Abstract-noun forming suffixes

Under Fabb's approach the deverbal suffixes -age (steerage), -al (betrayal) -ance (annoyance), -ment (containment), and -y (assembly) all share a selezionai restriction that does not allow attachment to already suffixed verbs. A closer look at the data reveals, however, that such a selectional restriction is rather ad hoc and conceptually inadequate.

The only verbs that contain a suffix in English are those that end in -ify, -ize, -ate, and -en. Thus it is only to one of these verb types that the above mentioned suffixes can potentially attach. Let us look at derivatives in -ify, -ize, and -ate first, because they share certain characteristics that distinguish them from -en derivatives with respect to the kinds of derivational processes by which these derived verbs are related to their corresponding abstract nouns.

Standard textbooks or dictionaries, if not the native speaker's intuition, tell us that verbs in -ify end up regularly with -ification as the nominalizing suffix attached to the root, verbs in -ize take -ation, and verbs in -ate take -ion. Following the common assumption that these three suffixes are phonologically conditioned allomorphs of -ation, we can simplify our statement and say that verbal derivatives in -ify, -ize, and -ate all take -ation as their nominalizing suffix. This is evident if we look at the impossible derivatives given in (3). The suffixes -age, -al, -ance, -ment, and -y are just as impossible as the inappropriate phonologically conditioned allomorphs:

My claim is that the failure of these five nominalizing suffixes to attach to already suffixed verbs is not due to a selectional restriction associated with them, but a natural consequence of the organization of derivational morphology in general, and verb nominalization in particular. I propose that it is the verbal suffix, hence the base, that is responsible for the choice of the nominalizing suffix that can be attached to it. Under this view we are dealing not with selectional restrictions of the nominalizing suffixes (i.e. "affix-driven" restrictions), but with restrictions imposed by the base, in this case the suffix of the base word, on the kind of suffix it can take. Let us call these kinds of restrictions 'base-driven'.

The postulation of base-driven restrictions in this case is conceptually superior to the affix-driven approach, since the possible combinations of suffixes have to be posited anyway. The derivational gaps then follow naturally. If we, however, conceptualize the restrictions as affix-driven, we can only predict the impossible combinations involving -age, -al, -ance, -ment, and -y, but we fail to predict the possible ones, featuring -ation. Let us look at this argumentation in more detail.

Under an affix-driven approach we have to state that -ion attaches to, among others, verbs ending in -ate, that -ification attaches to the stems of verbs in -ify and -ation attaches to verbs in -ize. Such a statement may be adequate to account for existing derivatives involving these suffixes, but there is no way to predict, let alone explain, why the derived verbs do not take -age, -al, -ance, -ment, or -y as suffixes. In order to save this generalization under an affix-driven approach, one would have to state additional selectional restrictions of just these suffixes. Under a base-driven approach, the fact that -age, -al, -ance, -ment, or -y do not attach to the derived verbs under discussion can be predicted and need not be stipulated, since the verbal suffixes in question take only the competing -ation suffix (and the general suffix -ing to which no restriction seems to apply).

With regard to verbal bases in -en I will argue that all nominalizing suffixes in question can in principle be attached to it, provided that other fac tors (e.g. of, for example, semantic or phonological provenance) do not intervene. At closer inspection, Fabb's generalization turn out to be not completely accurate. While he seems to be right in his observation that no derivatives in -enage, -enal, -enance, and -eny can be found in dictionaries or seem possible according to today's native speakers' judgments (cf. for example the unacceptability of *darkenance, *blackenage, *strengthenal, *wideny)¹, a small number of -enment forms is attested (e.g. enlightenment). Before we turn to the latter cases we will deal with the question as to how the unacceptable and unattested derivatives can be ruled out without making reference to a selectional restriction of -en.

There are in principle two ways to account for the combinatorial properties of -en with regard to abstract-noun-forming suffixes, of which one involves an etymological constraint, the other paradigmatic forces.

Turning again to the case of verbal -en derivatives, which are [- Latinate] due to the native suffix and its exclusive attachment to native bases, the etymological constraint predicts that they cannot combine with [+ Latinate] suffixes. The Latinate Constraint can therefore nicely account for the fact that -ation, for instance, may not attach to -en. Deverbal -ation almost exclusively occurs with non-native stems, which warrants its characterization as [+ Latinate] (see also, for example, Marchand 1969, Jespersen 1942, Giegerich 1998). Only few counterexamples seem to be attested, of which Marchand (1969:260) gives flirtation, starvation, botheration, backwardation (a stock exchange term), jobation (colloquial, OED), plus a number of jocular or vulgar words. Interestingly, these are all lexically governed, i.e. the fact that -ation attaches to these stems has to be listed as a special property of the respective stems², whereas Latinate stems may take -ation productively, provided that semantic or phonological constraints do not interfere. In fact, this kind of special listing seems to be the only way to circumvent the effect of the Latinate Constraint. The jocular or vulgar words among the counterexamples even support the etymological constraint, because their pragmatic effect originates in the systematic violation of the constraint. Thus, the use of a learned suffix with non-learned stem usually has a humorous effect.

Given the Latinate etymology of the nominalizing suffixes -ance, -age, -al, -y, one may be tempted to ascribe the absence of a preceding suffix -en to the Latinate Constraint. This is, however, not without problems, since with -ance, -age and -al the number of derivatives involving native bases is so large (see Marchand 1969: 248, 234-237, respectively) that these suffixes may be characterized as [± Latinate], allowing in principle their combination with -en, contrary to the facts. The only suffix that indeed seems unquestionably [+ Latinate] is -y (see, for example, Marchand 1969:285).

In view of the inability of the Latinate Constraint to explain the selectional restrictions of -en with regard to -ance, -age, -al, -y, let us examine an alternative solution. The domains of all four nominalizing suffixes -ance, -age, -al, -y have in common that they are highly restricted. No matter, whether one would opt for an analysis that views the domains as lexically-governed, or as defined by complex rules, it seems that both -ance and -al are restricted to verbal bases that have final stress.³ This generalization already means that all bases exhibiting non-final (primary) stress are automatically excluded from this domain. It is exactly this fact that does not allow -ance and -al to attach to verbal -en derivatives, and not a putative selectional restriction that forbids their application to suffixed stems. Again the prosodic restriction on possible bases or derivatives has to be stated anyway to account for the fact that other (derived or non-derived) verbs with non-final stress do not take one of these suffixes.

The distribution of the two remaining suffixes -age and -y seems to be entirely lexically governed, i.e. there is no way to predict which verbal stems they attach to. Steerage, for example, is not the result of a synchronically productive or semi-productive morphological rule, but has to be stored individually.⁴ The same is true for words like assembly, treaty, inquiry.⁵ In any case, a selectional restriction of the form proposed by Fabb is redundant. This leaves us with only one suffix combination that is still problematic, verbal -en followed by nominalizing -ment, to which we now turn.

A closer investigation of sources like the OED and Lehnert (1971) shows that there are some counterexamples to Fabb's claim. The following forms are attested:

The status of these counterexamples is of course arguable since, firstly, the majority may be considered esoteric in the sense described above. Secondly, one of the bases ((be-)dizen) involves a bound root, which casts some doubts on the morphological status of -en in this form. Another problem with this derivative is that derivatives of the form [bound root – suffix - ___ ] are not legitimate counterexamples to Fabb's claims, since he explicitly says that his group 1 suffixes "are never found in the environment [word - suffix - ____ ]" (1988:532). We will nevertheless include bases of the form [bound root - suffix] in our data in order to show that the more specific restrictions we are going to propose can account for both types of complex bases, whereas it is unclear how a model like Fabb's would deal with group 1 suffixes that attach to [bound root - suffix] combinations. The rare cases where the second suffix may be considered to be sensitive to the morphological make-up of the base word will be discussed in more detail below (see for example the combination -ment-al). Note that with respect to his groups 2 and 4 Fabb does not refer to the word/bound root distinction, which suggests that the suffixes of this group are insensitive to the morphological status of the root even in his approach.⁶

Coming back to the examples in (4), I will show in the following that these derivatives are not isolated, idiosyncratic formations, but that they are systematically derived words. We will see that the small number of attested derivatives in -enment results from a number of restrictions on - ment that have to be stated anyway in a reasonably adequate description of -ment.

In general, the specification of the domain of -ment is extremely difficult, and most of the generalizations that have been proposed in the literature are very crude and often accompanied by numerous counterexamples. The following characteristics of -ment derivatives have been observed. Jespersen (1942:376) notes that there is a tendency of -ment to attach to disyllabic bases with stress on the second syllable. Although this is only a tendency and not an exceptionless generalization, it means that verbs ending in -en are not likely candidates for -ment suffixation, which correlates with the near-absence of -enment derivatives. Another observation with regard to derivatives in -ment is that there is a strikingly high number of bases that contain a prefix (like a-, re-, dis-). This may well be related to the tendency mentioned by Jespersen, since prefixed monosyllabic stems will be necessarily disyllabic with ultimate stress. The preference for prefixed stems is especially obvious with stems containing the prefixes eN- and be-, which seem to take -ment obligatorily (cf. also Jespersen 1942:376, Chapin 1967, quoting a paper by Emonds 1966, and the remarks in the OED). For our problem this generalization is especially interesting, because it allows the prediction that verbs containing one of these prefixes AND our suffix -en will, if anything, take -ment.

The number of possible bases of the form eN/be-X-en is relatively small. The OED lists the following forms:

Ignoring the fact that the vast majority of these verbs are obsolete or rare, we can say that of the available 31 bases, three (bedizen, enlighten, enliven) are attested in combination with -ment, and with four derivatives (enchasten, enfasten, enlengthen, bedizen) there are attested combinations with -ment which, however, lack the prefix en-/be-. In all of the latter cases it is striking that the prefix is redundant in the respective base forms, as evidenced by the non-prefixed synonymous (and probably more common) forms chasten, fasten, lengthen, and dizen. The fact that of 31 possible candidates three to seven forms (depending on the analysist's decision to include or exclude chasten, fasten, lengthen and dizen) are attested to have actually undergone the proposed word formation process speaks for the regularity of -enment derivations.⁷ Furthermore, the remaining 'idiosyncratic' examples in (4), namely (re-)awakenment, disheartenment, and worsenment can be explained on different grounds. Thus, although the bases of (re-)awakenment and disheartenment do not conform to the disyllabic pattern observed by Jespersen, they both contain prefixes, and, as was mentioned above, this property seems to play a role in the domain of -ment suffixation. Furthermore, we should not exclude the possibility of analogical formations (cf. (to) better : betterment = worsen : worsenment).

From the foregoing discussion we can draw the conclusion that there seems to be a considerable degree of regularity in the apparent irregularity of -enment derivatives. They are morphologically possible formations, but are subject to independent systematic restrictions imposed on -ment which reduce the number of possible instantiations of -enment forms considerably. The proposed restrictions can thus account for both the existence and the scarcity of -enment derivatives, whereas under Fabb's approach these forms are entirely idiosyncratic.

To summarize our discussion of deverbal nominalizations, we can state that three of the four suffixed verb types in English, namely -ize, -ify, and -ate, are in domains of nominalization processes different from the do mains of the five nominalizing suffixes under discussion. The failure of the nominalizing suffixes in question to attach to the derived verbs is therefore not due to a selectional restriction of these suffixes but a natural consequence of the domain-specificity of the morphological processes under discussion. This is also true for the fourth type, -en derivatives, which is systematically excluded from the domains of -anee, -age, -al, and -y. With regard to the only remaining possible combination, -enment, we saw that although there is only a small number of attested counterexamples to Fabb's claim, these counterexamples are systematic in nature, and that the specific restrictions on -ment can account for the combinatorial properties of -ment in a more adequate fashion than Fabb's generalization. For all of the deverbal nominalizations discussed above it is therefore unnecessary at best to posit the kinds of selectional restrictions Fabb evokes to account for the data.

Let us turn to the denominal abstract-noun-forming suffixes -age )orphanage), -hood (nationhood), -ism (despotism), -y (robbery), which are again claimed by Fabb not to attach to already suffixed bases. Empirically, this position is highly questionable. Absenteeism, expansionism, libertarianism are only a few of the many documented and not at all strange examples, and if there are abortionists (listed in the OED) in this world, there seems to be no reason (at least no morphological one) why their frame of mind could not be labeled abortionism (not listed in the OED). The same holds for -hood 'person, personality, sex, condition, quality, rank' (OED) as in creaturehood, farmerhood, beggarhood, loverhood, to mention only few examples. Documented derivatives of the type N-age involving suffixed bases are, among others, cooperage, lighterage, porterage.

Looking at the number of forms featuring the suffixes under discussion, one cannot, however, overlook that suffixed bases are clearly in the minority, which calls for some explanation. Apart from extra-linguistic mechanisms at work, which may always influence the productivity of certain word formation patterns, a look at the meaning of the suffixes reveals that the low rate of suffixed bases may be a consequence of the semantics of -age, -hood, and -ism. Of the nouns that end in a suffix, a large portion are abstract nouns, in which case the stacking of another abstract-noun suffix leads to uninterpretable results. Thus, words like *concentrationhood or * concentrationage are hard to interpret, to say the least. Hence, the semantics of the bases and suffixes involved rules out a great number of possible derivations.⁸ A look at the above mentioned counterexamples to Fabb's claim adds fuel to this argument, since the majority of these nouns involve person- nouns as bases, where the said semantic problems do not arise.

Contrary to what Fabb puts forward, denominal noun-forming -y is regularly found with bases ending in the agentive suffix -er )archery, patchery, pottery),⁹ of which, strangely enough, Fabb gives an example himself: robbery.

In sum, with respect to the suffixes under discussion, Fabb's claim is empirically wrong, and apparent limitations of productivity can be explained by the semantics of the processes involved.

¹ The only form in the OED2 which manifests at least one of the combinations of suffixes is festenance, festynens (fasten + ance, 'confinement, durance') characterized as obsolete and scientific, with the last citation dating from 1533.

² This goes hand in hand with idiosyncrasies of various kinds. Thus, for example, thunderation or botheration are primarily used as exclamations.

³ There is only one counterexample attested for -al, namely burial, which only as the result of a folk etymology of the original form buriels became associated with deverbal -al. There are few counterexamples involving non-final stress with -ance, cf. utterance, but their number is very small. Note that final stress seems not to be a sufficient condition for the application of these suffixes (cf. * advance-ance, *advance-at). This warrants the conclusion that, unless other general constraints are detected, we are not dealing with a productive rule.

⁴ An OED search for all twentieth century neologisms with deverbal -age only yields six forms: coverage, creepage, frettage, narratage, spillage, stewage. The small number of neologisms is indicative of the non-productivity of the process, and the attested neologisms should be analyzed as local analogies.

⁵ The way the storage of this kind of information is conceptualized depends on one's theory of the mental lexicon. Under a base-driven approach, it is part of the lexical entry of the verb which kind of nominalizing suffix it may take (besides the general suffix -ing, which seems to be applicable to practically all verbal bases), under an affix-driven approach this information is stored as part of the lexical entry of the suffix. In a word-based morphology these nouns are simply stored as wholes, possibly with a redundancy rule expressing their relatedness with respect to the nominalizing suffix. For the purposes of this investigation we need not decide on this issue, because the points being made here may be accomodated for under each of the approaches just mentioned.

⁶ For our discussion of groups 2 and 4 this means that also derivatives involving a bound root can count as genuine counterexamples.

⁷ We do not discuss here the fact that so many of the base words are obsolete or rare, and the possible consequences this may have had for the diachrony of -enment derivatives. In very broad terms, it seems that taking into account the obsoleteness or rarity of the bases would only further reduce the number of available bases, which would add further strength to our argument that the process is regular.

⁸ The exact determination of the semantic restrictions involved is a matter of future research.

⁹ This pattern is obviously related to the complex suffix -ery.

مواضيع ذات صلة

Person-noun-forming suffixes

The largest group of suffixes (28 out of 43): Suffixes that do not attach to already suffixed words

An alternative account

The data

Fabb (1988)

The combinability of derivational suffixes

Productivity: A summary

Stratal constraints: Level ordering, Lexical Phonology, and the Latinate Constraint

Blocking

The unitary output hypothesis

The unitary base hypothesis

Structural restrictions: General mechanisms