumes a continuum of such changes along which sound change, morphophonemic extension, rule reordering and extension, all types of analogical changes, as well as some types of semantic change are found. The domains in which these changes apply define their properties: the broader the domain, the more regular and more systematic the changes are—sound changes (as defined by the Neogrammarians) are not constrained by non-phonetic/non-phonological factors and thus very likely to be regular. Analogical extension and leveling apply to large classes of entities (paradigms and sets of lexical items) and are therefore relatively systematic but less regular than sound changes. Changes like folk etymology apply only to individual words and are, as a consequence, quite sporadic. Semantic changes are most sporadic because they are generally restricted to small subsets of the lexicon. A further important point is that this continuum correlates with the degree of morphological and semantic information these changes are sensitive to. Sound changes are not sensitive to morphology or to semantics, analogical extension and leveling are more sensitive to morphological form than to semantic information, sporadic types of analogy like folk etymology are highly sensitive to semantics, and semantic change is most sensitive to this type of information.

Another approach is taken by Anderson (2015, p. 282), who assumes that there is no need for a special mechanism of analogy that operates on the level of morphology. Rather, phenomena like analogical extension, leveling, and the like can be explained by the mechanisms of inductive, deductive, and abductive change that apply to other domains of grammar as well and that play a crucial role in the process of grammar transmission from generation to generation.