Abdelkader Fassi Fehri University Mohammed V Souissi, Rabat [email protected]

VERBAL AND NOMINAL PARALLELISMS AND EXTENSIONS*

July 20, 2005

Reports & Documents no 8, IERA, Rabat

________ * This work has been presented at various occasions, notably at the Colloquium on “Language and Logic”, High School of Teachers, Meknes, April 2005, the Univ. of Paris VII, Invited lecture, May 2005, and the Linguistic Society of Morocco Lecture Series, June 2005. I would like to thank the audiences there, and acknowledge fruitful remarks and comments by Jacqueline Guéron, Mohammed Amine, Marie-Thérèse Vinet, Pierre Pica, Driss Seghrouchni, Nadia Aamiri, Danièle Godard, Ahmed Berrissoul, Majdouline Nhibi, Khalid Lachheb, and Abderrzak Tourabi. Errors are mine.

2

Despite the abundant literature on noun and verb parallelisms in terms of structures and functions, it is still dominated by an absence of clarity of how these parallelisms are characterized and identified in the various proposals. For example, there is no agreement among linguists about the conceptual content of the cross-categorial feature system needed, nor is there agreement about the number of classes to be used for both nouns and verbs, or the nature of functional heads and projections. The most common classification in recent years is based on two classes for nouns (count/mass), and four classes for verbs (e.g. the four Vendlerian classes), with a discriminative cardinality between the two categories. Although the most traditional classification of verbs has been binary (action/state), or even ternary (with ‘process’ as a third term), in addition to the present most spread quaternary classification, I know of no clearly articulated and conceptually simple proposal which makes the appropriate parallelisms both transparent and motivated.1 Based mainly on Arabic data, the analysis put forth in this article is an attempt to make verbal and nominal classifications fall out from one and the same unified system, based on two valued features: [± atomic] and [± singulative]. The latter induces four classes for both nouns and verbs. The system adopted turns out to be not only conceptually and empirically motivated, but it has far-reaching consequences for appropriately describing grammatical phenomena that escaped so far any adequate treatment for the formation and interpretation of event nouns and unit event nouns (both of which can function as cognate objects), and intensive predicates. Furthermore, more adequate (partial) architectures of the vp/CP and the np/DP domains are motivated and unified. In section1, I restate some of the foundations and arguments in favor of the new proposed system to treat nominal classes, already provided in previous work of mine. In section 2, I extend this system to verbal classes, and explore further extensions to event nouns and unit nouns, as well as intensive predicates. In section 3, I compare my system to some other systems which come partially close to mine, namely Jackendoff’s (1991), Verkuyl’s (1993), and Rothstein’s (2004).

1. The nominal domain In Fassi Fehri (2003b, 2004b), nominal expressions are classified into four distinct classes of objects: kinds K, individuals I, groups G, and masses M. The four instances are exemplified in Arabic (1)-(4), and English (5): (1) ´akal-tu tamr-an ate-I date-acc I ate dates (literally ‘date’). (2) ´akal-tu tamr-at-an ate-I date-unit-acc I ate a date. (3) laqii-tu fariiq-an met-I

team-acc

I met a team.

3 (4) štaray-tu zayt-an bought-I oil-acc-n I bought oil. (5) a. I like dates. b. I ate a date. c. I met a committee. d. I bought oil. This quadripartite system differs from the widely spread binary system found in the literature, based on a single valued [± count] feature. Enough criticism of the latter has been put forth in Fassi Fehri (2003b, 2004b), as well as Fassi Fehri & Vinet (2004). I will return to part of it here. In the references mentioned, it is proposed that np is dominated by classifier phrase ClP, and the latter by number phrase NbP, in a hierarchical functional structure such as the following (NumP for numeral phrase, and QP for quantifier phrase): (6) [ QP[ DP[ NumP[ NbP [ ClP[ np]]]]]] A mereological feature system is postulated in order to characterize the whole and parts of nominal expressions, taking into account their ‘integrity’. Moreover, it is argued to be a more appropriate classificatory system. 1.1. A nominal mereology The system is built on two features, designated as atomic and singulative. The first feature characterizes the integrity of the whole, and the second that of parts.2 The four nominal classes then cross-classify as follows (- is the negative or unspecified value): (7) a. I: [+ atom, + sing] (tamr-at-an ‘a date’) b. K: [- atom, + sing] (tamr-an ‘dates’) c. G: [+ atom, - sing] (fariiq-a-n ‘a team’) d. M: [- atom, - sing] (zayt-a-n ‘oil’) A binary system is insufficient and inadequate to account for the various properties and distributions of these classes. Thus the kind ‘date’ in (1) behaves as singular, although its English counterpart in (5a) is plural. It refers to an unspecified number of dates, which can be one, two, or more. But ‘date’ in (2), which is derived from the kind form in (1) via a suffix classifier –at, can only refer to a date unit, which is unique in this construction. In other words, the np in (2) is atomic and associated with cardinality 1, contrary to that in (1). However, the interpretation in both cases requires that the objects referred to have integrity, or are not divisive. The atomic/cardinality property makes (1) countable (or ‘numerable’), but (2) is not, hence the following individual/kind distinction : (8) ‘akal-tu talaat-a tamaraat-in ate-I three-acc date-unit.pl-gen "I ate three (individual) dates". (9)* ´akal-tu talaatat-a tamr-in ate-I three-acc date-acc "I ate three dates (literally ‘date’)". As for Mass in (4), its parts are not integral, and they can always be potentially divided. Thus every part of ‘oil’ is ‘oil’. Likewise, the cumulation of the parts of Mass does not lead to the formation of an integral whole. Consequently, Mass is neither

4 atomic nor singulative, and it has no cardinality associated with it, nor any integral parts. Finally, Group in (3) has a cardinality 1 associated with it, and hence can be counted, as well as pluralized.3 Its internal semantics implies that it has parts, but these are not N-parts, in the sense of Fassi Fehri (2003b), Nicolas (2001); cf. infra. 1.2. Distinctive properties With Mass, two properties are often associated. Distributivity is a property of Mass so that ‘any part of a mass object which is W is itself W’ (cf. Cheng 1973). An amendation of this definition, based on Nicolas (2001), is adopted here (cf. Fassi fehri 2003b, 2004b for motivation): (10) Distributivity N refers distributively if it applies to any N-part of what it applies to. Quine (1960) proposed that mass nouns like ‘water’ or ‘furniture’ have the characteristic semantic property to refer cumulatively, so that every sum of the parts which are named ‘water’ is itself ‘water’. Cumulativity can be defined as follows: (11) Cumulativity N refers cumulatively if, whenever it applies separately to each of two N-parts, it also applies to the two N-parts considered together. Unit nouns can be pluralized, as in (12), and so are kind nouns, as in (13). But the interpretation of each plural is different: (12) ´akal-tu tamaraat-in ate-I date-unit.pl-acc "I ate (many) dates". (13) ´akal-tu tumuur-an ate-I date.pl-acc "I ate: (a) (many) kinds of dates; (b) a lot of (units of) dates". In the first case, the plural forms a set of atoms, but in the second case it is a plural of sorts, or an amount plural (called also an ‘abundance plural’; cf. Fassi Fehri 2004b). It is also the case that kinds can be counted, and they are not incompatible with numerals. However, what is counted here are ‘sorts of objects’, rather than the objects themselves (compare with interpretation of (8)): (14) ´akal-tu talaat-at-a tumuur-in ate-I three-acc date-pl-gen "I ate three sorts of dates". These distinctions call for separating atomicity from singulativity. The distinction between kind and unit of kind (or individual) is also found in Chinese, a so-called classifier language. Kind nouns imply singulativity, which can be singular or not, being unspecified for atomicity: (15) wǒ kànjiàn gǒu le 1s see dog Asp "I see (a/the) dog(s)". When referring to an individual, a classifier phrase becomes necessary to force atomicity and cardinality:

5 (16) wǒ kànjiàn yī zhī gǒu 1s see a/one Cl dog "I saw a dog". Likewise, a classifier is necessarily introduced with a cardinal numeral, hence the ungrammatical (17b): (17) a. wǒ kànjiàn sān zhī gǒu 1s see three Cl dog "I saw three dogs". b. * wǒ kànjiàn sān gǒu 1s see three dog "I saw three dogs". Similar behavior is observed with quantifiers requiring individuals: (18) a. ji ge pingguo few Cl apples few apples b. mei ge ren every Cl man every man In Colloquial Moroccan Arabic, classifiers are also used to form unit nouns from kinds: (19) təffaah “apples” ( təffaah-at “apple-unit; an apple” But massifiers are also found, which derive Mass from Kind: (20) bgər "cattle, cows" ( bəgr-i "cow-M; beef" Likewise, a classifier is used in Tamazight Berber to derive unit from kind:4 (21) a. tini “date” Æ tini-t “date-unit” b. azru “stone” Æ t-azru-t “stone-unit” c. təffah “apples” Æ ta-tffah-t “apple-unit; an apple” The atomicity of Group can be tested via its cardinality, and its ability to pluralize as a set of groups, not sorts: (22) laqii-tu talaat-at-a furuq-in met-I three-acc team.pl-acc "I met three teams". (23) laqii-tu furuq-an met-I team.pl-acc "I met teams". But groups differ from individuals in that they manifest ‘plural’ behavior, unlike the latter. For example, they are compatible with reciprocal anaphora: (24) l-fariiq-u (* r-rajul-u) ntaqada bacd-u-hu bacd-an the-team-nom (*the man) criticized each-him each “The team criticized each other”. They can be used with plural concord : (25) a. n-naas-u y-aquul-uu-na haadaa the-people 3-say-pl-ind this "People say this". b. l-fariiq-u jtamac-a; tumma qarrar-uu haadaa the-team-nom met-3s. then decided-pl this "The team met; they then decided this". They support plural predication:

6 (26) a. The committee (* the man) is John and Fred. b. The committee gathered here. Their plural interpretation can take various other forms: (27) l-fariiq-u y-ata´allaf-u min ´acdaa´-in the-team-nom 3-compose-indic of members-gen "The team is composed of members". Their compatibility with the quantifier of (plural) individuals bidc “some, few» indicates that their individual parts might be named, although not with the same noun: (28) l-fariiq-u jtamaca bidcat-u ´acdaa´-in min-hu the-team-nom met few-nom members-gen of-it "The team met with a few members". Note, in comparison, that Kind, although not atomic and disallowing cardinals and plural, can occur with bidc, which then makes explicit its (potentially) atomic parts: (29) t-tamr-u ´akal-tu min-hu bidc-a tamaraat-in the-date ate-I of-it few-acc dates-gen "The dates, I ate few (units) of them". This contrast significantly with the behavior of Mass, which when used with bidc, can only be interpreted as quantifying over sorts, not over units: (30) z-zayt-u bidcat-u zuyut-in min-hu faasidat-un the-oil-nom few-nom oils-gen of-it bad-nom "The oil, few oils of it are bad". 1.3. Classifier and Number interaction I have shown that the nominal classification is quadripartite, and it relies on the characterization of the integrity/atomicity of the whole and parts. This ‘internal’ atomicity interacts with another atomicity, namely that of Number, and should not be confused with it, as amply argued for in Fassi Fehri (2003b, 2004b), and Fassi Fehri & Vinet (2004). It is shown in these references that there is no complementary distribution between Cl and Nb realizations, and that the ‘external’ atomicity marked by Nb is classification preserving. This is manifested, for example, by the different interpretation of Plural in (12) and (13) above, where plural of individuals differs from that of kinds. Likewise, contrasts are found between interpretation of numerals, depending on their classes (compare (8) and (14) above). These contrasts point to the fact that Numeral and Number are not incompatible with kinds (nor masses; cf. Fassi Fehri 2004b), contra Borer (2005a), but that they rather involve different interpretations, preserving the internal characteristics of the classes counted or pluralized. Furthermore, Cl and Nb are both projected hierarchically, with the Numeral higher than Number, as in (6) above, repeated here as (31) for convenience: (31) [ QP[ DP[ NumP[ NbP [ ClP[ np]]]]]] These clarifications are necessary to identify internal and external atomicity interactions in nominal structures. Similar distinctions are to be found in the verbal domain, as we will see.

7

2. The verbal domain In the event/verbal domain, Vendler (1967) has proposed four classes: (a) achievements, (b) accomplishments, (c) activities, and (d) states. Many later studies have explored this system, seeking to examine and determine adequate tests for identifying one class or the other. The quaternary system is illustrated by Arabic constructions (32)-(35), or their English translation counterparts: (32) wajada r-rajul-u l-hall-a found the-man-nom the-solution-acc “The man found the solution”. (33) ´akala r-rajul-u tuffaah-at-an ate the-man-nom apple-unit-acc “The man ate an apple”. (34) jaraa l-walad-u run the-child-nom “The child run”. (35) carafa r-rajul-u l-jawaab-a knew the-man-nom the-answer-acc “The man knew the answer”. These four classes can be made strictly parallel (and equivalent) to the four nominal classes mentioned above. The same two features used above can be used here to cross-classify: atomicity describes integrity of predicates as wholes, and singulativity integrity of their parts. The featured cross-classification and the correspondence come out as follows: (36) a. Achievement (= I): [+ atom, + sing] b. Accomplishment (= G): [+ atom, - sing] c. Activity (= K): [- atom, + sing] d. State (= M): [- atom, ∅ sing] Achievement has a single discrete event (which is ‘punctual’, not an interval). Accomplishment has many events grouped into a single one. To 'eat a piece of apple' has to be 'grouped' to form 'to eat an apple'. Accomplishment is atomic, though not singulative. Activity has different discrete events, but of unspecified quantity. States are (homogeneous) eventualities that are neither atoms nor of a specified quantity. Essential properties of VPs are then taken into account. Discrete events are singulative, and bounded events are atomic. Punctuality of achievements and durativity of accomplishments are derived through interaction of singulativity and atomicity in the first case, and absence of singulativity in the second case. Durativity is taken to be the unmarked case. Punctuality may be a property of only 'pure' achievements (as in e.g. Caudal 1999, Rothstein 2004). This four-way classification competes with a binary classification, which is widespread in the literature, and which uses a [± telic] feature. Bach (1986), among others, has equated the latter with the [± count] feature, so that the latter would apply to the event domain as well as the nominal domain. Telicity establishes two classes: (37) a. [+ telic]: achievements, accomplishments b. [- telic]: activities, states. The popular test for telicity is the adverbial in-X: (38) a. ´akala samakat-an fi i saacat-in ate fish-I-acc in hour-gen

8 "He ate a fish in an hour". b. * ´akala samak-an fii saacat-in ate fish-acc in hour-gen "* He ate fish in an hour". Obviously, the two members of the telic class do not exhibit similar behavior in various contexts, and they do not have the same internal structure. Achievements and accomplishments can be unified through the atomic value, which is manifested through counting the bounded event in the following two constructions: (39) ´akala r-rajul-u tuffaahat-an marr-at-ayni ate the-man-nom apple-unit-acc time-dual-acc “The man has eaten an apple two times”. (40) wajada r-rajul-u dirham-an marr-at-ayni found the-man-nom dirham-acc time-dual-acc “The man has found a dirham (a piece of money) two times”. The count reading indicates that the event can be counted as a unit. In fact, (39) and (40) are potentially ambiguous. They can count the number of eventualities (accomplishments or achievements) which have occurred twice. Under the count reading of eventualities, the man has eaten two apples. In the second reading, which is counting times, the man would potentially has eaten only one apple. This situation contrasts with the reading found with activities and states when constructed with numbers of times: (41) raqasa marr-at-ayni danced time-dual-acc “He danced two times”. (42) marida marr-at-ayni sickened time-dual-acc “He was sick two times”. There interpretation is limited to counting times. Based on these facts, one can establish equivalence between telic, count, and atomic: (43) [± telic] = [± count] = [± atom] In the positive case, one counts the eventuality, in the negative case its time. Telic means that the event is countable, not its time (which is irrelevant for the classification here). If telicity is atomicity, then we expect to find in the domain of events what we find in the domain of nominals, namely that the distinction between the two members of the class is necessary. When atomic, an event can be either [+ sing], and hence an individual I, or [- sing], and hence a group G. Clearly, there are distinctions between Achievement and Accomplishment, even though both are atomic. These distinctions amount to those between I and G. The same reasoning applies to the [- telic]/[- atom] value. Distinctions between activities and states would presumably be accounted for appropriately if they are made parallel to kinds K and masses M. The system hopefully allows further applications and predictions. Among those is an adequate characterization of derivations and interpretations of event nominals, named masdars by the Arabic tradition. A second case concerns nominals naming units of events. A third case is ‘intensive’ verbs. In treating appropriately these cases and others, our system proves to be more adequate on both empirical and conceptual grounds, compared to other systems.

9

2.1. Masdars as event kind nouns A masdar names an event. Its form from quadrilateral roots or more is quasicanonical and regular, and has normally raised no significant problems in the literature. Its canonical form from triliteral roots, though, is a matter of confusion, and many forms have been taken as canonical, depending on the vocalic pattern of the verb essentially. I believe, however, that event noun formation depends not only on form, but also on meaning. More precisely, it depends on the verb class. The unmarked canonical form comes as CaCC basically. Other forms are more marked, complex, or irregular. It is significant that the canonical form CaCC designates kind events: (44) a. jaraa jary-an run run-acc “He run a running”. b. raqasa raqs-an danced dance-acc “He danced a dancing”. Thus the canonical masdar form is a kind event noun. The kind event noun form contrasts with another canonical form (derived from it), which is more specific. The latter designates an event ‘unit’. It is morphologically marked by the same suffix that occurs on normal kind nouns, to derive a ‘noun unit’, hence its form CaCC-at: (45) a. jaraa jary-at-an run run-unit-acc “He run a run”. . b. raqasa raqs-at-an danced dance-unit-acc “He danced a dance”. c. ´akala ´akl-at-an ate eat-unit-acc “He ate an eat”. It is possible to count event units and interpret them accordingly. This situation contrasts with that of kind event nouns, which do not allow a bounded interpretation: (46) a. raqasa raqs-at-ayni danced dance-unit-dual “He danced two dances”. b. * raqasa raqs-ayni danced dance -dual “* He danced two dances”. This behavior with respect to Number (and also numerals) recalls that of the kind tamr and the unit tamr-at in the nominal domain. Telicity can be used as a test for establishing a parallelism between the event unit nominal and the direct object: (47) a. ´akala ´akl-at-an fii saacat-in ate eat-unit-acc in hour-gen "He ate an eat in an hour". b. ´akala tamr-at-an fii saacat-in ate date-unit-acc in hour-gen “He ate a date in an hour”.

10 In both cases, the nominal phrase functions as the ‘incremental theme’, which ‘converts’ an activity to an accomplishment, a behavior which has led some authors to postulate close derivational as well as semantic relationships between these two predicate classes (cf. e.g. Rothstein 2004). Canonical masdars can be shown to come only from these two classes, but not from achievements or states: (48) a. wajad-a “find” Æ * wajd-an “a finding” b. balagha “reach” Æ * balgh-an “a reaching” (buluugh “reaching”; cf. infra) c. wasala “arrive” Æ * wasl-an “an arriving” (wusuul “arriving”; cf. infra) (49) a. qabuha “become ugly”Æ * qabh-an “* an uglying” (qubhan “ugliness”) b. calima “know” Æ * calm-an “* a knowing” (cilm-an “knowledge”) c. carafa “know”Æ * carf-an “a knowing” (macrif-at “knowledge”) This simple picture of kind event or masdar formation and its meaning fits naturally in the system adopted. Given that masdars are formed only for kind events, or group events eventually, they should be either non-atomic and singulative, or atomic but non-singulative, respectively. In other words, masdars should exhibit integrity, regarding either their parts (K), or their whole (G), but not both. Their denotation must involve a form of ‘plurality’, so to speak. This is not the case for achievements (I), which are fully integral, both as wholes and as parts, nor of states (M), which have no integrity at all. The picture can be refined. Some achievements are not associated with masdars that are of the canonical form CaCC mentioned, but rather with an ‘internal plural’ form of it, CuCCuC, such as buluugh “reaching” and wusuul “arriving”, with respect to (48b) and (48c). It is reasonable to think of these forms as morphologically plural forms, which are formed to express extension, progression, and continuousness. It has been pointed out in the literature, after all, that distinctions have to be made between various achievements, some extended, and others not. Extended achievements are obtained via coercion (see e.g. Caudal 1999, Rothstein 2004). Positional verbs also refine the picture. These verbs have no canonical masdar CaCC, but rather ‘derived’ masdars, so to speak, like those mentioned for extended achievements: (50) a. jalasa (* jals-an) juluus-an sat sitting “He sat a sitting”. b. waqafa (* waqf-an) wuquuf-an stand up standing up “He stood up a standing up”. Doublets of masdars, including a canonical and a ‘plural’ form are found. Motion verbs, a subclass of activities, like xaraja “go out” and daxala “enter”, are expected to have canonically formed event nouns, xarj and daxl. But these masdars are affected to a specific meaning of these roots, basically ‘outcome’ and ‘income’ meanings, as in (51): (51) a. daxala l-maal-u daxl-an entered the-money entering-acc “The money came in”.

11 b. xaraja l-maal-u xarj-an went.out the-money going.out-acc “The money went out”. Consequently, motion verbs take a kind event noun only in the ‘internal plural’ form: (52) a. xaraja xuruuj-an (* xarj-an) went.out going.out-acc “He went out a going out”. b. daxala duxuul-an (* daxl-an) entered entering-acc “He entered an entering”. Apparent masdars of the canonical form come from stative verbs like fahima fahm-an “to understand”. But one might wonder whether these are real masdars. Evidence against their masdar nature comes from the fact that they form no event unit noun, hence behavoring like other stative verbs (cf.53b): (53) a. fahima fahm-an (* fahm-at-an) understood understanding-acc understanding-unit-acc “He understood an understanding“. b. kariha * karh-an (* karh-at-an) hated hating-acc hating-unit-acc “He hated (* a hating)”. In fact, the existence vs. absence of event unit nouns provides indirect evidence that verbs have potential canonical masdars or they do not, as we will see. For example, positional and motional verbs discussed above all have event unit nouns, which normally derive from the event kind noun. This indicates that the masdar is somehow only ‘accidentally’ missing: (54) a. jalasa jals-at-an sat sitting-unit-acc “He sat a sitting”. b. daxala daxl-at-an entered enter-unit-acc “He entered an entering”. We see then that modulo lexical/grammatical dissimilarities, forms of masdars and their meanings come as expected, and surprisingly regular, contrary to what the traditional literature on the matter has claimed. Our system predicts the properties of the computational core, besides coercive interpretation and lexical constraints. 2.2. Event unit nouns Event unit nouns, like masdars, come productively from activities. This is a strong prediction made by our system. Here are some examples: (55) a. ´akala ´akl-at-an ate eat-unit -acc “He has eaten an eat”. b. raqasa raqsat-an danced dance-unit-acc “He danced a dance”. Observe again that the event unit nominal constitutes the incremental theme that leads to telicity. As mentioned earlier, this nominal phrase is not compatible with a direct object. In the relevant interpretation, both constituents fulfill the same role:

12 (56)?? ´akala t-tuffaah-at-a ´akl-at-an ate the-apple-unit-acc eat-unit-acc "He ate an apple an eat". In this sentence, the kind event noun is possible only under an adverbial interpretation. The event unit noun does not come from achievements or states. This correlates with the fact that the event kind noun cannot be formed from these verbs either (recall (48) and (49) above): (57) a. wajada l-hall-a (* wajd-at-an) found the- solution-acc finding-unit-acc “He found the solution (* a finding)”. b. balagha l-qimmat-a (* balgh-at-an) reached the-summit-acc reach-unit-acc “He reached the summit (* a reaching)”. (58) a. kariha (* karh-at-an) hated hating-unit-acc “He hated (* a hating)”. b. qabuha (* qabh-at-an) uglied uglying-unit-acc “He became ugly (* an uglying)”. The non-availailability of the kind noun points to the absence of a source of derivation for the unit noun. The existence of event unit nouns of some verbs without the existence of canonical kind nouns may then appear at first sight problematic. There is, however, a distinct behavior of two classes of verbs, depending on their semantics. The class in (59), which includes positional verbs and motion verbs, form event unit nouns, but that of achievements verbs in (60) does not: (59) a. jalasa Æ juluus-an Æ jals-at-an “sit, sitting, a sitting” b. waqafa Æ wuquuf-an Æ waqf-at-an “stand up, standing up, a standing up” c. xaraja Æ xuruuj-an Æ xarj-at-an “go out, going out, a going out” (60) a. balagha Æ buluugh-an (*balgh-an) Æ *balgh-at-an "reach, reaching, a reaching” b. wasala Æ wusuul-an (*wasl-an) Æ * wasl-at-an "arrive, arriving, an arriving” That suggests that the verbs in (59) do in fact have a potential canonical kind event noun, from which the event unit noun is derived, but verbs in (60) do not have such a kind event noun, and hence the unit event noun cannot be so derived. The impossibility of deriving the unit event noun for achievements is then predicted, and the ‘internal plurality’ of the kind event noun of these verbs must be different from that found in (59). It correlates with the extension meaning due to coercion. The interpretation of (59), however, does not involve coercion, nor extension. We can think of the latter masdars as alternating forms of canonical forms, due to lexical dissimilarity. Both forms imply potential plurality, from which the canonical event unit is derived. Summarizing, the formation of masdars or pseudo-masdars, on the one hand, and that of unit event nouns, on the other, supports our new classification of verbal phrases, which differs substantially from the most spread one in the literature, i.e that based on a generalized binary [± count] distinction.

13

2.3. Intensive predicates and the Imperfective Paradox When we consider so-called intensive predicates, we find patterns that fit rather nicely with the previous distinctions. These predicates are formed only from a subclass of activities and accomplishments: (61) qattaca l-lahm-a (qataca) cut.intens the-meat-acc (cut) “He cut the meat into pieces” (62) ghallaqa l-´abwaab-a (ghalaqa) closed.intens the-doors-acc (closed) “He locked firmly the doors”. (63) a. jawwala (jaala) walked.intens took.walks “He took a lot of walks”. b. tawaffa (taaf-a) went.about.intens went.about “He went a lot about”. Likewise, change-of-state verbs in Moroccan Arabic like ftəh “open” can be constructed with the geminated intensive: (64) fttəh l-wərd (ftəh) opened.intens the-flowers (opened) “Flowers opened widely”. However, intensive forms of achievements or statives are not found: (65) * hammar-a l-ward-u reddened.intens the-flowers-nom “* Flowers reddened intensively”. (66) * wajjad-a l-hall-a found.intens the-solution-acc “* He found intensively the solution”. Note that event kind nouns and event unit nouns related to the above mentioned predicates are regularly formed: (67) a. taqtiic, taqtiic-ah “cutting into pieces, a cut into pieces” b. tatwiif, tatwiif-ah “going a lot about, a go a lot about” The contrasts between the two classes of predicates can be accounted for if we assume that verbal phrases (like nominal phrases) are dominated by a classifier phrase ClP, and that the latter is dominated by a Number Phrase NbP (instead of AspP, the widely spread assumption in the literature). Activities can obviously be distinguished from accomplishments, despite their similarities. For example, they behave differently regarding the Imperfective Paradox: (68) r-rajul-u y-a´kul-u → r-rajul-u ´akala the-man-nom 3-eat the-man-nom ate “The man eats” → “The man has eaten”

14 (69) y-a´kul-u t-tuffaah-at-a → ´akal-a → * ´akala t-tuffaah-at-a 3-eat the-apple-acc ate ate the-apple-acc “He eats the apple” → “He has eaten” → “He has eaten the apple” The activity implies an accomplishment, more precisely a singulative accomplishment. But the accomplishment does not, because it is not singulative. In the first case, we have an imperfective reading, expressing the fact that the eating started in time but has not come to an end. We may think of the imperfective then as taking an atomic event and stretching it over time, or de-atomizing/pluralizing the atomic event with respect to time. In fact, the atomic event is supposed to take place at many times. Perfective, on the other hand, is atomic, indicating that the process has taken place at an atomic time. Imperfective corresponds then to Plural, and Perfective to Singular. Atomicity may apply to times, or to units of events. Pluralizing might be de-atomizing as well as multiplying atoms. Imperfective can apply only to events which have different values for atomicity and singulativity. It cannot apply to states, because they have no singular or atomic potentials. It cannot apply to achievements, because their meaning depends on being both atomic and singulative, whereas the imperfective is [atom]. Perfective applies without problem to achievements, accomplishment, or activities, forcing them to be atomic, as in the following French example with the passé simple (simple past): (70) Il courut, mangea une pomme et arriva à 6h. “He run, ate an apple, and arrived at 6”. Note that Aspect, notably [± perfective] Asp (= Pfv), preserves internal/lexical aspect, or rather the internal properties of the predicate in terms of atomicity and singulativity, as I have shown. This makes Asp behave exactly like Number with respect to the internal classification. Having in mind the (external) Aspect/Number equivalence and the (internal) Aspect/Classifier equivalence, we can then propose that the functional structure of verbal predicates is as follows: (71) [ Nb/Pfv [Cl [VP]]

3. A comparison with other systems In evaluating feature systems, questions about conceptual and empirical adequacies arise. Our system appears to be conceptually simple, natural and homogeneous in terms of the features used. Furthermore, it is empirically more adequate in that it cross-classifies appropriately, hence accounting for distributions of subclasses that are not expected under other systems, as we have seen. In this section, I review briefly some of the systems available in the literature, and show their conceptual and empirical shortcomings. 3.1. Jackendoff’s (1991) system Jackendoff (1991) establishes a quadripartite classificatory system for objects (or things), based on two features: [± bounded] and [± internal structure]. Objects are then classified as follows (b for bounded, i for internal structure): (72) Object classification a. Individual:

[+b, -i]

‘a dog’

b. Group:

[+b,+i]

‘a team’

15 c. Substance (mass): [-b, -i]

‘oil’

d. Aggregate (kind): [-b, +i]

‘buses, cattle’

The feature [± bounded] indicates that the object has a boundary, or it does not. An apple has a boundary, but apples does not. The [± internal structure] feature distinguishes objects, the denotation of which entail “a medium comprising a multiplicity of distinguishable individuals” marked as [+ i], whereas substances and individuals lack such an entailment. Individuals then cross-classify negatively with masses, and have nothing in common with aggregates. Aggregates and groups crossclassify positively, although one is bounded, and the other is not. Groups are the most marked member of the system. Boundedness appears at first sight to be close to atomicity in my system, and internal structure close to singulativity. But this is not so. Jackendoff’s system appeals, in fact, to features of different ontological nature, and of somewhat unclear nature. For example, it is not clear what it means to talk about objects with no internal structure, given that all objects are supposed to have one, including masses which are distributive and cumulative by definition. As for empirical predictions of the system, it can be illustrated by examining the case of groups. Groups are the most marked member in this system, as I observed earlier, contrary to what is normally assumed in the literature. Greenberg (1972) and others, for example, take collectives to be the unmarked marker. This is supported by the fact that collectives are not morphologically marked, normally, compared to individuals which are marked in e.g. Arabic and Chinese. Likewise, kinds are marked by plural morphology in English, but not groups.5 As for structural parallelisms between events and objects, they are not established between the four classes of (72) and the four Vendlerian classes of verbal predicates. The parallelism is rather established with non-acknowledged (or non-established) classes of events, which are described as follows: (73) a. Closed (bounded) event: ‘John ran to the store’; ‘The light flashed’ b. Bounded iterative event: ‘The light flashed until dawn’ c. Unbounded homogeneous process: ‘John slept’ (not repetitive) d. Unbounded iterative process: ‘The light flashed continually’ States are not included in this classification. The analysis appeals to an additional feature, ‘iterativity’, to make the description fit. Moreover, it does not indicate how the nominal features values are paralleled by verbal ones. According to Jackendoff, the Vendlerian classes pertain to entire sentences rather than verbs, and in order to characterize them he appeals to new features: direction and dimension. The classification appears then as follows: (74) a. State: [- i [ - DIR] ] b. Activity:

[- b [

DIR]

]

c. Accomplishment: [+ b [DIM x, d. Achievement: [+ b, - i [DIM

DIR y]

]

0, DIR y]]

16 The system is then neither parallel nor comparable to the object system. Furthermore, it is much more complex, and far from the initial simple and elegant parallelism introduced by Bach (1986) and others, in which the count/mass nominal distinction parallels that of temporally bounded/unbounded eventualities. Jackendoff claims that lexical items can be conceptually decomposed into primitives, and raises the question of how one justifies primitives. To the question whether one primitive can be motivated better than another, Jackendoff answers negatively. It is only motivated in the context of the overall system in which it is embedded, which should then be justified on the grounds of its capacity for expressing generalizations and explaining the distinctions of the data, i.e. by the usual scientific standards of evaluation. Second, as for whether primitives must be really primitive, i.e. whether one is all the way at the bottom, he takes the view that this should not be a matter of worry, since a system can always be further decomposed. It seems to me, however, that Jackendoff is not perspicuous enough about evaluation, and one may wonder whether a primitive is relevant or natural to characterize a generalization or a distinction. For example, is it the case that [± count] is interchangeable with [± bounded], [± telic], or [± atomic]? The answer appears to me to be no. Third, the nature of composition of the overall system may matter: assuming that the feature system in syntax is binary, one might wonder how these features would be grouped together to fit in a system based on a count/non-count distinction. Indeed, he question of grammaticalization is a real one. 3.2. Verkuyl’s system Verkuyl (1993) assumes a tripartite classificatory system for eventualities, using two features: [± add to] and [± specified quantity]. He does not deal with objects. The first feature refers to a predicate which needs a complement or an extension via an object or a prepositional phrase to form an aspectual class. The second feature qualifies the complement as a specific quantifier phrase (= SQA). Thus although the author talks about an aspectuality system, the features used are not of aspectual nature. The first one points basically to constituency, and the second to quantity. The classification is then as follows: (75) a. Event: [+ ADD TO, + SQA] b. Process: [+ ADD TO, - SQA] c. State: [- ADD TO, ± SQA] Event cover both accomplishments and achievements, and Process is equivalent to activities. Moreover, an orthogonal distinction is made between what is temporal (± ADD TO) and what is atemporal (± SQA). The system fails to distinguish accomplishments and achievements, although they are distinct in terms of both aspect and tense behaviours. Questions arise also with respect to the nature and adequacy of the features used. Furthermore, no parallelism is established with the nominal domain. These characteristics of the machine make it difficult to compare it to the machine adopted here, regarding typically the fact that it is unable to capture the distributions and architecture parallelisms examined in this work.

17

3.3. Rothstein (2004) Rothstein claims that all verbal predicates have their denotations in the count domain, including stative verb predicates (which denote sets of countable stative eventualities). It is adjective predicates that denote in the mass domain. Two aspectual properties distinguish four verbal classes, depending on whether they ′naturally′ head telic VPs, and they naturally occur with the progressive: (76) a. Achievement: [+telic, -stage] b. Accomplishment: [+telic, +stage] c. Activity : [-telic, +stage] d. State : [-telic, -stage]. A VP is telic if it denotes a set of countable events, and expresses criteria for individuating atomic events. A Telicity Principle states that a VP is telic if it denotes a set of events X which is atomic (or a pluralization of an atomic set). A bare V denotes a set of events which in some context will count as atomic instances of V. So V denotes a set of singular entities, but not a set of atoms. VC is a subset of V consisting of events of V which are atomic in a context. Atomicity is defined as follows: (77) x is an atom of P in a context C if there is no y which is an atom of P in C such that x is a proper part of y. Atoms are things which are counted, and they have a cardinality of 1 (in line with Landman’s views; see the references). Counting is picking out individuals according to a particular criterion of individuation. There are predicates which are taken to be naturally atomic (such as dog), since they have internal individuating structure out of context, and those that need contexts (such as fence, table). Rothstein equates telicity with atomicity, and she uses sets to determine telicity: (78) A verbal P is telic if it denotes a set Vc, i.e. an atomic set contextually restricted by a context , or a pluralization of such a set. ( is the atomic measure function, at a particular time). Rothstein’s system comes close to mine in taking achievements and accomplishments to be telic/atomic, whereas states and activities are not so. Second, it is close in distinguishing singular sets and atomic sets. But other claims are problematic or unmotivated. For example, the [± stage] feature is disputable, because it does not deal with internal properties of vps, but rather with external combinability with the progressive, hence manifesting confusion in identifying hierarchical functions. As observed above, external Aspect is classification preserving, and therefore cannot be used for the purpose of determining internal classification. Other conceptual and empirical criticisms can be addressed to the system. For example, Rothstein describes achievements as naturally atomic predicates, and states natural atomicity as follows: (79) ‘Natural’ atomicity When a P is naturally atomic, its singular set and its atomic set are identical. Her singularity is close to my singulativity, and so is her atomicity with respect to mine. But the identity of values of the two features is just stipulated, rather than

18 derived from feature values which cross-classify. Likewise, the atomicity of accomplishments is construed as follows: (80) An accomplishment is atomic when it is the sum of an activity and an atomic BECOME event. Here too, the nature of atomicity is stated, rather than derived. Furthermore, one might wonder if there is evidence that accomplishments are really a sum of two events, or whether they parallel nominal groups, with the properties that I have postulated.6 More importantly, the origin of telicity as a specific mereological characteristic, or an atomic one, is rather unclear. Likewise, taking all the verb predicates to belong to the count domain, with exclusion of the mass domain, leaves no room for establishing the difference between counting times and counting eventualities, as I have observed with respect to the contrasts in (39) to (42) above. It does not leave room either for establishing direct parallelisms between nominal and verbal classes, an option that I believe is not conceptually or empirically motivated.

4. Conclusion In this article, I have provided bases for closer parallelisms between verb and noun classes and functional projections than ever been thought before. I have established a unified quadripartite classification for both nouns and verbs, based on the atomic and singulative features. If the analysis is correct, then verb predicates should be classified along the same lines that nominal predicates are classified, and both vps and nps should be dominated by classifier phrases. Likewise, parallelisms between (perfective) Aspect and Number provide substantial reason to think of the former as simply Number in the verbal domain. Further explorations of this system are needed, in particular with respect to Chomsky’s phases, as well as higher functional projections, related in particular to the tense domain.7

19 Notes 1. This judgment does not deny valuable efforts found in the literature, which started heading toward the right direction, among which Bach (1986), Jackendoff 1991), Rothstein (2004), Borer (2005 a &b), and Krazer (2002), to cite a few. As for the bases of the system proposed here, see Fassi Fehri (2003b and 2004b), as well as Fassi Fehri & Vinet (2004). 2. Greenberg attributes the first use of the term 'singulative' to Zeuss, who used it in Celtic for derivational formation which corresponds in these languages to the Arabic 'noun of unit'. Greenberg suggests that we are then dealing with a 'three term system', "… in which a collective which cannot be used with numerals is opposed to a singulative with its own singular and plural" (as in Arabic; p. 179). The basic opposition is then collective/singulative, rather than singular/plural (p. 181). The generic noun is equated with ism al-jins “kind noun” (p. 182). The classifier is an individualizer which performs the same function as a singulative derivational affix in languages with collective/singulative opposition (p. 184). Grenberg’s view can be summarized in the following tree: (i) NB / \ Col Sing / \ Sg Pl Since we take what appear with NB [± atom], Greenberg's singulative appears to correspond to [± atom], while its Col corresponds to [Ø atom]. Note that Greenberg's Col does not correspond to a Group, which is left unaccounted for. Mass is not dealt with either. Our notion of Sing crucially distinguishes Mass from Kind, and Collectives are treated as Groups, which are atomic. 3. This association is inspired by Rothstein (2004), following herself a proposal by Landman (see the references there). 4. Note that the same discontinuous affix t—t is used for the feminine formation, in parallel to what happens in Arabic: (i) aserdun «mule male” Æ t-aserdun-t “mule female” ; amazigh “ berber male” Æ t-amazigh-t “berber female” . 5. Jackendoff takes plurality, grinder, packager etc. as functions on the conceptual feature matrices in (72), and differences in entailment are a consequence of inference rules that refer to the features in question. 6. Note that in order to distinguish accomplishments from achievements, Rothstein appeals to two distinct principles of Cardinality, which she formulates as follows (her 82 a&b): (i) Cardinality Principle The cardinality of an Accomplishment e is the same as the cardinality of its BECOME e. (ii) Cardinality Principle 2 A singular BECOME e which is naturally atomic has a cardinality of 1 if it has an atomic argument. This seems to us to be lacking simplicity as well as motivation. 7. In Fassi Fehri (2003b), it is suggested that np/ClP and DP are phases of the nominal domain, which are parallel to vp and CP in the clausal domain. See also Guéron (2004) for appealing proposals with regard to space and temporal phases.

20

References Aikhenvald, A. 2001. Classifiers: A typology of noun categorization devices. Oxford, New York: Oxford University Press. Astarabaadii, R. 12th c. Sharh al-kaafiyyah. Beyrouth: daar t-tibaacah l-cilmiyyah. 1976. Borer, H. 2005a. Structuring Sense. In Name Only. New York: Oxford University Press. Borer, H. 2005b. Structuring Sense. The course of events. New York: Oxford University Press. Bunt, H.C. 1985. Mass Terms and model-theoretic Semantics. Cambridge: Cambridge University Press. Carlson, G. 1977a. Reference to Kinds in English. PhD diss. Univ. of Mass at Amherst. Carlson, G. 1977b. "A Unified Analysis of English Bare Plural". Linguistics & Philosophy 1.413-58. Carlson, G. & F. J. Pelletier eds. 1995. The Generic Book. Chicago: Univ. of Chicago Press. Caudal, P. 1999. “Achievements vs. accomplishments”. TALN. Cargèse. Cheng, C.Y. 1973. "Comment's on Moravcsik's paper". Approaches to natural language, ed. by J. Hintikka et al. Dordrecht: Reidel. Chierchia, G. 1998a. "Reference to Kinds across languages". Natural Language Semantics 6.339-405. Chierchia, G. 1998b. "Plurality of Mass Nouns and the Notion of "Semantic Parameter"". Events and Grammar, ed. by S. Rothstein, 53-103. Dordrecht: Kluwer. Chomsky, N. 1995. The Minimalist Program. Cambridge Mass: MIT Press. Chomsky, N. 1999. "Derivation by Phase". MITOPL 19. Cambridge: MIT. Cinque, G. 1999. Adverbs and Functional Heads. New York: Oxford University Press. Dobrovie-Sorin, C. 2002. "Adverbs of Quantification and Genericity: From DPs to NPs". Ms. Univ.of Paris VII. Dobrovie-Sorin, C. & B. Laka. 1996. "Generic bare NPs". Ms. Univ. of Paris VII and Univ. of Strasbourg. Fassi Fehri, A. 1993. Issues in the Structure of Arabic Clauses and Words. Dordrecht: Kluwer. Fassi Fehri, A. 1999. "Arabic modifying adjectives and DP structures". Studia Linguistica 53.105-154. Fassi Fehri, A. 2003a. "Arabic Perfect and Temporal Adverbs". Perfect Explorations, ed. By A. Alexiadou, M. Rathert, & A. von Stechow ed, 69-99. Berlin: de Gruyter.

21 Fassi Fehri, A. 2003b. "Nominal classes and parameters across interfaces and levels, with particular reference to Arabic". Linguistic Research 8.2.9-103. Rabat: IERA Publications. Fassi Fehri A. 2004a. "Temporal/Aspectual Interaction and Variation across Arabic Heights". The Syntax of Time, ed. by J. Guéron & J. Lecarme, 235-257. Cambridge, Mass: MIT Press. Fassi Fehri, A. 2004b. "Nominal classes, reference, and functional parameters, with particular reference to Arabic". Linguistic Variation Yearbook 4. Amsterdam: John Benjamins. Fassi Fehri, A. & M. T. Vinet 2004c. "Number and Classifier Distributions in Arabic and Chinese". Linguistic Research 9.1. Rabat: IERA Publications. Filip, H. 2005. “Telicity revisited”. Ms. Gestner, C. & M. Krifka 1987. "An outline of Genericity". SNS-Bericht 87-23. Univ. of Tübingen. Gillon, B. 1992. "Toward a common semantics for English count and mass nouns". Linguistics & Philosophy 15.597-639. Greenberg, J. 1972. "Numeral Classifiers and Substantival Number: problems in the Genesis of a Linguistic Type". Proceedings of the Eleventh Congress of Linguists, 166-193. Bologna: Società editrice il Milano. Guéron, J. 2003. "Inalienable possession and the interpretation of determiners". From NP to DP, vol. 2, ed. by M. Coene & Y. D'hulst, 189-220. Amsterdam: J. Benjamins. Guéron, J. 2004. "Tense, Person, and Transitivity”. Ms. Univ. of Paris III. Jackendoff, R. 1991. Parts and boundaries. Cognition 41.9-45. Kratzer, A. 1996. "Severing the external argument from its verb". Phrase structure and the Lexicon, ed. by J. Rooryck & L. Zaring, 109-138. Dordrecht: Kluwer. Kratzer, A. 2002. “The Event Argument”. Ms. Amherst: Univ. of Mass. Krifka, M. 1995. "Common Nouns in Chinese and in English". The Generic Book, ed. by G. Carlson and F.J. Pelletier, 398-411. Chicago : University of Chicago Press. Krifka, M. 1995. et al. "Genericity: an introduction", ed. by G. Carlson & F. J. Pelletier, 1-124. Landmann, F. 1996. "Plurality". The Handbook of Contemporary Semantic Theory.by S. Lappin, 425-457. Oxford: Backwell. Lewis, D. 1975. "Adverbs of quantification". Formal Semantics of Natural Languages, E. Keenan ed, 3-15. Cambridge: Cambridge University Press. Li, H-Y. A. 1999. "Plurality in a Classifier language". Journal of East Asian languages 8.75-99. Link, G. 1983. "The logical analysis of plurals and mass terms". Meaning, use, and interpretation in language, ed. by R. Bauerle et al., 303-323. Berlin: de Gruyter. Longobardi, P. 2003. "Toward a Unified Theory of Reference". Ms. Univ. of Trieste. Nicolas, D. 2001. "Do mass nouns constitute a uniform class?" Kansas WPL.

22 Pelletier, F. J. 1979. "Non-singular reference". Mass terms, ed. by F.J. Pelletier, 1-14. Dordrecht: Reidel. Pelletier, F. J. & L. Schubert.1989. "Mass expressions". The Handbook of Philosophical Logic, vol. 4, ed. by. D. Gabbay & F. Guenthner, 327-407. Dordrecht: Kluwer. Quine, W.V. 1960. Word and object. Cambridge, Mass: MIT Press. Rothstein, S. 2004. Structuring Events. London: Blackwell. Schwarzschild, R. 1996. Pluralities. Dordrecht: Kluwer. Sharvy, R. 1978. "Maybe English has no count nouns: notes on Chinese semantics". Studies in Language 2.345-365. Sharvy, R. 1980. "A more general theory of definite descriptions". Philisophical Review 89.4, 607-624. Sibawayhi, A. 8th c. al-kitaab. Cairo: Bulaaq. 1938. Simons, P. 1987. Parts. Oxford: Clarendon. Szabolcsi, A. 1994. "The Noun Phrase". The syntactic structure of Hungarian. Syntax and Semantics, vol. 27, ed. by F. Kiefer & K. E. Kiss, 179-274. San Diego: Academic Press. Wiese, H. & J. Maling (to appear). “ Beers, kaffi, and Schnaps:Different grammatical options for ‘restaurant talk’ coercions in three Germanic languages”. Journal of Germanic Linguistics. Wright, W. 1898. A Grammar of the Arabic Language. Translation from Caspari, with edition, corrections and additions. Third edition. Cambridge: Cambridge University Press. Zribi-Hertz, A. & H. Glaude. 2003. To appear. "Bare NPs and Deficent DPs in Haitian and French: from morphosyntax to referent construal". Bare nouns in Creole Languages, ed. by M. Baptista & J. Guéron. Amsterdam.