PHRASAL CATEGORIES IN WORD FORMATION RULES

It is'with the first property (l)a. above that the present paper is concerned. It is obvious why it is important to lexicalists that WFRs should not involve phrasal categories: the number of phrases generated by the syntax is infinite. Consequently these phrases cannot be available as input to WFRs in the form of a finite list in the lexicon. On the other hand, if such phrases had to be generated in the lexicon by means of rules, the lexicon would duplicate the function of the rules responsible for the generation of these phrases in the syntax. Thus Aronoff (1976:47) points out that


Introduction
This paper deals with one of the properties of the word formation rules of lexicalist morphology.These rules apply within the lexicon to form new words on the basis of already existing words.Roeper and Siegel (1978:202) recently summarized the properties of word formation rules (henceforth WFRs) as follows.
WFRs involve no phrasal categories b.
WFRs shift syntactic category c.WFRs involve no medial variables d.
WFRs have no extrinsic ordering e.
WFRs involve semantic compositionality f.WFRs permit statement of idiosyncratic information.
It is'with the first property (l)a.above that the present paper is concerned.It is obvious why it is important to lexicalists that WFRs should not involve phrasal categories: the number of phrases generated by the syntax is infinite.Consequently these phrases cannot be available as input to WFRs in the form of a finite list in the lexicon.On the other hand, if such phrases had to be generated in the lexicon by means of rules, the lexicon would duplicate the function of the rules responsible for the generation of these phrases in the syntax.Thus Aronoff (1976:47) points out that " of We be access to anything other than the base calls for rules a much more powerful sort than we would prefer to have.will therefore operate on the assumption that a WFR can cognizant only of information contained in its own base." If it could be shown that WFRs do in fact involve phrasal categories, then the very existence of these rules as a distinct type of rule could be questioned.
The central claim of this paper will be that Afrikaans l ) has verbal compounds which incorporate syntactic phrases as constituents.
It will be argued that if these verbal compounds have to be derived by means of WFRs, then WFRscannot be claimed to have the property (l)a.
To develop this argument, we have to consider a currently accepted lexicalist theory of word formation (~2), Roeper and Siegel's (1978) theory of verbal compounding in English (83) and the relevant Afrikaans verbal compounds (G4).

2.
A theory of word formation Roeper and Siegel (1978:200-204) provide a useful outline of a theory of the lexicon and its word formation component that draws on work by Halle (1973), Jackendoff (1975) and in particular Aronoff (1976).Within the framework of this theory the lexicon contains, among other things, a LEXICALCOREand a set of WORD FORMATION RULES.
The LEXICAL COREis a list of words containing t,7O sub-components: the atomic core and the complex core.The atomic core is a list of all those words that have no internal morphological structure.Thus words like church, house and ~will be part of the atomic core.The complex ~, on the other hand, is a list of those wordl3 that have been created by word formation rules and hence are morphologically structured.Words like housing and bedding are typical examples of words contained in the complex core.Words like possible and happY which appear to have morphological structure but which are not compositional in meaning, are listed in the atomic core 2 ).
The WORD FORMATION RULES(WFRs) are rules which operate completely within the lexicon.
Aronoff (1976:46) states that although WFRsmake reference to syntactic, semantic and phonological properties of words, they are totally separate from the other rules of a grammar.Words from the lexical core serve as input to WFRswhich create new words on the basis of these core words.Thus there is a WFRwhich creates words like housing and bedding by adding ~to the already existing words house and bed.
The output of WFRsmay either be inserted into s;yntactic structures via lexical insertion rules or they may be inserted into the complex core The properties which WFRs are claimed to have, have already been listed as (1) above.
In addition to WFRs the lexicon contains a number of other devices, including allomorphy adjustment rules and stress assignment rules, the nature and function of which are irrelevant to the aim of this paper.
To summarize their observations about the structure of the lexicon, Roeper and Siegel (1978:204) present a flow chart developed by Keyser and Carlson.

Lexical core
Word Formation Rules

Complex
From the flow chart (3) it is clear that the output of these rules which apply in the lexicon may either directly enter the syntax or be inserted into the lexical core where it will be stored in a long-term memory.

3.
A theory of verbal compounding are analysed during sentence processing.
The following remarks by Roeper and Siegel (1978:204) provide further clarification of this point. (4) "Words with particularly frequent affixes could not all be listed in the core.
For instance, the -~adverbs are so numerous that it would be inefficient to remember each one.It

De Villiers 42
Note that there are two points of entry one from the core and the other directly from the Thus certain newly formed words are not listed in the core, but Within the framework of the lexicalist theory outlined in 82, Roeper and Siegel (1978) propose a theory of verbal compounding for English.Amongst other rules they propose a transformation the Compound Rule which applies within the lexicon to form verbal compounds in English.
It is not at all clear which formal principles may serve as a basis for the distinction between those WFRs of which the output should be inserted into the lexical core and those WFRs of which the output should be inserted into syntactic base structures.
into the syntax WFRs.
serve as input to further WFRs.

3.1
Verbal compounds and root compounds Roeper and Siegel (1978:206)  (iii) may be unpredictable in meaning.

The first Sister Principle
The central principle in Roeper and Siegel's account of verbal compounds in English (1978:208) is the First Sister (FS) Principle. (5) First Sister (FS) Principle All verbal compounds are formed by incorporation of a word in first sister position of the verb.
Thus, a word may be incorporated in a verbal compound only if it can appear in the FS position of the verb that is in the position immediately to the right of the verb.
The fact that *huge going does not exist, therefore, follows from the fact that huge m~not appear in the FS position of the verb~.

FS
The existence of good-looker, by contrast, is explained by the fact that good can appear in the FS position of the verb look, as in (7).
v FS look .good The FS principle is expressed in the Compound R~e which Roeper and Siegel (1978:209) propose for the formation of verbal compounds.This rule, presented as (8) below, moves a word from the FS position to the left of a verb.
(13) [X empty] ~L N coffee J Any number of empty optional subcategorization frames may occur between the filled frame and the verb.
In such cases the words inserted by the rule (12) will not be in the FS position of the verb.
To overcome this problem, Roeper and Siegel (1978:212) propose an adjustment rule the VARIABLE DELETION RULE to delete all the empty phrases between the verb and the word inserted by rule ( 12).Empty subcategorization frames deleted by this rule will not be available for lexical insertion in syntax.
6 ) The general form of the Variable Deletion Rule is presented in ( 14).

3.4
The Compound Rule and English verbal compounds Roeper and Siegel (1978:213-217) argue that the properties of the Compound Rule (8) are consistent with the formal properties of WFRs (cf.(1) above) and that this rule must therefore be regarded as a WFR.
In their discussion (1978:213-214) of the property (l)a., i.e. the restriction that WFRs may not involve phrases, they state that the Subcategoriza- In 84 Roeper and Siegel (1978:217-225) discuss several apparent verbal compounds which seem to be counter-examples for their FS principle (cf. (5) in ~3.2 above).They argue that these examples are not in fact verbal compounds but that they should be analysed as either (i) phrase structure generated sequenc es or (ii) derivations with root compounds as bases.In this discussion they develop a number of diagnostics for verbal compounds.These diagnostics are presented in the form of the questions in ( 16).
If the answer to the question (a) and anyone of the remaining questions, is affirmative, then the form in question is a verbal compound that will obey the FS principle.

The formation of verbal compounds in Afrikaans
Afrikaans verbal compounds appear to be formed in accordance with the theory of verbal compounding proposed by Roeper and Siegel (1978) for English.Thus, the example koffiedririker ("coffee drinker" ) can be analyzed as follows.
The verb drink ("drink") and its subcategorization frames form the input to an affixation rule (cf.(lO) above) which attaches an empty frame to the left of the verb drink and the affix ~to the right of it.
The Subcategorization Insertion Rule (cf.( 12) above) is responsible for the insertion of a word, the noun koffie ("coffee") into the relevant subcategorization frame of the verb, drink.Any optional subcategorization frames which might appear between the verb drink and the noun, koffie will be deleted by the Variable Deletion Rule (cf.( 14) above) .The Compound Rule (cf.(8) above) moves the noun koffie ("coffee") and inserts it into the empty frame provided by the affixation rule to the left of the verb drink ("drink").
Although Afrikaans verbal compounds are formed in accordance with the theory of verbal compounding outlined in ~3, Afrikaans has a class of complex forms which seem to be problematic within the lexicalist theory outlined in 82.These complex forms have the properties of verbal compounds and obey the FS principle but they include phrases.Because WFRs may not involve phrases, one would attempt to avoid analysing these forms as verbal compounds.However, it will be argued that these complex forms cannot be analysed as anything but verbal compounds and that they appear, therefore, to be counter-examples to the claim that WFRs do not involve phrasal categories.
The complex forms under consideration belong to different classes in terms of the phrases which they incorporate and will be discussed separately.

Adverbial phrases incorporated in verbal compounds
A first set of examples which will be considered comprises a class of  Roeper and Siegel (1978:218) formulate the following criterion for deciding whether or not a particular form is a sequence generated by lexical insertion into phrase structure.
(20) "In a PS-generated sequence each lexical item must have a separate representation in the lexical core." If a lexical item in a putative phrase structure generated sequence does not exist independently, it follows that the considered example cannot be analysed as a phrase structure generated sequence.However, this does not mean that verbal compounds can never include independently existing words (cf.Roeper and Siegel (1978:226)).
The independent existence of each lexical item is not the only criterion for deciding whether a particular example is to be analysed as a phrase structure generated sequence.Roeper and Siegel (1978:223)  Further evidence that the putative adjective har~("loud lt ) in ( 22)c.
cannot be analysed as an adjective is the fact that it lacks a typical morphological property of adjectives.
In Afrikaans certain adjectives have the suffix -e when used attributively.
Hard in fact has this property when it is used conventionally as an attributive adjective.
Thus compare the acceptable a.-form (harde) with the unacceptable b.form (hard).
all have adverbial readings.Laat ("late") in ( 22)a., for instance, does not relate to the putative noun ?slaEer ('tsleeperlt) but to the verb slaaE (ltsleep").Thus, baie-laat-slaper, must be paraphrased as "a person who sleeps very late ll and not as Ita sleeper who is very late lt .
In the same way the putative adjective in every other complex form in ( 22) relates to the ~incorporated in the putative noun and not to the noun itself.

Die harde geluid
The loud noise a.The analysis (25) of the examples in ( 17) however, cannot be accepted.
There is no way in whi ch it can be argued that the adverbs analysed as putative adjectives in ( 25) are indeed adjectives.
In each case the putative adjective modifies the adverb incorporated in the verbal compound and not the noun formed by verbal compounding.Thus, baie lekkerlagger should be paraphrased as "a person who laughs very heartily" and not as "a heartily laugher who is very".
With regard to the complex forms in (25)c.and d. further evidence that the putative adjectives cannot be analysed as adjectives exists.
As in the case of the putative adjective hard in (2 )c., besonder has the suffix -e when used as an attributive adjective.The analysis ( 27), however, must be rej ected.
The putative root-compound verbs in ( 27) cannot be analysed as verbs since the,r lack typical morphological and syntactic properties of Afrikaans verbs.Compare, for example, the acceptable (i)-forms to the unacceptable (ii)-forms in ( 28) and ( 29).One of the Afrikaans expressions corresponding to go to church contains the postposition toe ("to").The question of course, is whether or not the complex form kerk-toe-ganer has to be analysed as a verbal compound.It will be argued that this form can be analysed neither as a phrase structure generated sequence nor as a complex form formed by the adding of a verbal affix to a root-compound verb.
The only plausible analysis of the complex form in (33)b., it will be argued, assigns it the status of a verbal compound.The putative noun 8. ganer (I'goer ll ), however, does not exist as an independent noun of Afrikaans.
In accordance with the criterion (20) above, the analysis (34) in which the form kerk-toe-ganer ("church-goer") is analysed as a phrase structure generated sequence, cannot be accepted.
Further evidence that the analysis (34) does not hold follows from the fact that the phrase kerk toe ("to church") must be analysed as an adverbial phrase.
The phrase under consideration does not relate to the putative noun & ganer (llgoer") but rather to the verb ge.~("go").
In (35) it may be seen that this phrase does not function adjectivally in sequences similar to the putative phrase structure generated sequence in (34).The alternative analyses for the complex form kerk toe garter (tlchurch-goer") cannot be accepted.It follows that this form is a verbal compound which is a counter-example to the claim (l)a.above because it incorporates a phrase as first constituent.
Other complex forms which have to be analysed as verbal compounds but which contain the postposition toe (lito")are presented in (38).,The'question now arises whether or not there is a plausible analysis of these forms which does not assign them the status of verbal compounds.
It will be argued that the forms in (40) cannot be analysed as phrase structure generated sequences in which independently existing words have been inserted separately.An analysis along these lines is presented in (42).The analysis (42), however, cannot be accepted.
Under such an analysis one would expect the putative locative phrases to modify the putative nouns.
Thus a. would be a & lier who is in bed, b. a smoker who is behind the wall, and c. a ?sitter who is on the stoep.In fact, however, the smoker in b. for instance, could be in the dining room or anYWhere else at the moment of speaking, but his smoking is habitually done behind the wall.
Thus it seems that the locative phrases in (42) do not modify the putative nouns but rather the verbs l~("lieU), ~(Usmoke") and~("sit  It WFRs do involve phrases, these phrases will either have to be listed in the lexical core or they will have to be generated in the lexicon by means of rules.The first alternative is not acceptable because the number of phrases is infinite and therefore cannot be stored in a finite list.
On the other hand, the second alternative is also not acceptable: if these phrases were generated by means of rules in the lexicon, these rules would duplicate the function of the rules responsible for generating the corresponding phrases in the syntax.
Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123 Y range over empty subcategorization frames After the application of the Rules of Affixation, the Sub categorization Insertion Rule and the Variable Deletion Rule, the Compound Rule (8) applies.This rule removes a word from the FS position appears after application of the Subcategorization Rule where it and inserts it into an empty frame to the left of a verb, the empty frame being supplied by the Rules of Affixation.
(22)  in which the examples in (17) are analysed as PSgenerated sequences, each consisting of an adverb, an adjective and a noun, cannot be upheld.A second alternative would be to analyse the Afrikaans examples in (17)as sequences into which an adjective and a noun have been inserted separately.On this analysis the noun would, in each case, be a verbal com- *Hy besonder klein 'skryf Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2*Hy het gebaie laat kom It is clear that the putative root-compound verbs in (27) cannot be analysed as verbs.Thus the analysis (27) cannot be accepted.It is therefore difficult to avoid the conclusion that the only remaining possibility is to analyse the examples presented in (17) as verbal compounds T) If the examples in (17) are verbal compounds, it follows that at least those WFRs responsible for the forming of the expressions in (17) involve phrasal categories.Further examples of complex forms in Afrikaans containing phrases of the type adverb + adverb, which I believe cannot be analysed as anything but Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123 De Villiers 56 verbal compounds, are presented below.(1978:242) show that some English verbal compounds appear to violate their FS principle as they incorporate words which may only appear in the FS position of the verb when preceded by a preposition.Thus church in the verbal compound church~goer can only appear in the FS position of the verb ~when it is preceded by the preposition to.account for this fact, they (1978:242) reformulate their Subcategorization Insertion Rule.This rule can now insert a preposition immediately Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123 to the left of the word which it inserts in the FS position of a verb.The question for Roeper and Siegel, now, is how to explain the fact that such prepositions never occur in English verbal compounds.They argue that this question does not pose a real problem.Their Variable Deletion Rule which was formulated to delete all empty frames which occur between the verb and the frame filled by Sub categorization Insertion, will now delete "everything that falls between verb and [+ word]" (1978: 242).Thus the preposition inserted in front of a word in the FS position of a verb is deleted before the Compound Rule applies.In this way the Compound Rule does not involve a phrase but a single word (cf.(l)a).
in (33)b.apparently incorporates a phrase cons.isting of a noun and a postposition.

Stellenbosch
Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123       De Villiers 58 A first analysis in terms of which the complex form in (33)b. is not treated as a verbal compound, assigns it the status of a phrase structure generated sequence into which independently existing lexical items have would be to assign the form (33)b. the status of a complex form in which a verbal affix is attached to a root-compound verb.(36) cannot be accepted.The put~tive root-compound verb kerk-toe-gaan cannot be analysed as a verb.It exhibits neither the typical syntactic not the typical morphological properties of Afrikaans verbs.Stellenbosch Papers in Linguistics,Vol.2, 1979, 39-69 doi: 10.5774/2it is clear that kerk toe gaan (lIchurchto go") does not function as a single verb unit, neither when used in the present tense (cf.(37)a.)norwhen used in the past tense (cf.(37)b.).
going the act of going to the game reserve Stellenbosch Papers in Linguistics,Vol.2, 1979, 39-69 doi: 10.5774/2walking along the road All the complex forms in (39) have the properties of verbal compounds and obey the FS principle.Altho;Ugh they have phra.sesas first constituents, they cannot be analysed as anything but verbal compounds.Thus the complex forms in (39) together with those in (38) and (33)b.constitute counterexamples to the claim that WFRs do not involve phrases.4.3 Prepositions incorporated in verbal compoWe turn next to complex forms in Afrikaans which appear to be verbal compounds but which incorporate a prepositional phrase as first constituent.Consider the forms in (40) which incorporate locative prepositional phrases.(40 ) a. in-die-bed-Ieer in the bed lier a person who habitually lies in bed Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2usually sits on the stoep Each of the complex forms in (40) has the properties of verbal compounds and obey the FS principle as is illustrated in (41).forms presented in (40) have to be analysed as verbal compounds, they will be counter-examples to the claim (l)a.that WFRs do not involve phrases.
choosing between leadersStellenbosch Papers in Linguistics,Vol.2, 1979, 39-69 doi: 10.5774/2-0-1234.4.Other phrases incorporated in verbal compoundsIn the preceding sections several counter-examples to the claim that WFRs do not involve phrasal categories were dealt with.Afrikaans has additional verbal compounds which are counter-examples to this claim (l)a.Instead of analysing these forms in detail, I simply list them below.Interestingly some of these forms include more than one phrase.-hand-in-die-sak-stanery with the hand in the pocket standing the standing with one's hand in one'Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123 b.met-die-handsak-onder-die-arm-lopery.with the handbag under the arm walking the walking with a handbag under one'theory of morphology outlined in 82, WFRs constitute a distinct type of rule which functions within the lexicon and which is totally separate from the other rules of a grammar.One of the properties that WFRs are claimed to have is that they do not involve phrases.Afrikaans has numerous examples of complex fo~g which have to be analysed as verbal compounds and whi ch are formed by means of a WFR such as the Compound Rule (8).These verbal compounds involve a phrase as first constituent and thus constitute counter-examples to the claim that WFRs do not involve phr ases .
De Villiers 50 not only on account of the fact that the putative nouns do not exist independently, but also because the putative adjectives have adverbial readings and must be analysed as adverbs.
argue that the examples in (21) below cannot be analysed as PS-generated sequences, Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123With regard to the independent existence of the putative nouns in (22), let us consider the sentences (23) below.FollowingRoeper and Siegel   Stellenbosch Papers in Linguistics, Vol. 2, 1979, 39-69 doi: 10.5774/2-0-123 In terms of a second alternative analysis the complex forms in (40) could be claimed to be formed by adding a verbal affix to a root-compound verb.It seems that the complex forms in (40) must be analysed as verbal compounds and are thus counter-examples to the claim that i~Rs do not involve phrases.Apart from locative prepositional phrases, other prepositions can also be included in Afrikaans verbal compounds.