<<

Draft for comments (2015) - Do not cite without author’s permission

1 The neuter suffix -ik- in Bantu

2 Sebastian Dom 3 KongoKing Research Group, Department of Languages and Cultures - Africa, 4 Ghent University, Belgium 5

6 1. Introduction 7 The agglutinative system of verbal suffixes is a fairly well-known grammatical feature of Bantu 8 languages among (typological) linguists. One would thus assume that in the tradition of Bantu 9 linguistics, these suffixes have been extensively analyzed and described. Unfortunately, this 10 can only be said to be true for a small number of all verbal suffixes reconstructed for Proto- 11 Bantu. The few verbal suffixes that have been the topic of a large number of detailed syntactico- 12 semantic studies are the applicative (Alsina & Mchombo 1990, 1993; Baker 1988, 1990; 13 Bostoen & Mundeke 2011; Bresnan & Moshi 1990; Cann & Mabugu 2007; Creissels 2004; 14 Dammann 1961; Demuth 1998; De Kind & Bostoen 2012; Harford 1993; Kähler-Meyer 15 1966)(Marten 2003, 2011, Marten & Kula 2014, Nakamura 1997, Ngonyani 1998a, 1998b, 16 2000, Ngonyani & Githinji 2006, Port 1981, Zeller & Ngoboka 2006), the passive (Biloa 1994, 17 Fleisch 2005, Givón & Kawasha 2006, Kawasha 2007, Kimenyi 1988, Kula & Marten 2008, 18 Stappers 1967, Trithart 1979, Woolford 1994) and the reciprocal (Dammann 1954, Dom et al. 19 forthcoming, Maslova 2000, 2007, Mchombo 1990, 1999, Mchombo & Ndunga 1994, 20 Mchombo & Ngalande 1980, Mudzingwa 2008, Mugane 1999, Ndayiragije 2002). Quite many 21 suffixes found in (the?) are still poorly described both from a syntactic and 22 semantic viewpoint, e.g. the extensive, the neuter, the positional, the repetitive, the separative, 23 and the tentive.1 A first aim of this is to provide an in-depth analysis of one such suffix, 24 the neuter (the term used in Schadeberg 2006). To my knowledge, there are only few studies 25 dedicated specifically to this suffix (Dubinsky & Simango 1996, Khumalo 2009, Mischke 1994, 26 Mchombo 1993, Seidl & Dimitriadis 2002), including one unpublished and unfinished draft 27 (Schadeberg 2004). In other articles, the middle suffix is concisely treated as part of a bigger 28 study, e.g. of an investigation of the verbal suffix inventory in some Bantu languages (e.g. 29 Ondo-Mebiame 2007, Westphal et al. 1974), and often neither a semantic description nor 30 syntactic analysis is provided. Many grammars of Bantu languages (of which most date back 31 to the first half of the 20th century) have a section on this middle marker. However, these never 32 exceed the length of a paragraph, often including only (some) lexical examples. 33 As I will show in the following sections, the middle suffix -ik- in Bantu languages marks (i) 34 anticausative constructions, (ii) ‘agentless passive’ construction, (iii) potential constructions; 35 when suffixed onto perception it can mark (iv) stimulus constructions and/or (v) 36 evidential constructions. The majority of the specialized articles on this suffix is written in a 37 generative framework and only treat functions (i)-(iii). Therefore, the second main objective of 38 the article is to provide the first typologically-oriented description of this middle marker in 39 Bantu, and to highlight the additional functions (iv) and (v). Typological works on

1 I use the names of the suffixes proposed by Schadeberg (2006). 1

Draft for comments (2015) - Do not cite without author’s permission

40 anticausativity, potentials (also termed ‘potential passives’, ‘middles’, ‘middle passives’ or 41 ‘medio passives’; Haspelmath 1987: 9), perception verbs or evidentiality generally do not 42 provide much information on Bantu languages (if any at all, besides Swahili; but see Botne 43 1995, 1997, on evidential markers in Lega). The present study thus provides typologists with 44 an extensive account on the expression of these categories and constructions in a variety of 45 Bantu languages. Moreover, the previous studies on the middle suffix have only been concerned 46 with syntactic properties of the constructions marked by this verbal morpheme, generally 47 focusing on the differences between the (anticausative) middle and passive. In addition to a 48 description of the syntactic behavior of these middle constructions in Bantu, I will also give a 49 semantic and pragmatic analysis of the middle marker and its functions. 50 The study is based on a language sample of 117 Bantu languages. It is a convenience sample 51 (Bakker 2011: 106) in that the data consist of all the information that was available to me. The 52 sources of the data can be categorized in three types. The first are descriptive grammars. The 53 body of descriptive grammars consulted for this study represents the entire collection of 54 descriptive grammars of Bantu languages of the library of African Languages and Literature of 55 the University of Ghent. The second type are articles either dedicated to or summarily reporting 56 on the middle suffix. The third type consists of corpus material. New data has been extracted 57 from (i) the corpus of Deo Kawalya, a PhD student at Ghent University and native 58 speaker of Luganda, with whom I have closely collaborated for the interpretation of the 59 Luganda examples and elicitation of additional examples, and (ii) from the Kikongo corpus of 60 the KongoKing research group, located at the University of Ghent. A full list of the Bantu 61 languages included in the sample is provided in appendix. 62 The paper is structured as follows: in section 2 an introduction is provided regarding the 63 middle marker, including a discussion of the terminology and the various present-day forms of 64 the suffix attested in the Bantu domain. Section 3 discusses the first function of the middle, 65 which is to mark anticausative constructions. Section 4 treats agentless passive constructions, 66 which have the same structure as anticausatives, but do not have anticausative verbs 67 as their predicate. In section 5, the anticausative/agentless passive constructions are compared 68 with passives, with regard to certain syntactic and semantic features such as oblique agentive 69 phrases, instrumental phrases, agent-oriented , and resultative vs. perfective aspect. 70 Section 6 treats the potential construction, providing an in-depth semantic analysis unifying the 71 interrelated concepts of force dynamics, modality and subjectivity. The seventh section 72 discusses two specialized functions of the middle marker when combined with perception 73 verbs, i.e. stimulus constructions and evidentiality. Conclusions are summarized in section 8.

74 2. The middle marker in Bantu 75 A first point worthy of discussion is the terminology used to label this marker. The following 76 list sums up the commonest terms found in the sources to refer to one and the same verbal 77 suffix. Some labels are only attested once, others are more recurrent, ‘neuter’ being the most 78 frequently used term.

79 Labels used for the middle marker

2

Draft for comments (2015) - Do not cite without author’s permission

‘Capable’ form neutro-passive derived intransitive passive Impositive potential neutrale/mediale Form (German) reflexive impositive intransitif positif (French) resultative Intransitive spontaneous intransitive Subjectivform (German) static medio-passive stative neuter tolerative neuter passive neuter-directive neuter-stative

80 Schadeberg (2006: 75) comments on the term ‘neuter’ that it “has the disadvantage that it 81 suggests a general syntactic function (closer to passive than to active ).” Schadeberg (ibid.) 82 continues that the suffix combines most often with verbs of destruction and experiencer verbs, 83 and thus that the label ‘neuter’ “does not express the specific link with the two semantic 84 categories mentioned.” However, the alternative Schadeberg proposes, namely ‘neutro- 85 passive’, does not seem to solve the problem even though he writes that it is “a more precise 86 semantic-syntactic label.” Several studies have shown that there are fundamental differences 87 between anticausative and passive constructions (among others Alexiadou et al. 2006, Kallulli 88 1999 for Indo-European languages, and Dubinsky & Simango 1996 and Seidl & Dimitriadis 89 2002 for Bantu languages). In this sense, it would be misleading to add ‘passive’ to the label. 90 From a typological perspective ‘neuter’ can be confusing in that it is often used in the context 91 of gender for languages with a tripartite gender distinction based on sexes (male, female and 92 neuter gender). Schadeberg (2006: 75) does convincingly argue that other labels such as stative 93 or intransitive are too abstract or better suited for other Bantu suffixes. Terms such as 94 ‘potential’, ‘stative’, ‘medio-passive’, ‘spontaneous’ or ‘resultative’ cover only one part of the 95 functions of the middle marker. I therefore opt to label the suffix with the more general term 96 ‘middle marker’ in the sense that it encodes a number of ‘deagentivized syntactic patterns’ 97 (Kulikov 2013: 265). 98 The middle marker has been reconstructed for Proto-Bantu as *-ɪk-, with a second-degree 99 closed (and even further back for Proto-Niger-Congo as *-ke; Voeltz 1977 in 100 Hyman 2007: 151). The Proto-Bantu form has evolved into a variety of manifestations in 101 present-day languages due to the diachronic 7-to-5 vowel merger and sonorization and lenition 102 of the consonant *k. Some Bantu languages have lost the opposition between [ɪ] and the first- 103 degree vowel [i] with the subsequent omission of the former. This process, in which the seven 104 vowel system of Proto-Bantu has been reduced to five in some present-day languages, is also 105 called the 7-to-5 Vowel Merger (Schadeberg 1995). Other Bantu languages have retained a 106 seven vowel system, either in its initial or in a modified form. However, in very few grammars 107 of languages having retained the aperture opposition, are the vowels [ɪ] and [ɛ] distinguished 108 from [i] and [e] by the authors. The list of attested forms of the middle suffix, presented below, 109 is therefore not exhaustive. Moreover, the closed vowel can alternate with the mid-close vowels

3

Draft for comments (2015) - Do not cite without author’s permission

110 [e] or [ɛ] due to the synchronic morpho-phonological rule of vowel height harmony (Hyman 111 1999).

112 Attested forms of the middle marker in present-day Bantu languages 113  -ik-, -ɪk-, -k-,-ig-, -iχ-, -iɣ-, -ih-, -iɦ-, -i- 114  -ek-, -eg-, -eɣ-, -eχ-, -eh-, -eɦ-, -ey-, -e-, -ɛ, -ə

115 3. Anticausative constructions 116 One function of the middle suffix is to mark the anticausative member of the 117 causative/anticausative alternation. Some examples are given in (1)-(3). The constructions in 118 (a) are the transitive-causative member of the alternation pair, and the constructions in (b) are 119 the anticausative counterparts.

120 (1) BENA (G63)2 (Morrison 2011: 370) 121 a. u-mu-ana a-haa-deeny-ile u-tu-bihi igólo.3 122 AUG1-CL1-child SM1-PST-break-FV AUG13-CL13-tree yesterday 123 ‘The child broke the twig yesterday.’ 124 b. u-tu-bihi a-haa-deeny-ih-ile igólo. 125 AUG13-CL13-tree SM13-PST-break-MID-FV yesterday 126 ‘The twig broke yesterday.’

127 (2) TSWANA (S31) (Creissels 2002: 403) 128 a. Ngwana o thubile mae. 129 ƞwàná ɷ́ -thùb-íl-é màɩ́. 130 CL1:enfant SM1-casser-PRF-FV CL6:œuf 131 ‘L’enfant a cassé les œufs.’ (‘The child has broken the eggs.’) 132 b. Mae a thubegile. 133 màɩ́ á-thúb-èχ-ìl-è. 134 CL6:œuf SM6-casser-MID-PRF-FV 135 ‘Les œufs se sont cassés.’ (‘The eggs are broken.’)

136 (3) KIGA (JE14) (Taylor 1985: 147) 137 a. Rutafa y-aa-yat-a amate. 138 Rutafa SM-PST-spill-FV milk

2 For each Bantu language I give the code of Maho’s (2009) revized version of Guthrie’s referential classification (1971) of the Bantu languages. 3 Using examples from different articles, there is no unified way of glossing which I could adopt. Different authors gloss differently. I have tried to conform most of the various systems into a unified way of glossing in this paper. However, I have not been able to resolve some inconsistencies. For example, some authors have indicated the class number of the prefixes, and others have not. Since I am not familiar with the grammar and noun class system of every language treated in this paper, examples taken from sources in which the class number is not indicated lack them here as well. Some authors separate the noun class marker from the nominal stem by means of a dash, whereas others do not but indicate the class number in the gloss by means of a colon. [you don’t ned a long discussion here; simply mention that YOU follow Leipzig glossing Rules!] Again, I do not know for all the languages which part of the noun is the prefixal morpheme and which is the nominal root. Apart from such minor inconsistencies, I have tried to make the glossing system used here as consistent as possible. 4

Draft for comments (2015) - Do not cite without author’s permission

139 ‘Rutafa spilt the milk.’ 140 b. amate g-aa-yat-ik-a. 141 milk SM-PST-spill-MID-FV 142 ‘The milk (got) spilt.’

143 An anticausative is a detransitivized construction in which the object of the corresponding 144 causative construction is encoded as the , and the original subject is omitted. Verbs 145 participating in the causative/anticausative alternation constitute a specific subgroup of material 146 (Halliday & Matthiessen 2014: 213) or physical action verbs. Physical action verbs constitute 147 the prototypical case of transitive verbs (Haspelmath 2011: 545, Næss 2007: 15, Shibatani 148 2006: 259, Tsunoda 1985: 387), i.e. with a prototypical agent volitionally and consciously 149 exerting force on a completely affected, prototypical patient resulting in a change of state of the 150 latter (Hopper & Thompson 1980). It is well-known that the anticausative derivation is 151 restricted to certain types of verbs (Keyser & Roeper 1984, Kulikov 1998, Levin & Rappaport 152 Hovav 1994). What remains a topic of debate is which semantic feature(s) these verbs have in 153 common. Haspelmath (1993: 93) proposes ‘the absence of agent-oriented meaning 154 components’ as the central concept underlying the anticausative verbs, because “the inchoative 155 member implies the absence of an agent, [and therefore] it cannot contain agent-oriented 156 semantic elements.” Anticausative constructions thus are often said to express situations which 157 occur spontaneously, i.e. without the participation of a causing agent. Song (1995: 213) 158 provides a critical account of Haspelmath’s analysis, stating that the feature ‘agent-oriented 159 meaning components’ is a rather vague concept seeing how Haspelmath does not specifically 160 define what these components should be. In line with this, Kulikov (1998: 143) argues that 161 posing a universal semantic feature for the class of anticausative verbs is untenable due to the 162 fact that specific criteria are probably language-specific. He does not, however, disregard 163 spontaneity but follows Song in that it is too vague a notion to be a useful concept. Kulikov 164 (1998: 150) proposes a concrete sub-component of the notion spontaneity that might be at the 165 heart of the group of anticausative verbs, namely entropy increase. Entropy is defined as a 166 general decrease in energy. Kulikov (1998: 147) has found two semantically coherent groups 167 of verbs in Vedic that can be subsumed under the notion of entropy increase, viz. those of 168 destruction and those of destructuring, namely “spontaneous processes which result in 169 destroying some natural or artificial system or organism” (ibid.).4 He (1998: 151-2) does remain 170 rather precautious in posing that the concept of entropy increase is the central feature of the 171 anticausative verbs cross-linguistically. However, it is probably no coincidence, then, that in 172 his general description of the Bantu middle suffix, Guthrie (1967: 92) writes that “it is found 173 […] with a small class of radicals which is at the same time a semantic class (« verba 174 destruendi »)” (emphasis SD), and later Schadeberg (2006: 75) states that this verbal suffix “is 175 best represented with […] verbs of destruction.” 176 I want to argue here that spontaneity, which has often been used by linguists in the 177 description of the semantics of anticausative verbs (Davidse 1992: 109, Haspelmath 1987: 15, 178 1993: 90, Levin & Rappaport Hovav 1994: 61), is a misleading concept and should be discarded

4 Kulikov (1998: 152) makes a valid distinction between entropy increase on the one hand and destruction on the other, in that the latter can be a process executed to obtain a goal, i.e. the creation of a new structure. He exemplifies this by the idea of slicing vegetables (destruction) in order to make a salad (creation of new structure). 5

Draft for comments (2015) - Do not cite without author’s permission

179 altogether in the discussion on the semantic features underlying the anticausative group of 180 verbs. A combination of semantic and syntactic properties of anticausative constructions have 181 led linguists to argue mistakenly that situations which are inherently causative (viz. events of 182 physical action, of which anticausatives constitute a subgroup) can be ‘self-instigated’ or 183 ‘spontaneous’. Following Levin & Rappaport Hovav (1994: 52), verbs participating in the 184 causative/anticausative alternation have a dyadic predicate structure. They typically comprise 185 two subevents:

186 break-transitive: [x CAUSE [y become BROKEN] 187 (Levin & Rappaport Hovav 1994: 52)

188 The anticausative construction only reports on the second subevent, and therefore allows the 189 speaker to foreground the resulting situation of a causal chain of events and background the 190 first, causing event by leaving it unmentioned. There are a number of reasons for a speaker to 191 make this decision. One can wish to de-agentivize certain situations in order to conceptually 192 decrease or conceal the influence of a causing agent (which is why anticausative constructions 193 are frequently used in bureaucratic and political discourse; see Goatly 1996). The agent might 194 also be unknown to the speaker, or not relevant in the discourse, e.g. when one wants to make 195 a general statement, as in the Luganda example in (4).

196 (4) LUGANDA (JE15) (LCK) 5 197 […] kubanga n'empiso bw'emenyeka tebagizza mu ssasa […]. 198 kubanga ne e-N-yiso bwe e-meny-ik-a 199 because even AUG9-CL9-needle when SM9-break-MID-FV 200 te-ba-gi-dd-iz-a mu li-sasa […]. 201 NEG-SM2-OM9-return-CAUS-FV LOC18 CL5-workshop 202 ‘[…] because even when a needle breaks it is not taken back to the workshop […].’

203 Perhaps the most important reason for backgrounding the first event/agent with regard to the 204 notion of ‘spontaneity’ is that the causing event does not necessarily have to involve an animate 205 entity. The causing event of a change-of-state event in an anticausative construction can involve 206 a human/animate referent (as in (4) above, where it is possible that a child takes a needle and 207 breaks it), but also impalpable micro-organisms, as in (5) where the destruction of the boat’s 208 heartwood can be caused by fungi. Closely related to impalpable physical entities are (natural) 209 forces such as the wind (e.g. the door closes), heat (e.g. the vegetables are cooking), or electric 210 and mechanical systems (e.g. the motor is running).

211 (5) LUGANDA (JE15) (LCK) 212 […] kubanga omutima bwe gusigalamu eryato livunda mangu ne lyonooneka. 213 kubanga o-mu-tima bwe gu-sigal-a-mu e-li-ato 214 because AUG3-CL3-heart if SM3-remain-FV-LOC18 AUG5-CL5-boat 215 li-vund-a mangu ne li-yonoon-ik-a. 216 SM5-rot-FV quickly and SM5-destroy-MID-FV

5 LCK stands for ‘Luganda Corpus of Deo Kawalya’ 6

Draft for comments (2015) - Do not cite without author’s permission

217 ‘[…] because if the heartwood remains there the boat rots quickly and is destroyed.’

218 It need not even be one single entity, but can also be a series of events that leads to a resulting 219 change-of-state of an affected entity. For example, when stating the chair broke, it does not 220 necessarily have to be so that a single, concrete physical entity volitionally and purposely acted 221 upon the chair causing it to break, but rather that by repeatedly sitting on it for a long period of 222 time the material wore out and eventually broke. In this case the speaker, wanting to assert that 223 the chair is broken, cannot refer to any specific causing entity. However, what remains 224 impossible is that an entity in perfect condition spontaneously changes its state without any 225 causing event(s) having physically affected it. It is therefore misleading to state that an 226 anticausative construction presents an event as occurring spontaneously: physical action verbs 227 such as anticausatives typically imply causing events bringing about a resulting change-of-state 228 in an affected participant. The absence of such causing events in anticausative constructions is 229 not a way of conceptualizing the event as self-initiated, but is rather the result of pragmatic 230 motivations of the speaker. Moreover, animate entities are higher on the hierarchy, 231 and furthermore higher in referentiality and topicality than inanimates, (natural) forces and 232 multiple causing events. Inanimates are also less prototypical to function as agents (de Swart et 233 al. 2008: 134), which is why they are less likely to be encoded as syntactic (subject) arguments 234 in a construction.

235 4. Agentless passive constructions 236 The middle suffix also marks de-agentivized verbs with verbs that do not belong to the semantic 237 group of anticausatives. As such, it can be suffixed onto physical action verbs (6) and verbs of 238 visual (7) and auditory perception (8). In perception constructions it is the sensed or perceived 239 entity, i.e. the stimulus or phenomenon, that is encoded as the subject and the senser/perceiver 240 is omitted.

241 (6) LUGANDA (JE15) (LCK) 242 Abalimi baffe mwenna mufube bufubi okulima obutale bujja kufunika. 243 a-ba-limi ba-ffe mu-enna mu-fub-e 244 AUG2-CL2-farmer PP2-POSS1PL PP18-all SM2PL-work.hard-SBJ 245 bu-fubi o-ku-lim-a o-bu-tale bu-jj-a 246 CL14-exertion AUG15-CL15-cultivate-FV AUG14-CL14-market SM14-come-FV 247 ku-fun-ik-a. 248 CL15-get-MID-FV 249 ‘(To) all our farmers, just strive to cultivate, the market will be found (lit. be got).’

250 (7) LUGANDA (JE15) (LCK) 251 Teyaddayo kulabikako. 252 te-a-a-dd-a-yo ku-lab-ik-a-ko. 253 NEG-SM1-REM-come.back-FV-CLT1 CL15-see-MID-FV-LOC17 254 ‘S/he was not seen again.’

255 (8) LUGANDA (JE15) (LCK)

7

Draft for comments (2015) - Do not cite without author’s permission

256 Nakiru: (Eddoboozi lye liwulikika ku muzindaalo ng’akyuka erudda n’erudda.) 257 Nakiru: ( e-li-loboozi li-e li-wulikik-a ku 258 Nakiru: AUG5-CL5-voice PP5-POSS1 SM5-be.heard:MID:MID-FV LOC17 259 mu-zindaalo nga a-kyuk-a e-lu-dda ne e-lu-dda.) 260 CL3-speaker CNJ SM1-turn.around-FV AUG11-CL11-side and AUG11-CL11-side 261 ‘Nakiru: (her voice is heard on the speaker turning from one side to another.)’

262 Lexical examples of agentless passive verbs in Bantu languages other than Luganda are 263 provided in Table 1 below.

264 Table 1. Non-anticausative verbs with middle marker Language SVB6 EVB Source

LONDO (A11) -rɔ̀ nd- ‘sew’ -rɔ̀ ndɛ- ‘be sewn’ Kuperus (1985: 198) -sórw- ‘shave’ -sórwɛ- ‘be shaved’

FANG (A75) -ádzə̀ b- ‘bury’ -ádzə̀ bə̀ - ‘be buried’ Ondo-Mebiame (2007: 122)

BÉO (C45A) -kamo- ‘kill’ -kamo(l)e- ‘be killed’ Gérard (1924: 74)

BEYA (D25) -gul- ‘sell’ -gulik- ‘be sold’ Botne (2006: 437)

MITUKU (D13) — -bundik- ‘be taken’ Stappers (1973: 13)

KIKUYU (E51) -haat- ‘sweep’ -haatik- ‘be swept’ Barlow (1951: 123)

-ri- ‘eat’ -riik- ‘be eaten’

KAMBA (E55) -iw- ‘hear, feel’ -iwik- ‘be heard, felt’ Whiteley & Muli (1962: 106)

SHAMBALA (G23) -šund- ‘punish’ -šundik- ‘be Kotz (1909: 32) punished’

RWANDA (JD61) -vúg- ‘say’ -vúgik- ‘be said’ Coupez (1985: 19)

RUNDI (JD62) -imb- ‘dig’ -imbik- ‘be dug’ Ménard (1908: 195)

-tem- ‘cut’ -temek- ‘be cut’

GUSII (JE42) -ßuat- ‘hold, -ßuatek- ‘be held’ Cammenga (2002: seize’ 381)

6 I am not using the term SVB, i.e. ‘simple base’, in a strict sense, viz. a verb root without any suffix, but rather use it to refer to any verb base (simple or extended) which is not derived by the middle marker. The (simple or extended) verb base under the heading SVB is subsequently marked in the row headed with EVB, i.e. ‘extended verb base’, by the middle suffix. 8

Draft for comments (2015) - Do not cite without author’s permission

CHOKWE (K11) -hond- ‘wring’ -hondek- ‘be wrung’ Van Den Eynde (1960: 38)

KILUBA (L33) -fúl- ‘forge’ -fúlik- ‘be forged’ Vandermeiren (1912: 153)

BEMBA (M42) -seb- ‘sift’ -sebek- ‘be sieved’ Noël (1935: 94)

SENGA (N41) -luw- ‘forget’ -luwik- ‘be forgotten’ Ranger (1928: 187)

YAO (P21) -oko- ‘save, -okok- ‘be saved, Whiteley (1966: 39) rescue’ rescued’

MABIHA (P25) -put- ‘wipe’ -putik- ‘get wiped’ Harries (1940: 131)

TSWANA (S31) -rat- ‘love’ -ratêg- ‘become Cole (1955: 196) loved’

-batlêg- ‘be sought, -batl- ‘seek’ be wanted’

-dirêg- ‘become -dir- ‘do, make’ done’

SESOTHO (S33) -tlam- ‘bind’ -tlameh- ‘be bound’ Van Eeden (1941: 199) -seny- ‘destroy’ -senyeh- ‘be destroyed, be rotten’

265 Marking an agentless passive construction, -ik- can also suffix onto ditransitive verbs, as in (9). 266 The Luganda verb -w- ‘give’, which has been shown to be the most prototypical ditransitive 267 verb (Goldberg 1992: 54, Kitillä 2006), takes the middle marker resulting in the derived two- 268 participant verb base -week- ‘be given’.

269 (9) LUGANDA (JE15) (LCK) 270 Bw'aba nga mu kirowoozo kye alaba nga tekiweeka muntu oba nga kiyitirivu oba nga 271 tekigoberera mateeka, […]. 272 bwe a-ba-a nga mu ki-rowoozo ki-e a-lab-a 273 if AUG2-PP2-CON CNJ LOC18 CL7-opinion PP7-POSS1 SM1-think-FV 274 nga te-ki-w-ik-a mu-ntu oba nga ki-yitirivu 275 that NEG-SM7-give-MID-FV CL1-person or CNJ CL7-excessive 276 oba nga te-ki-goberer-a ma-teeka, […]. 277 or that NEG-SM7-follow:APPL:APPL-FV CL6-law 278 ‘If in his opinion he thinks that it is inappropriate that it is given to someone or it is too 279 much or that it doesn't follow the law, […].’

9

Draft for comments (2015) - Do not cite without author’s permission

280 The example shows that the theme participant, which is not overtly expressed but anaphorically 281 marked on the verb tekiweeka by the SM of class 7 (referring to kintu ‘thing’) (Bresnan & 282 Mchombo 1987) is in subject position when a is derived by the middle marker. 283 The recipient participant muntu immediately follows the verb, which is often the primary object 284 position in Bantu languages (Bearth 2006: 124). The agent is omitted from the clause, although 285 it is impossible to imagine that semantically something is given to someone without there being 286 a giver. 287 In the light of the middle suffix’s function as a marker of agentless passive constructions 288 with perception verbs, a revision is needed of Schadeberg’s (2006: 75) definition that “verbs 289 with [the middle marker] indicate that the subject is potentially or factually affected by the 290 action expressed by the verb” (emphasis S.D.). The perceived entity is typically not affected by 291 the perceiver, and thus affectedness does not constitute a central element in a semantic 292 definition of the middle marker but is rather dependent on the verb type it combines with.

293 5. Comparing the anticausative/agentless passive with the productive passive 294 construction 295 The constructions marked by the middle suffix discussed in the previous two sections in a way 296 resemble the productive passive construction. The most obvious way in which the two differ is 297 that the productive passive is typically marked by a different suffix, reconstructed for Proto- 298 Bantu as *-ʊ-/-ibu-. What they have in common is that both constructions are de-agentivized 299 clauses in which the patient-like participant is in subject position and the agent-like participant 300 is demoted (Kulikov 2013, Shibatani 1985). The passive and anticausative constructions can, 301 from a semantic viewpoint, be relatively easily distinguished in that the former is usually not 302 restricted to any type of verb class, whereas the anticausative construction is limited to verbs 303 participating in the causative/anticausative alternation. The most problematic case is thus the 304 semantic and pragmatic difference between an agentless passive construction marked by the 305 middle suffix and a prototypical passive construction marked by the productive passive suffix.

306 5.1. Grammatical features regarding the agent-like participant

307 5.1.1. Oblique agentive phrases 308 The most well-known feature distinguishing anticausatives from passives is the ability of the 309 latter to take an oblique agentive phrase, whereas the former generally does not (10) (Alexiadou 310 et al. 2005: 176, Härtl 2003: 892, Haspelmath 1987: 7).

311 (10) NDEBELE (S44) (Khumalo 2009: 166, 168) 312 a. isi-valo sa-val-ek-a (*ngu Thabo). 313 CL7-door SM7-shut-MID-FV by Thabo 314 ‘The door closes (*by Thabo).’ 315 b. isi-valo sa-val-w-a (ngu Thabo). 316 CL7-door SM7-shut-PASS-FV by Thabo 317 ‘The door was closed (by Thabo).’

10

Draft for comments (2015) - Do not cite without author’s permission

318 In the anticausative construction (10a) the presence of the oblique agentive phrase ngu Thabo 319 renders the clause ungrammatical, and is therefore impossible. The passive construction in 320 (10b), on the other hand, has no such restrictions. This feature can be extended to agentless 321 passive constructions with physical action verbs marked by the middle suffix as well, as in (11).

322 (11) CHEWA (N31b) (Dubinsky & Simango 1996: 751) 323 a. mbale zi-na-tsuk-ik-a (*ndi Naphiri). 324 plates SM-PST-wash-MID-FV by Naphiri 325 ‘The plates were washed (*by Naphiri).’ 326 b. mbale zi-na-tsuk-idw-a ( ndi Naphiri). 327 plates SM-PST-wash-PASS-FV by Naphiri 328 ‘The plates were washed (by Naphiri).’

329 Interestingly, this feature seems to be restricted to middle constructions with physical action 330 verbs only. As shown in (12) and (13), the senser in perception constructions can be 331 reintroduced via an oblique phrase.

332 (12) NDONGA (R22) (Fivaz 1986: 111 in Fleisch 2005: 123) 333 ókinó ndjoká o-y-a-tál-ik-á ká-á-ntu a-yéhe. 334 CL9:film DEM9 AFF-SM9-PRF-watch-MID-FV INSTR-CL2-person PP2-many 335 ‘This film is seen by many people.’

336 (13) CHEWA (N31b) (Mchombo 1993: 6) 337 nkhání yá Katenje y-a-mv-ik-á kwá anthu 338 CL9:story PP9:CON Katenje SM9-PRF-hear-MID-FV PP17:to CL2:people 339 ônse. 340 PP2:all 341 ‘The story of Katenje has been heard by all the people.’

342 It is important to note that the preposition kwá preceding the senser in (13) is different from that 343 of an agentive phrase in a Chichewa passive construction, which is ndi as can be seen in (11b). 344 The preposition kwá is a combination of the noun class prefix 17, which has a locative meaning 345 (in Bantu more generally; Katamba 2006: 116), and the associative morpheme -a. A literal 346 interpretation would be that the sound travelled to and has arrived at (locative), i.e. has been 347 heard by, the listeners. Thus, while ndi is reserved for oblique phrases introducing agent 348 participants (among others, see (19) below where it is used to introduce instrumental NPs), kwá 349 seems to function as the preposition for sensers in Chichewa. 350 In Swahili, constructions with cognitive verbs and an overt senser such as those with verbs 351 of perception in (12) for Ndonga and (13) for Chichewa are impossible. In (14) the cognitive 352 verb -tambul- ‘know’ is marked by the middle suffix, but the construction is ungrammatical 353 due to the presence of the oblique phrase with the senser mjini ‘town(speople)’.

354 (14) SWAHILI (G42) (Seidl & Dimitriadis 2002: 246) 355 * a-na-tambul-ik-a na mjini. 356 SM-PRS-know-MID-FV by town 357 ‘She is well-known by the town (the townspeople).’

11

Draft for comments (2015) - Do not cite without author’s permission

358 However, cognitive constructions that take mental sources as the initiator of their basic 359 construal (sometimes referred to as cognitive constructions of the please-type, as opposed to 360 those of the like-type as ‘know’ in (14); Halliday & Matthiessen 2014: 247-48) do seem to allow 361 the demoted mental source in an oblique phrase, as illustrated in (15).

362 (15) SWAHILI (G42) (Schadeberg 2004: 11, 12) 363 a. -patika na mauti ‘être surpris par le mort’ [sic.] < -patika ‘be available’ 364 b. -patika na shida ‘éprouver des difficultés’

365 The middle verb -patik- ‘be available’ receives an idiomatic meaning through the addition of 366 some specific participants such as mauti ‘death’ or shida ‘difficulties’. This resulting non- 367 compositional or idiomatic meaning expresses a cognitive event with the mental source as the 368 initiator, viz. ‘be surprised (by death)’ and ‘feel/have (difficulties)’. Although no clausal 369 examples are provided, the overt presence of the mental source in the lexical examples seems 370 to indicate that they can be expressed in an oblique phrase with na, which is also used in Swahili 371 to mark the oblique agentive phrase in passive clauses. 372 Moreover, when the verb in Swahili cognitive constructions such as (14) is additionally 373 extended with the productive reciprocal marker -an-, the collocation of an oblique senser phrase 374 is allowed. This is shown in (16) with the verb -tambul- ‘know’ and in (17) with the perception 375 verb -on- ‘see’.

376 (16) SWAHILI (G42) (Seidl & Dimitriadis 2002: 246) 377 a-na-tambul-ik-an-a na mjini. 378 SM-PRS-know-MID-REC-FV by town 379 ‘She is well-known by the town (the townspeople).’

380 (17) SWAHILI (G42) (Seidl & Dimitriadis 2002: 246) 381 hu-on-ek-an-i na watu siku hizi. 382 NEG.SM-see-MID-REC-FV.NEG by people days these 383 ‘You haven’t been seen around by people these days.’

384 The examples in (16) and (17) have been rejected by G. Mertens, a near-native speaker and 385 teaching assistant of Swahili at Ghent University (personal communication). Mertens stated 386 that he has never encountered a clause with a middle-reciprocal verb followed by a prepositional 387 na-phrase. This indicates that the constructions are not acceptable for all Swahili speakers, and 388 perhaps are only grammatical in some dialectal varieties (although these examples should be 389 evaluated by more native speakers in order to make any decisive statements about them). Seidl 390 & Dimitriadis (2002: 239, footnote) elaborate that two of their informants speak the Zanzibari 391 variety, and one other has been consulted for the standard variety. Concerning the constructions 392 above, they do not indicate which informant has produced the examples and whether they have 393 been checked by the other two informants. Nevertheless, Seidl & Dimitriadis make a 394 generative-oriented analysis of the neuter based on the assumption that the examples above are 395 correct. They argue that stativisation (which is the term they use for the function of the middle 396 marker) in Swahili does not involve an external reduction operation (a concept which they do 397 not elaborate), but that:

12

Draft for comments (2015) - Do not cite without author’s permission

398 “the lexical operation triggered by the Swahili [middle] morpheme is a variant of saturation 399 […] which [they] refer to as arbitrarization. Like ordinary saturation, arbitrarization binds 400 the suppressed argument by an existential quantifier. But the introduced variable is marked 401 as “arbitrary” […] meaning it cannot be given a specific denotation.” (Seidl & Dimitriadis 402 2002: 260; italics in original)

403 Presuming that the examples in (16) and (17) are acceptable and grammatical, my analysis is 404 different from that of Seidl & Dimitriadis. First of all, (14) illustrates that a senser cannot be 405 reintroduced in a cognitive construction with an agentless passive argument structure (in 406 contrast to Chichewa, as shown in (13)). However, when a verb marked by the middle suffix is 407 extended by the reciprocal -an-, the senser can be reintroduced in an oblique phrase (16). The 408 sociative/reciprocal suffix is highly polysemous in Bantu, having a wide number of functions 409 and expressing various notions such as reciprocity (either productive or lexicalized), collective 410 situations, chaining situations, repetitiveness, intensive/extensive, comitative/instrumental, and 411 anticausativity (Bostoen et al. forthcoming). The various meanings of the sociative/reciprocal 412 suffix can be described more generally as denoting a process involving multiple participants 413 and/or events. In the literature various terms have been used and proposed to describe 414 polysemous markers involving reciprocity. Lichtenberk (2000), in his article on the reciprocal 415 marker in some Oceanic languages, subsumes the different meanings under the more general 416 semantic concept of plurality of relations: “there is plurality of relations in an overall situation 417 […] if what can be considered to be basically one and the same relation holds more than once 418 either between one or more participants and the event/state they are involved in, or between the 419 relevant entities” (Lichtenberk 2000: 34). Creissels & Nouguier-Voisin (2008: 291) label this 420 broader category ‘co-participation’, viz. “constructions that imply a plurality of participants in 421 the event they refer to, without assigning them distinct roles.” Maslova (2007) refers to this as 422 a polyadic event type or structure, for which she provides the following definition:

423 “Some type of participation in an event constitutes a polyadic participant role if it must 424 be shared by minimally two separate participants. An event structure counts as polyadic if 425 it involves such a role.” (Maslova 2007: 336; italics and underlining in original)

426 When used in Swahili cognitive constructions marked by the middle -ik- (as in (16) and (17) 427 above), the sociative suffix -an- no longer expresses reciprocity, but does not lose its function 428 as marker of plurality of relations/co-participation/polyadic participation. Plurality of relations 429 can, furthermore, be subdivided into the notions of plurality of participants and plurality of 430 events. In combination with the middle suffix, which normally grammatically deletes the 431 participant in subject position of the basic construction, the semantic notion of plurality of 432 relations, and more specifically that of plurality of participants expressed by the sociative 433 marker allows the oblique expression of the demoted senser participant. This ‘addition or 434 preservation of an implicit participant’-function of the sociative/reciprocal morpheme, invoked 435 by the plurality of relations semantics, is not uncommon as it has been shown that it is also 436 involved in Bantu antipassive constructions (Bostoen et al. forthcoming, Dom et al. 437 forthcoming), where the sociative/reciprocal allows the grammatical deletion of the patient, 438 which nevertheless remains implied at the semantic level.

13

Draft for comments (2015) - Do not cite without author’s permission

439 The data presented in this section point toward a difference in the grammatical behavior of 440 agentless passive -ik- constructions in Bantu with regard to the expression of the initiator, which 441 seems to be motivated by the opposition between physical action verbs and non-physical actions 442 verbs. The prototypical agent of physical action events (and thus also of events expressed by 443 anticausative verbs) is typically inexpressible in constructions marked with the middle suffix. 444 However, initiating participants of cognitive events (sensers, perceivers, or mental sources) are 445 not prototypical agents. In a cognitive event, the initiating senser does not actively exert a force 446 affecting a patient with a resulting change-of-state of the latter (see also Hopper & Thompson 447 1980: 269-270, who argue that “verba sentiendi” are “less active verbs”). Thus, a senser is quite 448 distinct from an agent of a physical action verb, which motivates the grammaticality of the 449 former being overtly expressed in a cognitive middle construction. It is important to note, 450 however, that the oblique NPs denoting sensers are always generic (everyone, (towns)people). 451 The only exceptions that I have found are the Swahili examples in (15) denoting cognitive 452 events with the mental source as initiator.

453 5.1.2. Instrumental phrases 454 Instruments generally imply the presence of an agentive participant controlling them in order 455 to execute a certain action. Therefore, instrumental phrases are not allowed in anticausative and 456 agentless passive constructions, as shown in (18) and (19a), in contrast to prototypical passive 457 constructions in which instrumental NPs are unproblematic (19b).

458 (18) SWAHILI (G42) (Seidl & Dimitriadis 2002: 257) 459 * chungu ki-li-pig-ik-a kwa nyundo. 460 cooking.pot SM-PST-break-MID-FV with hammer 461 ‘The cooking pot was hit with a hammer.’

462 (19) CHEWA (N31b) (Dubinsky & Simango 1996: 752) 463 a.*kalata i-na-lemb-ek-a ndi pensulo. 464 letter SM-PST-write-MID-FV with pencil 465 b. kalata i-na-lemb-edw-a ndi pensulo. 466 letter SM-PST-write-PASS-FV with pencil 467 ‘The letter was written with a pencil.’

468 The following Swahili example seems to contradict the ungrammaticality of instruments in 469 middle marked constructions.

470 (20) SWAHILI (G42) (Seidl & Dimitriadis 2002: 257) 471 a-li-chom-ek-a kwa maneno haya. 472 SM-PST-stab-MID-FV with words these 473 ‘He was stabbed by these words.’

474 However, when taking a closer look at the semantics of the instrument adjunct, it becomes clear 475 why it is possibly allowed in a middle marked construction in this case. The instrumental 476 adjuncts in (18) and (19) refer to material objects (a hammer or pencil) which are generally 477 used by a forceful, volitional, agentive participant. Furthermore, there has to be direct contact 478 between the instrument and the affected patient (in case of the hammer and the cooking pot in

14

Draft for comments (2015) - Do not cite without author’s permission

479 (18)), or the instrument has a central role in the event, i.e. the pencil is indispensable in order 480 to create the letter in (19). The instrument maneno haya ‘these words’ in (20), on the other 481 hand, is not a material entity, and furthermore only experienced by the patient. There is no 482 direct contact or causation and the patient is not materially affected (or damaged) by the 483 instrument. The mental effect actually originates from within the patient participant. The 484 speaker of the words might not have had the intention of hurting the person in question, thus 485 possibly lacking in forcefulness, volitionality and agentivity. Thus, the low amount of 486 prototypical agentivity normally associated with the instrumental NP possibly allows for its 487 presence in (20).

488 5.1.3. Agent-oriented adverbs and purpose clauses 489 A third feature generally mentioned in discussions on the difference between passives and 490 anticausatives is the possibility to combine with agent-oriented adverbs. The example in (21) 491 illustrates that constructions marked by the middle suffix cannot collocate with agent-oriented 492 adverbs, whereas in productive passive constructions they do not affect the grammaticality of 493 the clause.

494 (21) NDEBELE (S44) (Khumalo 2009: 168) 495 a. *isi-valo sa-val-ek-a ngabomo. 496 CL7-door SM7-shut-MID-FV deliberately 497 *‘The door closed deliberately.’ 498 b. isi-valo sa-val-w-a ngabomo. 499 CL7-door SM7-shut-PASS-FV deliberately 500 ‘The door was closed deliberately.’

501 Purpose clauses imply that the event of the main clause has been brought about by a 502 volitional agent for a certain reason, which is elaborated in the subordinate purpose clause. As 503 shown in (22), the middle marked construction cannot combine with a purpose clause, whereas 504 this is unproblematic in a passive clause. No translation is provided in the original source for 505 (22a), due to the fact that the construction is ungrammatical.

506 (22) CHEWA (N31b) (Mchombo 1993: 17) 507 a.*mphâtso zi-na-sókónez-ek-a kutí pa-sa-khál-é ku-konděr-a. 508 CL10:gifts SM10-PST-mix-MID-FV that SM16-NEG-be-SBJ CL15-favor-FV 509 b. mphâtso zi-na-sókónez-ědw-a kutí pa-sa-khál-é ku-konděr-a. 510 CL10:gifts SM10-PST-mix-PASS-FV that SM16-NEG-be-SBJ CL15-favor-FV 511 ‘The gifts were mixed up so that there should be no favouritism.’

512 5.2. Aspectual difference: resultative meaning of the middle marker 513 Although this is generally less elaborately discussed in the literature on anticausativity, 514 anticausative constructions typically imply (objective) resultative aspect (Nedjalkov & 515 Jaxontov 1988: 9). Passive constructions, on the other hand, rather have a perfective aspectual 516 meaning. This is illustrated in the following example (23). The negation makes the aspectual 517 distinction even more clear.

518 (23) CHEWA (N31b) (Dubinsky & Simango 1996: 750)

15

Draft for comments (2015) - Do not cite without author’s permission

519 a. nyemba si-zi-na-phik-ik-e. 520 beans NEG-SM-PST-cook-MID-FV 521 ‘The beans were not cooked.’ 522 b. nyemba si-zi-na-phik-idu-e. 523 beans NEG-SM-PST-cook-PASS-FV 524 ‘The beans were not cooked (at all).’

525 The anticausative clause in (23a) can be asserted in a situation where the beans are in a cooking 526 pot on a fire, but only half cooked. On the other hand, the passive construction in (23b) can 527 only be uttered when the beans have been left untouched, i.e. they have never been acted upon 528 for cooking. 529 As I have argued in section 3, the event structure of situations denoted by anticausative verbs 530 is a causal chain of events that can be represented as Event1 RESULT Event2. For pragmatic 531 reasons, the speaker chooses a syntactic configuration that only reports on the resulting Event2 532 by means of the anticausative construction. The definition of ‘resultative’ given by Nedjalkov 533 & Jaxontov (1988: 6) overlaps for a great part with the event view conceptualized by an 534 anticausative construction:

535 “the term resultative is applied to those verb forms that express a state implying a previous 536 event. The difference between the stative and the resultative is as follows: the stative 537 expresses a state of a thing without any implication of its origin, while the resultative 538 expresses both a state and the preceding action it has resulted from.”

539 The concept of the ‘resultative’ given by Nedjalkov & Jaxontov is, however, too strict in that 540 their definition narrows resultatives down to the expression of states only. This, however, 541 should not always be the case in resultative meaning of anticausative constructions. The stative 542 meaning of a resultative is the outcome of an interaction between tense-aspect categories and 543 punctuality of events. Punctual events such as break are less often talked about in the present 544 progressive due to what has been coined in Cognitive Grammar as the ‘epistemic problem’ 545 (Langacker 2001: 263), i.e. “the conceptual difficulty of observing and at the same time 546 linguistically reporting, or even merely identifying, punctual or very short (i.e., typically 547 dynamic) events in the present” (Brisard & Meeuwis 2011: 32, italics in original). Durative 548 events are not restricted by the epistemic problem, and anticausative verbs such as cook or open 549 can combine with the present progressive, e.g. The beans are cooking or The garage door is 550 opening. These constructions do not express states, but report on the resulting Event2 of a causal 551 chain of events in which the first event is someone acting on the beans and the garage door. It 552 is only when combined with past tense or the perfect that anticausative constructions with 553 durative verbs express states, e.g. The door is/was closed or The beans are/have been cooked. 554 Thus, the resultative meaning of anticausative constructions should be defined in a broader way 555 as expressing the resulting event or state of a causal chain of events, implying a previous event.

16

Draft for comments (2015) - Do not cite without author’s permission

556 6. Potential constructions 557 The potential construction expresses, in very general terms, the attribution of a quality, denoted 558 by the verb, to the subject of the clause. The potential is often marked by the same means as the 559 passive, e.g. in Hausa and Modern Greek (Haspelmath 1987: 7). However, in Bantu languages 560 potential constructions are marked by the middle suffix, as illustrated in the Swahili example 561 in (24).

562 (24) SWAHILI (G42) (Driever 1976 in Seidl & Dimitriadis 2002: 255) 563 barua hii i-me-som-ek-a. 564 letter DEM SM-PRF-read-MID-FV 565 ‘This letter is readable.’

566 Potentials are intransitive constructions in that they typically involve one, patient-like 567 participant to which is attributed a quality, such as ‘being readable’ in (24). Interestingly, one- 568 place predicates, i.e. basic intransitive verbs, can also figure in potential constructions. In that 569 case, it is an adjunct that functions as the subject of the potential clause. The Swahili verb -lal- 570 ‘sleep’ in (25) and (26) is an , taking the locative adjuncts godoro ‘mattress’ 571 (25) or kitanda ‘bed’ (26) as its subjects in the potential constructions.

572 (25) SWAHILI (G42) (Seidl & Dimitriadis 2002: 254) 573 godoro li-na-lal-ik-a. 574 mattress SM-PRS-sleep-MID-FV 575 ‘This mattress can be slept on.’

576 (26) SWAHILI (G42) (Seidl & Dimitriadis 2002: 254) 577 kitanda ki-na-lal-ik-a vizuri. 578 bed SM-PRS-sleep-MID-FV well 579 ‘This bed sleeps well.’

580 The fact that locative adjuncts can function as subjects in potential constructions marked by the 581 middle suffix again shows that affectedness is not a central property of the middle marker’s 582 semantics. Although sleeping is (often) done on beds and mattresses, the act of sleeping does 583 not have as its purpose to affect these locational entities. Thus, whereas the anticausative and 584 agentless passive functions of the middle suffix require the verb to have a basic two-participant 585 configuration, its potential function differs in that it can combine with basic one-participant 586 predicates. 587 It has been noted for Ndebele potential constructions that they can combine with 588 instrumental NPs. As shown in (27), the instrumental phrase ngengquamu ‘with a knife’ does 589 not render the potential construction ungrammatical, as would be the case in other Bantu 590 languages for anticausative/agentless passive constructions.7

591 (27) NDEBELE (S44) (Khumalo 2009: 169)

7 Khumalo (2009: 169) does not discuss the grammaticality of anticausative/agentless passive constructions with instrumental phrases. Therefore, I can only compare the potential construction with the instrumental phrase with anticausative/agentless passive constructions in other Bantu languages. 17

Draft for comments (2015) - Do not cite without author’s permission

592 i-N-kukhu ya-qum-ek-a ngengqamu.

593 AUG9-CL9-chicken SM9-cut-MID-FV with.knife 594 ‘The chicken was cutable with a knife.’

595 This is possibly a language-specific feature of potential constructions, given that the similar 596 Swahili example in (28) is ungrammatical. Although kijungu hiki ‘this little pot’ is a recipient 597 (i.e., not the thematic role but an entity in which something is placed) rather than a prototypical 598 instrument, the two are closely related in that the pot is a necessary tool used by the cook to 599 prepare the chicken.

600 (28) SWAHILI (G42) (Seidl & Dimitriadis 2002: 257) 601 * kuku ki-li-pik-ik-a kwa kijungu hiki. 602 chicken SM-PST-cook-MID-FV with little.pot DEM 603 ‘The chicken was cookable with this little pot.’

604 Note also that the verb in (28) is an anticausative verb, in contrast to the one in (27). Further 605 research should investigate whether an equivalent of the Ndebele construction is licit in Swahili, 606 and whether the verb class (anticausative or not) has any effect on the grammaticality of a 607 potential construction with an instrumental phrase. 608 In the following sections I will analyze the potential semantics in more detail. In order to 609 adequately describe the potential’s meaning and function, a combination of three linguistic 610 concepts is needed: (i) force dynamics, (ii) modality, and (iii) subjectivity.

611 6.1. Force dynamics 612 Force dynamics is a notion elaborated by Leonard Talmy in his work on cognitive semantics 613 (1985, 1988, 2000: Chapter 7). It is a semantic category expressing “how entities interact with 614 respect to force” (Talmy 2000: 409), such as exertion of and resistance to force, overcoming of 615 resistance, blockage of force exertion and removal of blockage. The interaction of force 616 revolves around two central entities, the agonist or focal force entity, and antagonist or opposing 617 force entity (Talmy 2000: 413). In a force relation, the agonist can have a tendency toward rest 618 (29a,b) or action (29c,d). The force of the opposing antagonistic entity can be bigger than the 619 force exerted by the agonist, thus overcoming (29a), hindering (29c) or blocking (29d) its 620 tendency toward rest or action, or it can be smaller, in which case the agonist’s tendency is 621 manifested (29b,c).

622 (29) a. The ball kept rolling because of the wind blowing on it. 623 b. The shed kept standing despite the gale wind blowing against it. 624 c. The ball kept rolling despite the stiff grass. 625 d. The log kept lying on the incline because of the ridge there. 626 (Talmy 2000: 416)

627 One such specific force interaction is a letting relation (Talmy 2000: 419). This relation refers 628 to a situation in which the antagonist does not oppose the force exertion of the agonist, so that 629 there is no blocking or hindrance. In other words, the antagonistic entity ‘lets’ the agonist 630 perform the desired action. It is this relation that is expressed in a potential construction. For 631 example, in the Luganda potential clause in (30), it is asserted that an implied agonistic agent

18

Draft for comments (2015) - Do not cite without author’s permission

632 can drink the water without the antagonistic patient ‘the water’ hindering its ingestion. The 633 speaker furthermore elaborates the favorable properties of the antagonistic entity such as it 634 being soft and fresh.

635 (30) LUGANDA (JE15) (LCK) 636 […] eno nga ya mazzi maseeneekerevu, mawoomerevu, gamirika mu kuganywa, […]. 637 eno nga e-a ma-zzi ma-seeneekerevu, ma-woomerevu, 638 DEM9 like PP9-CON CL6-water PP6-soft PP6-fresh 639 ga-mir-ik-a mu ku-ga-nyw-a, (...). 640 SM6-swallow-MID-FV LOC18 CL15-OM6-drink-FV 641 ‘[…] this one has soft, fresh water, that can be swallowed (?is swallowable) in drinking, 642 […].’

643 The opposite relation holds when a potential construction has negative polarity. It then 644 expresses a hindering relation, in which the antagonistic entity’s opposed force hinders (31) 645 or blocks (32) the execution of the agonist’s action.

646 (31) LUGANDA (JE15) (LCK) 647 […] era mmwe abeetwala ku mbaga ze batabayiseeko mwandyesanga mwalya dda 648 ebitaliika. 649 era mmwe a-ba-ee-twal-a ku N-baga 650 and PRON2PL AUG2-SM2-REFL-take-FV LOC17 CL10-wedding 651 zi-e ba-ta-bayis-e-ko mu-andi-e-sang-a 652 PP10-REL SM2-NEG-invite-PRF-LOC17 SM2PL-CND-REFL-find-FV 653 mu-a-li-a dda e-bi-ta-li-ik-a. 654 SM2PL-REM-eat-FV short.time.ago AUG8-SM8-NEG-eat-MID-FV 655 ‘[…] and those of you who invite themselves to weddings where you are not invited 656 might find that you have already eaten the inedible.’

657 As we can observe in (31), the hindrance relation does not automatically imply that the agonist’s 658 force is nullified. The unwelcome guests have, probably without their knowledge, eaten goods 659 that either have a bad taste or are not meant to be eaten, i.e. the opposing force did not 660 completely stop the agonists of eating the food. What is important, though, is that the eaten 661 entities manifested properties that made them unpleasant to eat. Their ‘force’ (in non-causal 662 events, force has to be understood in a more abstract sense than the purely physical notion of 663 force) thus complicated or hindered an agent feasting on them. In (32), however, a speaker 664 states that without the accompaniment of something else, alcohol does not ‘let’ itself be drunk 665 by an implied agonistic agent.

666 (32) LUGANDA (JE15) (LCK) 667 […] omwenge omukalu tigujja kunyweka, […]. 668 o-mu-enge o-mu-kalu ti-gu-jj-a ku-nyw-ik-a, […]. 669 AUG3-CL3-alcohol AUG3-CL3-plain NEG-SM3-come-FV CL15-drink-MID-FV 670 ‘[…] alcohol alone will be undrinkable, […].’

19

Draft for comments (2015) - Do not cite without author’s permission

671 In a study on English middle constructions, which also express potentiality (Haspelmath 672 1987: 7), Davidse & Heyvaert (2007) make a categorization of different middle types. The two 673 most important types with regard to potential constructions in Bantu are (i) the process-oriented 674 middle, which “focuses on whether the properties of the entity construed as the subject are 675 conducive to the action as such” (2007: 67), and (ii) the facility-oriented middle which “focuses 676 on whether the properties of the subject-entity are conducive to carrying out the process easily 677 or with difficulty” (ibid.). The second type always involves constructions with adverbs such as 678 easily, well, or with difficulty. Bantu potentials can be categorized as process-oriented in that 679 they do not elucidate whether the process expressed by the verb can be carried out easily or 680 with difficulty, but asserts only that the subject lends itself to the execution of the action. 681 However, a basic potential construction can combine with adverbs of manner, expressing 682 facility-oriented potentiality (33).

683 (33) TSWANA (S31) (Creissels 2002: 403) 684 Mae a thubega motlhofo. 685 màɩ́ á-thúb-ɛ́χ-à mɷ̀ tɬhɷ̀ fɷ̀ . 686 CL6:œuf SM6-casser-MID-FV facilement 687 ‘Les œufs se cassent facilement.’ (‘The eggs break easily.’)

688 The close semantic relation between the two types of middles is furthermore demonstrated by 689 the fact that Seidl & Dimitriadis give two interpretations for the process-oriented Swahili 690 potential construction in (24) above, viz. this letter is readable or this letter could be read easily. 691 The second interpretation relates more closely the facility-oriented middle type. Moreover, 692 Muriungi (2008) also gives a facility-oriented interpretation (“an “easy reading: Z was easy to 693 V”; Muriungi 2008: 3) of the simple potential construction in Kitharaka (34).

694 (34) THARAKA (E54) (Muriungi 2008: 3) 695 mbûri n-î-ûrag-îk-ir-e. 696 CL9:goats FOC-SM9-kill-MID-PRF-FV 697 ‘The goat was easy to kill.’ (It wasn’t strong.)

698 It remains to be established whether the facility-oriented interpretation of potential 699 constructions without manner adverbs is due to a misreading of the authors attributable to the 700 close similarity between process- and facility-oriented types, or whether potential constructions 701 can be classified as both process-oriented and facility-oriented (although then without the overt 702 expression of the facility) in some Bantu languages.

703 6.1.2. Actuality status of the letting relation 704 Actuality refers to whether something has taken place, takes place at the moment of speaking 705 or will take place in the future. Palmer (1977) provides an account of actuality in relation with 706 the English modal verbs can and will, and defines it as the taking place of the event expressed 707 by the . Here, I will discuss the actuality of the letting relation in potential 708 constructions. This differs from the actuality of events in that in a potential construction the 709 speaker does not overtly state whether the event expressed by the lexical verb has been 710 actualized.

20

Draft for comments (2015) - Do not cite without author’s permission

711 The three factors that I will discuss here on which the actuality status of the letting relation 712 in a potential construction is dependent, are tense distinction, polarity and conditionality. The 713 status is relatively straightforward in potential constructions in the present and future tense. The 714 letting relation is actualized in a potential construction in the present tense, as can be observed 715 in (30) above, or is claimed to hold in the future when the clause has future tense marking (32). 716 If a positive potential construction is in the past tense, it is implied that the state of letting no 717 longer holds at the moment of speaking. Negative potentials interact semantically different with 718 the past tense. In (35), for example, the speaker asserts that during a specific period in the past, 719 land was easy to buy. What is implied is that the letting relation is no longer actualized at the 720 moment of speaking. Because people talk more frequently about what is and can or will be than 721 about what is not, it seems hypothetically logical that the positive potenial is more frequently 722 attested in the present and future tense than in the past. This was borne out by a query of the 723 Luganda corpus. Although no extensive query was performed, I did not find any positive 724 potential construction in the past tense. The example in (35) has been elicited by D. Kawalya 725 and is nevertheless fully grammatical. This instantiates the above hypothesis that in natural 726 language, this construction is not frequently used.8

727 (35) LUGANDA (JE15) (D. Kawalya, personal communication) 728 Ku mulembe gwa Muteesa ettaka lyali ligulika. 729 Ku mu-lembe gwa Muteesa e-li-taka li-a-li 730 LOC17 CL3-reign CON3 Muteesa AUG5-CL5-land SM5-PST-be 731 li-gul-ik-a. 732 SM5-buy-MID-FV 733 ‘On Muteesa’s reign, land was affordable.’

734 Negative potentials interact semantically different with respect to past tense. In (36), it is not 735 necessarily implied that the letting relation no longer holds. Although translated in English in 736 the present, embuzi tezaabalika literately means ‘goats were uncountable’. Here, it is not 737 necessarily implied that the hindrance relation is no longer actualized (e.g. ‘and nobody knows 738 how many were stolen up until today’), nor that the hindrance is still in effect at the time of 739 speaking (e.g. ‘but we conducted an investigation and found he stole three hundred of our 740 livestock’). Past negative potential constructions are thus neutral with respect to the actuality 741 status, and probably one of the two is picked out by the listener through implicature if the 742 present actuality status of the hindrance relation is not mentioned by the speaker in the 743 preceding or following discourse.

744 (36) LUGANDA (JE15) (LCK) 745 Kabaka n’agenda e Sayi e Kyagwe, n’anyagayo ente lunaana, embuzi tezaabalika. 746 Kabaka ne a-gend-a e Sayi e Kyaggwe ne 747 king and SM1-go-FV CNJ Sayi LOC23 Kyaggwe and 748 a-nyag-a-yo e-n-te lu-naana

8 Note also that the VP of a past positive potential construction in Luganda is obligatorily construed with an auxiliary expressing tense-aspect. Although the auxiliary is only required for this specific construction (i.e., past positive potential), it is definitely an interesting development in the light of the case study of the constructionalization of potential constructions in Chichewa (see section 6.4. below). 21

Draft for comments (2015) - Do not cite without author’s permission

749 SM1SG-loot-FV-LOC23 AUG10-CL10-cow CL11- eight.hundred 750 e-n-buzi te-zi-a-bal-ik-a. 751 AUG10-CL10-goat NEG-OM10-REM-count-MID-FV 752 ‘And the King went to Sayi in Kyaggwe and looted (from there) eight hundred cows and 753 countless goats.’

754 The clause in (37) below is different from the potential constructions above in that it is a 755 conditional clause. In terms of Sweetser’s (1990: 113ff.) pragmatic account of conditional 756 constructions, the example in (37) is a content conditional meaning that “the realization of the 757 event or state of affairs described in the protasis is a sufficient condition for the realization of 758 the event or state of affairs described in the apodosis.” An important effect of conditionality on 759 the potential is that it enhances the status of the letting value. In contrast to a simple proposition 760 such as this success is obtainable, which implies that the status of the letting value is actualized 761 at the moment of speaking, the conditional asserts that a requirement has to be met before the 762 success is obtained by the agonist. Thus, even though the potential verb is marked for the 763 present tense, the status of the letting value is only actualized when the condition expressed in 764 the protatis is realized.

765 (37) LUGANDA (JE15) (LCK) 766 Obuwanguzi buno bufunika ng’omwana wo agenze mu ssomero. 767 o-bu-wanguzi buno bu-fun-ik-a nga o-mu-ana 768 AUG14-CL14-succes DEM14 SM14-get-MID-FV if AUG1-CL1-child 769 u-o a-gend-e mu li-somero. 770 PP1-POSS2SG SM1-go-PRF LOC18 CL5-school 771 ‘This success is obtainable if your child has gone to school.’

772 In Gusii, the actualization is marked by means of suffix reduplication. If the verb of a 773 potential construction has one middle suffix, the letting relation is actualized (38b). However, 774 if the middle suffix is reduplicated, the implied reading is one in which the letting relation will 775 hold in the future (38c), but not yet at the moment of speaking. The author has not glossed the 776 examples in the original source, but gives the basic verb in isolation which allows us to identify 777 the middle marked verb (indicated in bold) in the constructions.

778 (38) GUSII (JE42) (Cammenga 2002: 381) 779 a. -miñɔk- ‘run’ 780 aßaana ηkomiñɔka ßare. 781 ‘The kids are running.’ 782 b. -miñɔɣɛk- ‘be actually runnable’ (i.e. it is possible to hold a race there, for example). 783 ase aria ηkomiñɔɣɛka are. 784 ‘That place is runnable’ (i.e. it is possible to hold a race there, for example). 785 c. -miñɔkɛɣɛk- ‘to be (potentially) runnable’ 786 ase aria ηkomiñɔkɛɣɛka are. 787 ‘That place is (potentially or latently) runnable’ (i.e. it may be used as a racecourse, 788 for example, but not readily so: it will first have to be made fit for that purpose).

22

Draft for comments (2015) - Do not cite without author’s permission

789 6.2. Modality: Participant-internal possibility 790 Potentiality is a semantic concept often discussed in works on modality, and closely related to 791 the notion of ability (Palmer 2001: 70) and possibility (Palmer 2001: 79). Informally, a potential 792 construction can be paraphrased as ‘can be V-ed’ in English, thus expressing a positive 793 possibility of the subject to be affected by the implied agent. Possibility pertains here to the 794 fact that the event expressed by the verb is not yet actualized (Palmer 2001: 70), to the speaker 795 wanting to indicate that it is possible to undertake the denoted action and furthermore that s/he 796 is quite certain that the desired effect of the action on the patient (or more generally the 797 endpoint) will be obtained. 798 The notion of ability is related to the antagonistic subject’s force exertion, i.e. its 799 involvement in the letting relation. With a potential construction the speaker makes a statement 800 about the ability of the antagonist subject to exert its opposing force in a force interaction 801 between the implied agent and the affected patient, namely the speaker indicates that its ability 802 to exert counterforce is low. As the force-exerting ability is internal to the subject of the 803 construction, the potential can be subsumed under the modal domain of ‘participant-internal 804 modality’:

805 “This term refers to a kind of possibility or necessity internal to the participant engaged 806 in the state of affairs. In the case of possibility we are dealing with a participant’s ability 807 (capacity) (…).” (Van der Auwera & Plungian 1998: 80)

808 The modality expressed by the potential is thus dynamic participant-internal modality (Palmer 809 2001: 76-77, Van der Auwera & Plungian 1998: 81). Palmer (2001) contrasts dynamic with 810 deontic modality in that:

811 “with deontic modality the conditioning factors are external to the relevant individual, 812 whereas with dynamic modality they are internal. Thus, deontic modality relates to 813 obligation or permission, emanating from an external source, whereas dynamic modality 814 relates to ability or willingness, which comes from the individual concerned.” (Palmer 815 2001: 9-10)

816 Again, we can say that the potential construction is a proposition in which something is asserted 817 about the willingness of the antagonistic subject to allow the desired force exertion of the 818 implied agonist, thus expressing dynamic modality. Both deontic and dynamic modality can be 819 subsumed under the general type of event modality, which is concerned with potential future 820 events (Palmer 2001: 8).

821 6.3. Subjectivity 822 Subjectivity is often involved in modal assertions (Palmer 2001: 33, 75). Finegan (1995: 4) 823 states that one of the three areas having been the focus in the study of subjectivity is “a 824 locutionary agent’s expression of the modality or epistemic status of the propositions contained 825 in utterances” (italics in original). Slightly different definitions of the notion of subjectivity 826 have been provided by linguists working on the topic. For example, Cuyckens et al. (2010: 1) 827 write that “in linguistic theorizing, [subjectivity] refers, broadly speaking, to the centrality of 828 the speaker in language”, whereas Finegan (1995: 1) defines it as “the involvement of a

23

Draft for comments (2015) - Do not cite without author’s permission

829 locutionary agent in a discourse, and the effect of that involvement on the formal shape of 830 discourse (…).” The potential construction is a subjective statement in that it expresses an 831 evaluation of the speaker of the likelihood of a state of affairs (Nuyts 2001: 383). This means 832 that the speaker is fully responsible for his/her evaluation. The subjective value of a potential 833 construction is closely related to the evidential construction marked by the same suffix (section 834 7.2.). Nuyts (2001: 386) argues that subjectivity is best defined “in terms of a difference in the 835 status of the evidence and the epistemic evaluation based on that evidence from the perspective 836 of the (knowledge of the) interaction partners in that situation.” As we will see, evidential 837 constructions in Luganda typically express the speaker’s uncertainty of the likelihood of the 838 state of affairs due to an indirect observation of evidence of that state of affairs. 839 Having separately treated the three semantic concepts involved in the semantics of the 840 potential construction, we can now join them in a unified definition of the potential’s meaning: 841 a potential clause expresses the subjective judgment of a speaker concerning the participant- 842 internal ability of the antagonist-subject’s force dynamics. A similar analysis has been made by 843 Davidse & Heyvaert (2007: 62) for middle constructions in English, namely that “middles 844 express a judgment of how the antagonist force exerted by the inanimate subject entity impacts 845 on the agonist’s action.” They furthermore argue that the anticausative construction expresses 846 an objective assertion about a certain state of affairs, and is linked to the middle/potential “by 847 a process of SUBJECTIFICATION, viz. a shift from the description of a verifiable state of affairs 848 to a subjective statement of dynamic modality” (Davidse & Heyvaert 2007: 73). Their semantic 849 analysis of the English middle is comparable, if not identical, to the description of the potential 850 construction in Bantu.

851 6.4. A case of constructionalization: potentials in Chichewa 852 In Chichewa, the potential construction differs morphosyntactically from the 853 anticausative/agentless passive construction. The most important development is a 854 morphological change in which the subject is no longer marked on the verb by means of the 855 subject marker in the potential, but instead via an attributive prefix (which is a concatenation 856 of the infinitival prefix ku- and the connective -a; Corbett & Mtjenje 1987: 11, footnote). 857 Compare the anticausative construction in (39) with the potential construction in (40).

858 (39) CHEWA (N31b) (Dubinsky & Simango 1996: 760) 859 mbale zi-na-sw-ek-a. 860 plates SM-PST-break-MID-FV 861 i. ‘The plates were broken.’ 862 ii.*‘The plates were breakable.’

863 (40) CHEWA (N31b) (Dubinsky & Simango 1996: 759) 864 mwana uyu ndi wo-kumbatil-ik-a. 865 child DEM is ATTR-embrace-MID-FV 866 ‘This child is embraceable.’

867 Furthermore, if the lexical verb is part of the predicate, as in (40), it differs from the 868 anticausative/agentless passive in that the lexical verb is non-finite. Tense and aspect are

24

Draft for comments (2015) - Do not cite without author’s permission

869 subsequently encoded on an auxiliary.9 However, the lexical verb can also function as an 870 attributive , in which case there is no auxiliary as shown in (41) and (42).

871 (41) CHEWA (N31b) (Mchombo 1993: 8) 872 mbûzi zó-b-ěk-a 873 CL10:goat ATTR10-steal-MID-FV 874 ‘stealable goats’ (lit. ‘goats that can be stolen’)

875 (42) CHEWA (N31b) (Mchombo 1993: 8) 876 chitumbúwá chó-dy-ěk-a 877 CL7:pancake ATTR7-eat-MID-FV 878 ‘an edible pancake’ (lit. ‘pancakes that can be eaten’)

879 The grammatical change in which potential constructions have become structurally different 880 from the anticausative/agentless passive constructions is an example of constructionalization, 881 or “the dynamic reorganization of associations in the constructional network of a language” 882 (Trousdale & Norde 2013: 36; also Croft 2001: 127). In this case, the reorganization that has 883 taken place is that of the potential construction entering the network of attributive constructions. 884 Other predicative attributive constructions are identical to the potential in their morphosyntactic 885 structure, i.e. the presence of an auxiliary having the subject marker and tense-aspect 886 morphology, and the lexical verb agreeing with the subject by means of the attributive prefix 887 (43).

888 (43) CHEWA (N31b) (Kiso 2012: 84) 889 a-dza-khal-a o-zizir-a. 890 SM-FUT-become-FV ATTR-be.cold-FV 891 ‘It (the water) will be cold.’

892 Note that the attributive prefix is formally different for non-verbal predicates, such as mfupi 893 ‘short’ in (44). In case of , it consists of the pronominal prefix (here u-) instead of the 894 infinitival prefix, and the connective.

895 (44) CHEWA (N31b) (Dubinsky & Simango 1996: 761) 896 Chibwe a-na-li wa-mfupi. 897 Chibwe SM-PST-be ATTR-short 898 ‘Chibwe was short.’

899 Thus, the potential construction is differentiated from the anticausative, no longer solely by the 900 verb class (anticausative vs. non-anticausative verbs), but also by a different construction type. 901 In terms of grammaticalization theory, the potential has first been reanalyzed as expressing 902 attributive semantics, viz. the attribution of a quality (the low ability to exert force opposition), 903 and subsequently the potential meaning was complemented with morphosyntactic change,

9 The auxiliary ndi in (40) differs from regular auxiliaries in Chichewa in that it is a lexicalized “used for nominal predication in sentences referring to the present […]”, and “is always unmarked for tense-aspect and person […]” (Kiso 2012: 25). The copula in (44), for example, does take agreement and past tense marking. 25

Draft for comments (2015) - Do not cite without author’s permission

904 losing morphology of finite verbs and, by analogy with other attributive constructions, 905 attracting morphology of non-finite verbs expressing attributive qualities. 906 That grammatical change is a gradual process, often with intermediate stages in which old 907 and new meanings and/or constructions are attested at the same time (Hopper & Traugott 2003: 908 52), is illustrated by the fact that attributive constructions with anticausative verbs are 909 ambiguous between a potential and anticausative reading (45).

910 (45) CHEWA (N31b) (Dubinsky & Simango 1996: 760) 911 mbale zi-na-li zo-sw-ek-a. 912 plates SM-PST-be ATTR-break-MID-FV 913 i. ‘The plates were broken.’ 914 ii. ‘The plates were breakable.’

915 7. The middle marker with perception verbs

916 7.1. Stimulus constructions 917 Perception verbs have been categorized based on a variety of features. Some authors make a 918 distinction between ‘active’ versus ‘passive’ (Willems 1983: 150, Willems & Defrancq 2000: 919 8), ‘agentive’ versus ‘non-agentive’ (Willems 1983: 158, Willems & Defranq 2000: 9), and/or 920 ‘experiencer’ versus ‘stimulus’ (Usoniene 1999: 212) verbs. The well-known classification of 921 Viberg (1984: 121) combines all these features, making up three general event types of 922 perception verbs: controlled activity (e.g. look at), non-controlled experience (e.g. see), and 923 source-based copulative (e.g. look, seem, appear). If a perception verb is marked by the middle 924 suffix, it can figure in an agentless passive construction as in (12), repeated here as (46), a 925 potential construction (47), or a stimulus construction (48).

926 (46) NDONGA (R22) (Fivaz 1986: 111 in Fleisch 2005: 123) 927 ókinó ndjoká o-y-a-tál-ik-á ká-á-ntu a-yéhe. 928 CL9:film DEM9 AFF-SM9-PFV-watch-MID-FV INSTR-CL2-person PP2-many 929 ‘This film is seen by many people.’

930 (47) LUGANDA (JE15) (LCK) 931 Bw'olyengereza kiki ekirabika? 932 Bwe o-li-engel-ez-a kiki eki-lab-ik-a? 933 if SM2SG-OM5-ripe-CAUS-FV what REL7-see-MID-FV 934 ‘If you burn it, what can be seen?’

935 (48) KINDIBU (H16b) (KKC)10 936 I una yeno kuna mbazi lumoneka kwa wantu ne ansongi. 937 i u-na yeno ku-na m-bazi lu-mon-ik-a ku-a 938 CNJ PP1-be PRON2PL PP17-be CL9-outside SM2PL-see-MID-FV PP17-CON 939 wa-ntu ne a-nsongi.

10 KKC stands for KongoKing Corpus, which is a Kikongo corpus compiled by the KongoKing research group at Ghent University. 26

Draft for comments (2015) - Do not cite without author’s permission

940 CL2-person and CL2-rightousness 941 ‘So you also on the outside look righteous to others.’

942 In stimulus-oriented constructions, it is not atypical for the senser to be overtly expressed. In 943 (48), this is the generic NP wantu ‘people’. Similar to the overt senser in agentless passive 944 constructions, it seems as if this participant can only be denoted by generic NPs since no 945 examples have been attested in which a specific senser is present. Constructions expressing the 946 comparative stimulus situation ‘look like’ generally consist of a subject denoting the stimulus 947 entity and a complement denoting the source of the comparison, i.e. what the subject entity is 948 said to look like. These can be complex complements, e.g. relative clauses referring to concrete 949 entities such as ‘people who are appreciated in your eyes’ in (49), or complements denoting a 950 quality such as ‘nice and in order’ in (50).

951 (49) LUGANDA (JE15) (LCK) 952 […] tulioke tulabike nga abantu abasimibwa mu maso go, […]. 953 tulioke tu-lab-ik-e nga a-ba-ntu 954 so SM1PL-see-MID-SBJ like AUG2-CL2-person 955 a-ba-sim-ibu-a mu ma-iso ga-o, […]. 956 AUG2-SM2-appreciate-PASS-FV LOC18 CL6-eye PP6-POSS2SG 957 ‘[…] so that we look like people who are appreciated in your eyes, […].’

958 (50) LUGANDA (JE15) (LCK) 959 […] birabike bulungi nga biteredde. 960 bi-lab-ik-e bu-lungi nga bi-tereer-e. 961 SM8-see-MID-SBJ CL14-nice like SM8-be.in.order-SBJ 962 ‘[…] so that they look like they are nice and in order.’

963 The Luganda verb -bonek-, a reflex of the reconstructed Proto-Bantu form *-bon- ‘see’ and the 964 derived form *-bonek- ‘appear’, is most often used to express the appearing of the moon (D. 965 Kawalya, personal communication), as illustrated in (51).

966 (51) LUGANDA (JE15) (LCK) 967 […] bwe yatandika okwekubanga yokka, buli mwezi lwe gwabonekanga. 968 bwe e-a-tandik-a o-ku-e-kub-a-nga e-okka, 969 when SM9-REM-start:MID-FV AUG15-CL15-REFL-beat-FV-REP PP9-self 970 buli mu-ezi lwe gu-a-bonek-a-nga. 971 each.time CL3-moon when SM3-REM-appear:MID-FV-REP 972 ‘[…] when it [the drum] started beating itself, each time the moon appeared.’

973 The verb -bonek- has become lexicalized. The underived verb -bon- has been replaced by the 974 verb -lab- ‘see’ in present-day Luganda. In (52), -bonek- is used in a more modern and non- 975 idiomatic context to refer to the appearing of different kinds of food. It thus seems that the use 976 of -bonek- is (re)expanding its selectional restriction to express not only the appearance of the 977 moon but also that of other entities.

978 (52) LUGANDA (JE15) (LCK)

27

Draft for comments (2015) - Do not cite without author’s permission

979 Era olwo enkoko oba turkey […] lwe bibonekako n'ebikyepere ebirala nga Christmas 980 pudding ne bigendera okwo. 981 era olwo e-N-koko oba Ø-turkey lwe 982 and then AUG9-CL9-chicken or CL9-turkey when 983 bi-bonek-a-ko ne e-bi-kyepere e-bi-rala 984 SM8-appear:MID-FV-LOC17 with AUG8-CL8-nice.food AUG8-CL8-other 985 nga Christmas pudding ne bi-gend-er-a o-kwo. 986 like Christmas pudding and SM8-go-APPL-FV AUG17-DEM17 987 ‘And then when chicken or turkey […] appeared together with other kinds of nice food 988 like Christmas pudding and others.’

989 However, the middle marked verb -labik- is more commonly used to denote events of appearing 990 with entities other than the moon. In such stimulus constructions, the time (53) or location (54) 991 of appearance is often elaborated.

992 (53) LUGANDA (JE15) (LCK) 993 […] Bad Black alabikeko nga 31 omwezi ogujja. 994 Bad Black a-lab-ik-e-ko nga 31 o-mu-ezi 995 Bad Black SM1-see-MID-SBJ-LOC17 on 31 AUG3-CL3-month 996 o-gu-jj-a. 997 AUG3-SM3-come-FV 998 ‘[…] Bad Black to appear on the 31st of next month.’

999 (54) LUGANDA (JE15) (LCK) 1000 Erinnya Muhammadi lirabikira mu Kulaane emirundi ena, […]. 1001 e-li-nnya Muhammadi li-lab-ik-ir-a mu Kulaane 1002 AUG5-CL5-name Muhammad SM5-see-MID-APPL-FV LOC18 Koran 1003 e-mi-rundi e-na, (…). 1004 AUG4-CL4-time PP9-four 1005 ‘The name Muhammad appears four times in the Koran’.

1006 The event of appearing is closely related to perception events such as see and look (at), or, to a 1007 lesser extent, to ditransitive perception verbs such as show, reveal, bring in(to ‘a location’). 1008 The subject of appear corresponds to the complement of these agnate active (look at) or 1009 experiential (see) perception verbs, although this is a unidirectional alternation. Every subject 1010 of appear can figure as the complement of a perception construction, but not every complement 1011 can figure as the subject of a stimulus construction. This is because complements can be 1012 (sub)clauses denoting acts or facts. Thus, the sun appeared on the horizon is agnate to the police 1013 detective saw the sun on the horizon, but the traveler saw the train leave does not make sense 1014 as the train leave appeared, or even as the leaving train appeared. Thus, perception 1015 constructions do not alternate with stimulus constructions in the same way that 1016 causative/anticausative constructions do, although they do allow a similar but one-way 1017 alternation. This similarity also explains why stimulus constructions are marked by the same 1018 middle suffix as anticausatives.

28

Draft for comments (2015) - Do not cite without author’s permission

1019 The stimulus subjects in the examples above are all low in agency, i.e. they are not conscious, 1020 volitional, intentional or human/animate. However, this should not always be the case. In (55) 1021 the subject of the stimulus construction is the human entity Kabaka ‘king’. In the light of the 1022 discussion in the previous paragraph, the construction in (55) is agnate to ‘show (oneself) to’ 1023 rather than to ‘see’.

1024 (55) LUGANDA (JE15) (LCK) 1025 Era byabeeranga mu buli kigango Kabaka mwe yalabikiranga. 1026 era bi-a-ba-er-a-nga mu buli ki-gango Ø-Kabaka 1027 and SM8-REM-be-APPL-FV-REP LOC18 each.time CL7-palace CL1-king 1028 mu-e a-a-lab-ik-ir-a-nga. 1029 PP18-REL SM1-REM-see-MID-APPL-FV-REP 1030 ‘And they were always in each palace where the king usually appeared.’

1031 The relation between the stimulus-oriented construction and an active ditransitive construction 1032 with ‘show (oneself) to’ when having a human/animate entity as stimulus is more clearly 1033 brought out in (56), for which the English interpretation of -labik- is more naturally translated 1034 into ‘manifest oneself’ instead of ‘appear’, and could also be paraphrased as ‘he has showed 1035 himself at Busiro hill’.

1036 (56) LUGANDA (JE15) (LCK) 1037 “Yee tumaze okumulaba, era alabikidde wali ku lusozi Busiro.” 1038 yee tu-mal-e o-ku-mu-lab-a, era a-lab-ik-ir-e 1039 yes SM1PL-finish-PRF AUG15-CL15-OM1-see-FV and SM1-see-MID-APPL-PRF 1040 a-wali ku lu-sozi Busiro. 1041 AUG16-DEM16 LOC17 CL11-hill Busiro 1042 ‘“Yes, we have already seen him (the king), and he has manifested himself at Busiro 1043 hill.”’

1044 The defocusing and backgrounding of an initiating participant, i.e. the senser in a perception 1045 event, which I have argued to be one of the main pragmatic functions of the middle marker, is 1046 illustrated in (57). In this construction, the verb base -labik- is nominalized to ndabika ‘format’ 1047 by adding the nominal prefix of class 9 to it. Although the format of a text or document pertains 1048 to its visual aspects, i.e. the design, there is absolutely no implication of a senser participant. 1049 Nevertheless, the document’s format cannot make itself up ‘spontaneously’, but has to be 1050 created by a (backgrounded and demoted) causing agent.

1051 (57) LUGANDA (JE15) (LCK) 1052 Tekyetaagisa kubaako nfaanana oba ndabika egobererwa mu kuwandiika eddaame. 1053 te-ki-etaagis-a ku-ba-ko n-faanana oba n-labika 1054 NEG-SM7-require:CAUS-FV CL15-have-LOC17 CL9-lay-out or CL9-format 1055 e-goberer-u-a mu ku-wandiik-a e-li-laame. 1056 SM9-follow:APPL:APPL-PASS-FV LOC18 CL15-write-FV AUG5-CL5-testament 1057 ‘It does not require to have a lay-out or format to be followed in writing a testament.’

29

Draft for comments (2015) - Do not cite without author’s permission

1058 Concerning the expression of the senser in constructions where the derived perception verb 1059 denotes the event of appearing, examples with both generic (58) and specific (59) NPs have 1060 been attested, although in different languages. It is not yet clear whether this can be generalized 1061 for these constructions in Bantu more generally, or whether the presence of a specific senser is 1062 a language-specific development.

1063 (58) LUGANDA (JE15) (LCK) 1064 […] okuyitirira kwo kulabikenga eri bonna. 1065 o-ku-yitirir-a ku-o ku-lab-ik-e-nga 1066 AUG15-CL15-be.excessive:APPL:APPL-FV PP15-POSS2SG SM15-see-MID-SBJ-REP 1067 eri ba-onna. 1068 to CL2-all 1069 ‘[…] so that your superiority appears to all.’

1070 (59) KINDIBU (H16b) (KKC) 1071 Ye mu fuku mbonameso imonekene kwa Paulo. 1072 ye mu Ø-fuku m-monameso i-mon-ik-idi ku-a Paulo. 1073 and CL18 CL9-night CL9-vision SM9-see-MID-PRF PP17-CON Paul 1074 ‘During the night a vision appeared to Paul.’

1075 The senser Paulo in (59) is more accurately experiencing a vision rather than seeing a concrete 1076 entity in the ‘outer world’. Verbs expressing the act of ‘seeing’ have generally been defined as 1077 denoting a passive perception by an involuntary, non-agentive senser, and often there is a focus 1078 on the stimulus (Gruber 1967: 943; Willems 1983: 150, 158; forthcoming: 88-89). 1079 The Luganda example in (60) is a passivized stimulus construction. The senser is now in 1080 subject position, i.e. Kyannyanja marked anaphorically on the verb alabikiddwa by the subject 1081 concord a-, and the stimulus omuzimu in object position. In contrast to many Bantu languages, 1082 where agents in passive construction can only be expressed in oblique prepositional phrases, 1083 the agent in Luganda passive constructions immediately follows the passivized verb, as can be 1084 seen in (61).

1085 (60) LUGANDA (JE15) (LCK) 1086 Kyannyanja bwe yalaba Mbayiwa yali nga alabikiddwa omuzimu. 1087 Kyannyanja bwe e-a-lab-a Mbayiwa e-a-li nga 1088 Kyannyanja when SM9-REM-see-FV Mbayiwa SM9-REM-be like 1089 a-lab-ik-ir-u-a o-mu-zimu. 1090 SM1-see-MID-APPL-PASS-FV AUG3-CL3-spirit 1091 ‘When Kyannyanja saw Mbayiwa it was as if he had experienced a spirit.’

1092 (61) LUGANDA (JE15) (D. Kawalya, personal communication) 1093 Emmotoka yatwalibwa Musaka mu galagi. 1094 e-li-motoka a-a-twal-ibw-a Mukasa mu Ø-galagi. 1095 AUG5-CL5-car SM1-REM-take-PASS-FV Mukasa LOC18 CL9-garage 1096 ‘The car was taken to the garage by Mukasa.’

30

Draft for comments (2015) - Do not cite without author’s permission

1097 Because the argument structure in the passivized construction in (60) differs from the argument 1098 structure of a stimulus construction, it can be interpreted in English as an experiencer-oriented 1099 construction as shown in the translation. However, a syntactically more correct interpretation 1100 would be a topicalized passive construction such as ‘When Kyannyanja saw Mbayiwa it was as 1101 if to him had appeared a spirit’.

1102 7.2. Evidential constructions 1103 A previously unreported function of the middle suffix is its marking of evidential constructions. 1104 Even though stimulus constructions have also been largely neglected in the literature on the 1105 middle marker, lexical examples of derived stimulus-oriented perception verbs can often be 1106 found in wordlists in grammars. However, I have not found any information on or mentioning 1107 of the evidential meaning of perception verbs with the middle suffix. This section is therefore 1108 explorative, presenting some data on Luganda evidential constructions and tries to give a first 1109 descriptive analysis based on previous accounts of non-grammaticalized evidential verbs in 1110 non-Bantu languages. 1111 Evidentiality has been variously defined in broad and strict senses (among others, see Barnes 1112 1984, Chafe & Nichols 1986, Palmer 2001, Aikhenvald 2004). In the narrow interpretation, 1113 evidentiality has to do with the specification of the source of information a speaker bases his/her 1114 proposition on. A textbook example are the evidential markers in Tuyuca, a language spoken 1115 in Colombia and Brazil. A report of one and the same event can be marked in five different 1116 ways, depending on how the speaker acquired the information expressed in the construction 1117 (Barnes 1984: 257). The most basic distinction in languages with evidential markers is made 1118 between direct and indirect evidentiality (Matlock 1989: 215), i.e. whether the speaker has 1119 experienced the reported event him/herself through any of the senses (visual, auditory, 1120 olfactory) or whether the information was acquired indirectly (i.e. through evidence, by being 1121 told or by assumption). Languages also tend to differ in how fine-grained their system of 1122 evidentiality is. Additional senses attributed to the notion of evidentiality is the attitude of the 1123 speaker towards a situation and the degree of certainty about the information (Floyd 1996: 69, 1124 Matlock 1989: 215). 1125 Evidentiality is a category generally neglected in Bantu linguistics. An exception is Botne 1126 (1995, 1997) on evidential markers in Lega, a language spoken in the eastern region of the 1127 Democratic Republic of the Congo. Some of the evidential markers discussed for Lega have a 1128 pronominal source and thus differ from the evidential verbs discussed here in that they have a 1129 different source domain. The fact that evidentiality is a concept generally not discussed by 1130 Bantuists is due to the fact that the majority of Bantu languages do not have grammatical 1131 markers dedicated to the expression of evidentiality. However, comparable to West European 1132 languages which also lack such specialized markers (Nuyts 2001), Bantu languages have non- 1133 grammaticalized means of expressing evidentiality. In terms of Aikhenvald (2004: 105), when 1134 the middle marker is combined with visual perception verbs it has acquired certain semantic 1135 features of evidentials, and is thus an evidentiality strategy rather than a grammatical evidential 1136 marker. The fact that these evidential verbs originate from perception verbs in Bantu is a general 1137 trend observed cross-linguistically to be very frequent (references). A Luganda example of an 1138 evidential construction is provided in (62).

31

Draft for comments (2015) - Do not cite without author’s permission

1139 (62) LUGANDA (JE15) (LCK) 1140 Wazira, obunyiikivu bwa kitange mu kuweereza bulabika bwali bukwata nga 1141 kawumpuli nange bwankwata era naweereza nnyo. 1142 wazira, o-bu-nyiikivu bu-a Ø-kit-ange mu 1143 but AUG14-CL14-perseverance PP14-CON CL1-father-POSS1SG LOC18 1144 ku-weerez-a bu-lab-ik-a bu-a-li bu-kwat-a 1145 CL15-serve-FV SM14-see-MID-FV SM14-REM-be SM8-grasp-FV 1146 nga ka-wumpuli na-ange bu-a-n-kwat-a era 1147 like CL12-wumpuli CON-POSS1SG SM14-REM-OM1SG-grasp-FV and 1148 n-a-weerez-a nnyo. 1149 SM1SG-REM-serve-FV very.much 1150 ‘But, my father’s hard work in service seemed contagious like kawumpuli [a certain 1151 disease], it also caught up with me and I served a lot.’

1152 Evidentials originating from visual perception verbs, such as -labik- in Luganda, typically 1153 imply that the information is obtained visually. Moreover, as we can deduct from the English 1154 equivalent of (62), the evidential verb -labik- indicates that the speaker is not completely certain 1155 about the proposition. For example, in (63) the speaker does not assert that she is pregnant, but 1156 rather that through observing certain facts she thinks that she might be pregnant for three 1157 months. The evidential verb thus expresses a low(er) degree of certainty of the speaker. 1158 Uncertainty is often a logical corollary of indirect evidentiality, since the speaker has not 1159 personally witnessed the reported state of affairs and has thus obtained the information second 1160 handed. Therefore, the truthfulness of the proposition cannot be warranted by the speaker.

1161 (63) LUGANDA (JE15) (LCK) 1162 […] n'engeri gye ndabika nti ndi lubuto olwo olw'emyezi ng'esatu. 1163 ne e-n-geri gye n-lab-ik-a nti n-li lu-buto 1164 and AUG-CL9-way REL9 SM1SG-see-MID-FV that SM1SG-be CL11-pregnancy 1165 olu-o olwa e-mi-ezi nga e-satu. 1166 PP11-DEM CON11 AUG-CL4-month about PP4-three 1167 ‘[…] and because I seem to be pregnant for about three months.’

1168 Thus, the evidential verb -labik- in Luganda does not indicate that the state of affairs has been 1169 directly perceived by the speaker, but rather that it is inferred by him/her through the perception 1170 of observable evidence. 1171 The modality expressed by the evidential construction differs from that of the potential in 1172 that, as we have seen, the latter expresses dynamic event modality, whereas the former 1173 expresses evidential propositional modality. Palmer (2001: 8) describes the difference between 1174 epistemic and evidential propositional modality as follows: “(…) with epistemic modality 1175 speakers express their judgments about the factual status of the proposition, whereas with 1176 evidential modality they indicate the evidence they have for its factual status.” However, the 1177 additional uncertainty implied by the evidential construction also locates its meaning in the 1178 domain of epistemic modality. In Palmer’s (2001: 24) terms, it denotes Speculative, i.e. the 1179 expression of uncertainty, and Deductive, i.e. indication of inference from observable evidence, 1180 epistemic modality. The definition of the Deductive category largely overlaps with the

32

Draft for comments (2015) - Do not cite without author’s permission

1181 inferential category often found in evidential systems and the two are thus closely related to 1182 each other. 1183 Evidential constructions are also subjective statements in that they encode the speaker’s 1184 involvement in the proposition, viz. it indicates how the information was obtained by the 1185 speaker and furthermore his/her evaluation of the reliability and correctness of the information. 1186 In case of the Luganda evidential construction, the speaker indicates that s/he is not completely 1187 certain that the state of affairs expressed in the proposition is true. 1188 Further research on evidential constructions marked by the middle suffix should investigate 1189 the different complementation types, analyzing the degree of grammaticalization of the 1190 evidential verb and different construction types the evidential subsumes. More examples are 1191 needed to verify quantitatively whether -labik- or evidential verbs in other Bantu languages are 1192 restricted to inference through observable evidence, or whether some evidentials have expanded 1193 over different categories in the evidential domain, expressing knowledge-based evidentiality or 1194 evidence through hearsay. Most importantly, evidentiality as a linguistic category in Bantu 1195 languages should gain wider acceptance and recognition, resulting in more data from a diverse 1196 array of Bantu languages.

1197 8. Conclusion 1198 This article is meant as a first typologically-motivated description of a polysemous verbal suffix 1199 attested in a large number of Bantu languages, whose different functions can best be subsumed 1200 under the general notion of middle. It brings together data from the existing literature with new 1201 examples from Luganda and Kikongo corpora. Putting these data in a different perspective 1202 allows us to shed new light on the description of the functions of the middle marker. The most 1203 general function of the middle marker is, in abstract terms, to derive a clause with n participants 1204 from a clause with n+1 participants; in other words, to reduce the or participant number 1205 of a basic construction with one. A three-participant construction becomes a two-participant 1206 construction, a two-participant becomes a one-participant construction, and a one-participant 1207 construction results in a construction in which no participant is overtly expressed. In case of the 1208 latter, a locational or instrumental adjunct typically takes the grammatical role of subject. 1209 The latter can be generalized into a feature of middle marked constructions, namely that the 1210 subject role is typically denoted by a non-agentive entity partaking in the event denoted by the 1211 verb. Non-agentive is to be understood in a very broad and general sense subsuming a number 1212 of participant roles such as endpoint, patient, theme, phenomenon/stimulus, and the like, 1213 dependent on the verb class. 1214 It has been shown that the middle suffix marks anticausative constructions in Bantu, and, 1215 when the verb does not participate in the causative/anticausative alternation, an agentless 1216 passive construction. Both constructions have the same argument structure, although the 1217 agentless passive function of the middle marker is not restricted to a semantically defined group 1218 of verbs and can therefore combine with both three-place, ditransitive and one-place, 1219 intransitive predicates. As a contribution to the theoretic discussion on anticausatives, I have 1220 argued that ‘spontaneity’ is useless as a semantic concept for verbs that denote dyadic, caused 1221 events. I have furthermore stated that ‘spontaneity’ has been evoked in the description of

33

Draft for comments (2015) - Do not cite without author’s permission

1222 anticausatives due to the pragmatic function of the construction, viz. the backgrounding of a

1223 causing first event (Event1) and foregrounding of the resulting second event (Event2). 1224 In the existing literature, the potential function of the middle marker is usually described in 1225 general, impressionistic terms. Here, a detailed analysis is provided combining related concepts 1226 into an in-depth description of the potential’s semantics: a potential construction expresses a 1227 subjective statement of the speaker about the ability or willingness of the antagonistic force 1228 entity to lower its opposing force, resulting in a letting relation. The subjective, modal value of 1229 potential constructions has led to the constructionalization of the potential function in 1230 Chichewa. Although this is the only Bantu language for which this development has been 1231 reported, other languages might have undergone the same or a similar constructional change. 1232 The interactions resulting from a combination of the middle marker with (mainly visual) 1233 perception verbs have gone largely unnoticed in studies in the verbal suffix. Perception verbs 1234 can be construed from the viewpoint of the senser or from the viewpoint of the 1235 stimulus/phenomenon. The middle suffix marks such stimulus-oriented constructions. Finally, 1236 an explorative section is provided on evidential constructions in Luganda. The evidential verb 1237 -labik- is best translated as ‘seem’, which has been described as an evidential verb (see Aijmer 1238 2008 and references therein). Although it cannot be said for Bantu languages that evidentiality 1239 is an important part of the grammatical system, the speaker in Luganda (and probably speakers 1240 of other Bantu languages) can choose to express the fact that the reported state of affairs is 1241 based on inference from observable evidence, and furthermore that s/he is not fully certain of 1242 the truthfulness of the proposition (due to the indirect observation of evidence rather than direct 1243 perception of the state of affairs).

1244 Abbreviations

1245 1, 2, 3, ... number of prefix class 1246 APPL applicative 1247 ATTR attributive prefix 1248 AUG augment (pre-prefix) 1249 CAUS causative 1250 CL noun class prefix 1251 CLT 1252 CND conditional 1253 CNJ conjunct 1254 CON connective 1255 DEM 1256 FOC focus 1257 FUT future 1258 FV final vowel 1259 INSTR instrumental 1260 LOC locative 1261 MID middle marker 1262 NEG negative

34

Draft for comments (2015) - Do not cite without author’s permission

1263 OM object marker 1264 PASS passive 1265 PL plural 1266 POSS 1267 PP pronominal prefix 1268 PRF perfective 1269 PRON personal 1270 PRS present 1271 PST past 1272 REC reciprocal 1273 REFL reflexive 1274 REL relative 1275 REM remote past 1276 REP repetitive 1277 SBJ subjunctive 1278 SG singular 1279 SM subject marker

1280 References 1281 Aijmer, Karin. 2008. Seem and evidentiality. Functions of Language 16(1), 63-88. 1282 Aikhenvald, Alexandra Y. 2004. Evidentiality. Oxford: Oxford University Press. 1283 Alexiadou, Artemis, Elena Anagnostopoulou & Florian Schäfer. 2006. The properties of 1284 anticausatives crosslinguistically. Online document. 1285 URL: http://www.philology.uoc.gr/staff/anagnostopoulou/a11.pdf. 1286 Alsina, Alex & Sam Mchombo. 1990. The syntax of applicatives in Chicheŵa: Problems for a theta 1287 theoretic assymetry. Natural Language & Linguistic Theory 8(4), 493-506. 1288 Alsina, Alex & Sam Mchombo. 1993. Object asymmetries and the Chichewa applicative construction. 1289 In Sam Mchombo (ed.) Theoretical Aspects of Bantu Grammar, 17-45. Stanford: Center for the 1290 Study of Language and Information. 1291 Baker, Mark. 1988. Theta theory and the syntax of applicatives in Chicheŵa. Natural Language and 1292 Linguistic Theory 6, 353-389. 1293 Baker, Mark. 1990. Elements of a typology of applicatives in Bantu. In J. Hutchinson and V. Manfredi 1294 (eds.) Current Approaches to African Linguistics 7, 111-124. Dordrecht: Foris. 1295 Bakker, Dik. 2011. Language sampling. In Jae Jung Song (ed.) The Oxford Handbook of Linguistic 1296 Typology, 100-127. Oxford: Oxford University Press. 1297 Barlow, A. Ruffel. 1951. Studies in Kikuyu grammar and idiom. Edinburgh: William Blackwood & 1298 Sons Ltd. 1299 Barnes, Janet. 1984. Evidentials in the Tuyuca verb. International Journal of American Linguistics 1300 50(3), 255-271. 1301 Beart, Thomas. 2006. Syntax. In D. Nurse & G. Philippson (eds.) The Bantu Languages, 121-142. 1302 London/New York: Routledge 1303 Biloa, Edmond. 1994. Passive incorporation and clause structure. Journal of West African Languages 1304 24(1), 101-107.

35

Draft for comments (2015) - Do not cite without author’s permission

1305 Bostoen, Koen & Léon Mundeke. 2011. The causative/applicative syncretism in Mbuun (Bantu B87, 1306 DRC): Semantic split or phonemic merger? Journal of African Languages and Linguistics 32, 179- 1307 218. 1308 Bostoen, Koen, Sebastian Dom & Guillaume Segerer. Forthcoming. The antipassive in Bantu. 1309 Linguistics. 1310 Botne, Robert. 1995. The pronominal origin of an evidential. Diachronica 12(2), 201-221. 1311 Botne, Robert. 1997. Evidentiality and epistemic modality in Lega. Studies in Language 21(3), 509- 1312 532. 1313 Botne, Robert. 2006. Lega (Beya ) (D25). In D. Nurse & G. Philippson (eds.) The Bantu 1314 Languages, 422-449. London/New York: Routledge. 1315 Bresnan, Joan & Lioba Moshi. 1990. Object assymetries in comparative bantu syntax. Linguistic 1316 Inquiry 21(2), 147-185. 1317 Bresnan, Joan & Sam A. Mchombo. 1987. Topic, pronoun and agreement in Chicheŵa. Language 1318 63(4), 741-782. 1319 Brisard, Frank & Michael Meeuwis. 2011. Present and perfect in Bantu: The case of Lingála. Journal 1320 of African Languages and Linguistics 30, 21-43. 1321 Cammenga, Jelle. 2002. Phonology and morphology of Ekegusii, A Bantu language of Kenya. 1322 Cologne: Rüdiger Köppe Verlag. 1323 Cann, Ronnie & Patricia Mabugu. 2007. Constructional polysemy: the applicative construction in 1324 chiShona. Metalinguistica 19, 221-245. 1325 Chafe, Wallace & Johanna Nichols (eds.). 1986. Evidentiality: The linguistic encoding of 1326 epistemology. Norwoord, New Jersey: Ablex. 1327 Cole, Desmond T. 1955. An introduction to Tswana grammar. Cape Town: Longmans, Green and Co., 1328 Ltd. 1329 Corbett, Greville G. & Alfred D. Mtenje. 1987. Gender Agreement in Chichewa. Studies in African 1330 Linguistics 18/1, 1-38. 1331 Coupez, André. 1985. La dérivation verbale en (bantou J61). Le , 7-41. 1332 Creissels, Denis & Sylvie Nouguier-Voisin. 2008. Valency-changing operations in Wolof and the 1333 notion of co-participation. In E. König & V. Gast (eds.) Reciprocals and reflexives: theoretical and 1334 typological explorations, 293-309. Berlin/New York: Mouton de Gruyter. 1335 Creissels, Denis. 2002. Valence verbale et voix en tswana. Bulletin de la Société de linguistique de 1336 Paris, fasc. 1, 371-426. 1337 Creissels, Denis.2004. Non-canonical applicatives and focalization in Tswana. Syntax of the World’s 1338 Languages (SWL 1). Online document. 1339 URL: http://elisa.ddl.ish-lyon.cnrs.fr/fulltext/Creissels/Creissels_2004d.pdf. 1340 Croft, William. 2001. Radical construction grammar: Syntactic theory in typological perspective. 1341 Oxford: Oxford University press. 1342 Cuyckens, Hubert, Kristin Davidse & Lieven Vandelanotte. 2010. Introduction. In H. Cuyckens, K. 1343 Davidse & L. Vandelanotte (eds.) Subjectification, intersubjectification and grammaticalization, 1- 1344 26. Berlin: Mouton. 1345 Dammann, Ernst. 1954. Reziprok und Associativ in Bantusprachen. Zeitschrift der deutschen 1346 morgenländischen Gesellschaft 29, 163-174. 1347 Dammann, Ernst. 1961. Das Applikativum in den Bantusprachen. Zeitschrift der deutschen 1348 morgenländischen Gesellschaft 36, 160-169. 1349 Davidse, Kristin. 1992. Transitivity/ergativity: the Janus-headed grammar of actions and events. In 1350 Martin Davies & Louise Ravelli (eds.) Advances in Systemic Linguistics: Recent Theory and 1351 Practice, 105-135. London/New York: Printer.

36

Draft for comments (2015) - Do not cite without author’s permission

1352 De Kind, Jasper & Koen Bostoen. 2012. The applicative in ciLubà grammar and discourse: A 1353 semantic goal analysis. Southern African Linguistics and Applied Language Studies 30(1), 101- 1354 124. 1355 Demuth, Katherine. 1998. Argument structure and acquisition of Sesotho applicatives. Linguistics 36, 1356 781-806. 1357 Dom, Sebastian, Guillaume Segerer & Koen Bostoen. Forthcoming. Antipassive/associative polysemy 1358 in Cilubà (Bantu, L31a): A plurality of relations analysis. Studies in Language. 1359 Dubinsky, Stanley & Silvester Ron Simango. 1996. Passive and stative in Chichewa: Evidence for 1360 modular distinctions in grammar. Language 72(4), 749-781. 1361 Finegan, E. 1995. Subjectivity and subjectivisation: an introduction. In D. Stein & S. Wright (eds.) 1362 Subjectivity and subjectivisation, 1-15. Cambridge: Cambridge University press. 1363 Fivaz, D. 1986. A reference grammar of Oshindonga. Windhoek: The Academy [African Studies of 1364 the Academy 1]. 1365 Fleisch, Alex. 2005. Agent phrases in Bantu. In E.F.K. Voeltz (ed.) Studies in African Linguistic 1366 Typology, 95-127. Amsterdam: John Benjamins. 1367 Fleisch, Axel. 2005. Agent phrases in Bantu passives. In E.F.K. Voeltz (ed.) Studies in African 1368 Linguistic Typology, 93-112. Amsterdam: John Benjamins. 1369 Floyd, Rick. 1996. Experience, certainty and control, and the direct evidential in Wanka Quechua 1370 questions. Functions of Language 3(1), 69-93. 1371 Gérard, R. 1924. La langue lebéo: grammaire et vocabulaire. Brussels: Vromant & Co. 1372 Givón, Talmy & Boniface Kawasha. 2006. Indiscrete grammatical relations: The Lunda passive. In T. 1373 Tsunoda & T. Kageyama (eds.) Voice and Grammatical Relations: In Honor of Masayoshi 1374 Shibatani, 14-41. Amsterdam: John Benjamins. 1375 Goatly, A. 1996. Green grammar and grammatical metaphor, or Language and the myth of power, or 1376 Metaphors we die by. Journal of Pragmatics 25/4, 537-560. 1377 Goldberg, Adele. 1992. The inherent semantics of argument structure: The case of the English 1378 ditransitive construction. Cognitive Linguistics 3(1), 37-74. 1379 Gruber, Jeffrey S. 1967. Look and see. Language 43(3), 937-947. 1380 Guthrie, Malcolm. 1967. Bantu Grammatical Reconstructions. Africana Linguistica 3, 79-121. 1381 Guthrie, Malcolm. 1971. Comparative Bantu, vol. 2. London: Gregg Press. 1382 Halliday, Alexander Kirkwood & Matthiessen, Christian M.I.M. 2014. Halliday’s introduction to 1383 Functional Grammar (4th ed.). London/New York: Routledge. 1384 Harford, Carolyn. 1993. The applicative in Shona and lexical mapping theory. In S. Mchombo (ed.) 1385 Theoretical aspects of Bantu grammar, 93-111. Stanford: Center for the Study of Language and 1386 Information. 1387 Harries, Lyndon. 1940. An outline of Mawiha grammar. Bantu Studies 14(2), 91-146. 1388 Haspelmath, Martin. 1987. Transitivity alternations of the anticausative type. Arbeitspapier 5, 1-51. 1389 Haspelmath, Martin. 1993. More on the typology of inchoative/causative verb alternations. In B. 1390 Comrie & M. Polinsky (eds.) Causatives and Transitivity, 88-120. Amsterdam: John Benjamins. 1391 Haspelmath, Martin. 2011. On S, A, P, T, and R as comparative concepts for alignment typology. 1392 Linguistic Typology 15, 535-567. 1393 Hopper, Paul J. & Elizabeth Closs Traugott. 2003. Grammaticalization (2nd ed.). Cambridge: 1394 Cambridge University Press. 1395 Hyman, Larry M. 1999. The historical interpretation of vowel harmony in Bantu. In J.-M. Hombert & 1396 L.M. Hyman (eds.) Bantu Historical Linguistics: Theoretical and Empirical Perspectives, 235-295. 1397 California: CSLI.

37

Draft for comments (2015) - Do not cite without author’s permission

1398 Hyman, Larry M. 2007. Niger-Congo verb extensions: Overview and discussion. In D.L. Payne & J. 1399 Peña (eds.) Selected Proceedings of the 37th Annual Conference on African Linguistics, 149-163. 1400 Sommerville, MA: Cascadilla Proceedings Project. 1401 Kähler-Meyer, E. 1966. Die örtliche Funktion der Applikativendung in Bantusprachen. In J. Lukas 1402 (ed.) Neue afrikanistische Studien: Festschrift für A. Klingenheben, 126-136. Hamburg: Deutsches 1403 Institut für Afrika-Forschung. 1404 Kallulli, Dalina. 2007. Rethinking the passive/anticausative distinction. Linguistic Inquiry 38(4), 770- 1405 780. 1406 Kawalya, Deo, Koen Bostoen & Gilles-Maurice de Schryver. 2014. Diachronic semantics of the 1407 -sóból- in Luganda. International Journal of Corpus Linguistics 19/1, 60-93. 1408 Kawasha, Boniface. 2007. Passivization in Lunda. Journal of African Languages and Linguistics, 37- 1409 56. 1410 Keyser, Samual Jay & Thomas Roeper. 1984. On the middle and ergative constructions in English. 1411 Linguistic Inquiry 15(3), 381-416. 1412 Khumalo, Langa. 2009. The passive and stative construction in Ndebele: A comparative analysis. 1413 Nordic Journal of African Studies 18(2), 154-174. 1414 Kimenyi, Alexandre. 1988. Passives in Kinyarwanda. In M. Shibatani (ed.) Passive and Voice, 355- 1415 386. Amsterdam: John Benjamins. 1416 Kiso, Andrea. 2012. Tense and aspect in Chichewa, Citumbuka and Cisena: A description and 1417 comparison of the tense-aspect systems in three southeastern Bantu languages. Stockholm: 1418 Stockholm University. PhD dissertation. 1419 Kitillä, Seppo. 2006. The anomaly of the verb ‘give’ explained by its high (formal and semantic) 1420 transitivity. Linguistics 44(3), 569-612. 1421 Kotz, Ernst. 1909. Grammatik des Chasu in Deutsch-Ostafrika (Pare-Gebirge). Berlin: Georg Reimer. 1422 Kula, Nancy C. & Lutz Marten.2008. Argument structure and agency in Bemba Passives. In K. Legère 1423 & C. Thronell (eds.) Bantu Languages: Analyses, Description and Theory, 115-130. Cologne: 1424 Rüdiger Köppe. 1425 Kulikov, Leonid. 1998. Passive, Anticausative and classification of verbs: The case of Vedic. In 1426 Leonid Kulikov & Heinz Vater (eds.) Typology of Verbal Categories: Papers presented to 1427 Vladimir Nedjalkov on the occasion of his 70th birthday, 139-153. Tübingen: Max Niemeyer 1428 Verlag. 1429 Kulikov, Leonid. 2013. Middle and reflexive. In S. Luraghi & C. Parodi (eds.) The Bloomsbury 1430 Companion to Syntax, 261-280. London: Bloomsbury. 1431 Kuperus, Juliana. 1985. The Londo word. Its phonological and morphological structure. Brussels: 1432 Royal Museum for Central Africa. 1433 Langacker, Ronald W. 2001. The English present tense. and Linguistics 5, 251-272. 1434 Levin, Beth & Malka Rappaport Hovav. 1994. A preliminary analysis of causative verbs in English. 1435 Lingua 92, 35-77. 1436 Lichtenberk, Frantisek. 2000. Reciprocals without reflexives. In Z. Frajzyngier & T.S. Curl (eds.) 1437 Reciprocals: Forms and Functions, 161-178. Amsterdam/Philadelphia: John Benjamins Publishing 1438 Company. 1439 Maho, Jouni Filip. 2009. NUGL Online: The online version of the New Updated Guthrie List, a 1440 referential classification of the Bantu languages. Online document. 1441 Marten, Lutz & Nancy C. Kula. 2014. Benefactive and substitutive applicatives in Bemba. Journal of 1442 African Languages and Linguistics 35(1), 1-44. 1443 Marten, Lutz. 2003. The dynamics of Bantu applied verbs: An analysis at the syntax-pragmatics 1444 interface. In K.K. Lébikaza (ed.) Actes du 3e congrès mondial de linguistique africaine (Lomé 1445 2000), 207-221. Cologne: Rüdiger Köppe.

38

Draft for comments (2015) - Do not cite without author’s permission

1446 Marten, Lutz. 2011. Bemba benefactive constructions in the typology of applicatives. In Peter Austin, 1447 Oliver Bond, Lutz Marten & David Nathan (eds.) Proceedings of the 3rd conference on language 1448 description and theory, 183-192. London: SOAS. 1449 Maslova, Elena. 2000. Reciprocals and set construal. In Z. Frajzyngier & T. Walker (eds.) 1450 Reciprocals: forms and functions (volume 2), 161-178. Amsterdam: John Benjamins. 1451 Maslova, Elena. 2007. Reciprocal and polyadic (Remarkable reciprocals in Bantu). In V.P. Nedjalkov 1452 (ed.) Reciprocal Constructions (Volume 1), 335-352. Amsterdam/Philadelphia: John Benjamins. 1453 Matlock, Teenie. 1989. Metaphor and the grammaticalization of evidentials. Proceedings of the 1454 Fifteenth Annual Meeting of the Berkeley Linguistics Society, 215-225. 1455 Mchombo, Sam A. & Armindo S.A. Ngunga. 1994. The syntax and semantics of the reciprocal 1456 construction in Ciyao. Linguistic Analysis 24, 3-31. 1457 Mchombo, Sam A. & R.M. Ngalande. 1980. Reciprocal verbs in Chichewa: a case for lexical 1458 derivation. Bulletin of the School of Oriental and African Studies 43, 570-575. 1459 Mchombo, Sam A. 1990. Reciprocalization in Chichewa: A lexical account. Linguistic Analysis 21, 3- 1460 22. 1461 Mchombo, Sam A. 1993. A formal analysis of the stative construction in Bantu. Journal of African 1462 Languages and Linguistics 14, 5-28. 1463 Mchombo, Sam A. 1999. Quantification and verb morphology: The case of reciprocals in African 1464 languages. Linguistic Analysis 29(1-2), 182-213. 1465 Ménard, F. 1908. Grammaire . Alger: Imprimerie des Missionaires d'Afrique. 1466 Mischke, Gerda E. 1994. Neuter verbal extensions in Southern Sotho. South African journal of African 1467 languages 14(2), 78-82. 1468 Morrison, Michelle E. 2011. A reference grammar of Bena. Houston: Rice University. PhD 1469 dissertation. 1470 Mudzingwa, Calisto. 2008. The reciprocal and associative in Shona. Journal of Literature, Language 1471 and Linguistics 2(1). Online document. 1472 URL: http://www.scientificjournals.org/journals2008/articles/1322.pdf. 1473 Mugane, John M. 1999. The recalcitrant nature of the Bantu verbal morpheme -an-. Linguistic 1474 Analysis 29(1-2), 160-181. 1475 Muriungu, Peter Kinyua. 2008. Phrasal movement inside Bantu verbs: Deriving affix scope and order 1476 in Kîîtharaka. Tromsø: University of Tromsø (Center for Advanced Study in Theoretical 1477 Linguistics). PhD dissertation. 1478 Nakamura, Masanori. 1997. Object extraction in Bantu applicatives: Some implications for 1479 minimalism. Linguistic Inquiry 28(2), 252-280. 1480 Ndayiragije, Juvénal. 2002. Théories linguistiques et réciprocité en Chichewa: la leçon de Kirundi. In 1481 P. Sauzets & A. Zribi-Hertz (eds.) Typologie des langues d’Afrique et universaux de la grammaire, 1482 169-210. Paris: L’Harmattan. 1483 Nedjalkov, Vladimir P & Sergej Je. Jaxontov. 1988. The typology of resultative constructions. In V.P. 1484 Nedjalkov (ed.) Typology of resultative constructions, 3-62. Amsterdam: John Benjamins. 1485 Ngonyani, Deo & Peter Githinji. 2006. The asymmetric nature of Bantu applicative constructions. 1486 Lingua 116, 31-63. 1487 Ngonyani, Deo. 1998a. Properties of applied objects in Kiswahili and Kindendeule. Studies in African 1488 Linguistics 27(1), 67-96. 1489 Ngonyani, Deo. 1998b. Towards a typology of applicatives in Bantu. In I. Maddieson and T. 1490 Hinnebusch (eds.) Language History and Linguistic Description in Africa, 248-257. Trenton: 1491 Africa World Press. 1492 Ngonyani, Deo. 2000. The constituent structure of Kindendule applicatives. In V. Carstens & V. 1493 Parkinson (eds.) Advances in African Linguistics, 61-76. Trenton: Africa World Press.

39

Draft for comments (2015) - Do not cite without author’s permission

1494 Noël, R. 1935. Éléments de grammaire Kibemba. Luapula: Préfecture du Luapula supérieur des rév. 1495 pères salésiens de Saint Jean Bosco. 1496 Nuyts, Jan. 2001. Subjectivity as an evidential dimension in epistemic modal expressions. Journal of 1497 Pragmatics 33, 383-400. 1498 Ondo-Mebiame, Pierre. 2007. Verbal suffixes of derivation in Fang-Ntumu. South African Journal of 1499 African Languages 3, 117-127. 1500 Ondo-Mebiame, Pierre. 2007. Verbal suffixes of derivation in Fang-Ntumu. South African Journal of 1501 African Languages 3, 117-127. 1502 Palmer, Frank R. 1977. Modals and actuality. Journal of Linguistics 13(1), 1-23. 1503 Palmer, Frank R. 2001. Mood and modality (2nd ed.). Cambridge: Cambridge University Press. 1504 Port, Robert F. 1981. The applied suffix in Swahili. Studies in African Linguistics 12(1), 71-84. 1505 Ranger, A. Sidney B. 1928. Chinsenga handbook: A manual of the Nsenga language, spoken in the 1506 protectorate of Northern Rhodesia. London: Sheldon Press. 1507 Schadeberg, Thilo C. 1995. Spirantization and the 7-to-5 vowel merger in Bantu. In M. Dominicy & 1508 D. Demolin (eds.) Sound Change, 73-84. Amsterdam/Philadelphia: John Benjamins. 1509 Schadeberg, Thilo C. 2004. Kutendekana: socio-stative verbs in Swahili. Unpublished article. 1510 Schadeberg, Thilo C. 2006. Derivation. In D. Nurse and G. Philippson (eds.) The Bantu Languages, 1511 71-89. London/New York: Routledge. 1512 Seidl, Amanda & Alexis Dimitriadis. 2002. Statives and reciprocal morphology in Swahili. In P. 1513 Sauzets & A. Zribi-Hertz (eds.) Typologie des langues d’Afrique et universaux de la grammaire, 1514 239-284. Paris: L’Harmattan. 1515 Shibatani, Masayoshi. 1985. Passives and related constructions: A prototype analysis. Language 61(4), 1516 821-848. 1517 Shibatani, Masayoshi. 2006. On the conceptual framework for voice phenomena. Linguistics 44(2), 1518 217-269. 1519 Song, Jae Jung. 1995. Reviews of Bernard Comrie and Maria Polinksy (eds.) Causatives and 1520 Transitivity. Amsterdam/Philadelphia: John Benjamins, 1993. Lingua 97, 211-232. 1521 Stappers, Leo. 1973. Esquisse de la langue mituku. Tervuren: Royal Museum for Central Africa. 1522 Stappers, Léon. 1967. Het passief suffix -u- in de Bantoe-talen. Africana Linguistica 3, 139-145. 1523 Sweetser, Eve. 1990. From etymology to pragmatics: Metaphorical and cultural aspects of semantic 1524 structure. Cambridge: Cambridge University press. 1525 Talmy, Leonard. 1985. Force dynamics in language and thought. In W. Eilfort, P. Kroeber & K. 1526 Peterson (eds.) Papers from the 21st Regional Meeting of the Chicago Linguistic Society, 293-337. 1527 Chicago: University of Chicago. 1528 Talmy, Leonard. 1988. Force dynamics in language and cognition. Cognitive Science 12, 49-100. 1529 Talmy, Leonard. 2000. Toward a Cognitive Semantics, vol. 1: Concept Structuring Systems. 1530 Cambridge: MIT Press. 1531 Taylor, Charles. 1985. Nkore-Kiga. Sydney: Croom Helm. 1532 Traugott, Elizabeth Closs. 1989. On the rise of epistemic meanings in English: An example of 1533 subjectification in semantic change. Language 65(1), 31-35. 1534 Trithart, Lee. 1979. Topicality: An alternative to the relational view of Bantu passive. Studies in 1535 African Linguistics 10(1), 1-46. 1536 Trousdale, Graeme & Muriel Norde. 2013. Degrammaticalization and constructionalization: two case 1537 studies. Language Sciences 36, 32-46. 1538 Tsunoda, Tasaku. 1985. Remarks on transitivity. Journal of Linguistics 21(2), 385-396. 1539 URL: http://goto.glocalnet.net/mahopapers/nuglonline.pdf. 1540 Usoniene, Aurelia. 1999. Perception verbs revisited. Working Papers 47, 211-225.

40

Draft for comments (2015) - Do not cite without author’s permission

1541 Van Den Eynde, Karel. 1960. Fonologie en morfologie van het Cokwe. Leuven: Leuven University. 1542 PhD dissertation. 1543 Van Eeden, B.I.C. 1941. Inleiding tot die Studie van Suid-Sotho (Die "Sesoeto" van Basoetoland en 1544 dele van die O.V.S.). Stellenbosch: Pro-Ecclesia-Drukkery. 1545 Vandermeiren, J. 1912. Grammaire de la langue kiluba-hemba telle qu'elle est parlée par les Baluba 1546 de l'est (Katanga). Brussels: Ministère des colonies. 1547 Viberg, Åke. 1984. The verbs of perception: a typological study. In B. Buttersworth, B. Comrie & Ö. 1548 Dahl (eds.) Explanations for Language Universals, 123-162. Berlin: Mouton de Gruyter. 1549 Voeltz, Erhard.1977. Proto Niger-Congo extensions. Los Angeles: University of California. PhD 1550 dissertation. 1551 Westphal, E.O.J., J.R. Masiea, S.M. Tindleni, H.M. Jimba, I.V. Mzileni & M.T. Matiela. 1974. The 1552 verbal extensions in : A descriptive and comparative classification. 1553 Bulletin of the School of Oriental and African Studies, University of London 37(1), 213-222. 1554 Whiteley, W.H. & M.G. Muli. 1962. Practical introduction to Kamba. London: Oxford University 1555 Press. 1556 Whiteley, W.H. 1966. A study of Yao sentences. Oxford: Clarendon Press. 1557 Wierzbicka, Anna. 1994. Semantics and epistemology: The meaning of ‘evidentials’ in a cross- 1558 linguistic perspective. Language Science 16(1), 81-137. 1559 Willems, Dominique & Bart Defrancq. 2000. L’attribut de l’objet et les verbes de perception. Langue 1560 français 127, 6-20. 1561 Willems, Dominique. 1983. “Regarde voir” Les verbes de perception visuelle et la complementation 1562 verbale. In E. Roegiest & L. Tasmowski (eds.) Romanica Gandensia XX: Verbe et phrase dans les 1563 langues romanes, Mélanges offerts à Louis Mourin, 147-158. Ghent: George Michiels. 1564 Willems, Dominique. Forthcoming. Les constructions verbales en co(n)texte et en contraste: le cas du 1565 verbe observer. Linguistica Romanica. Stockholm: Stockholm University press. 1566 Woolford, Ellen. 1994. Why passive can block object marking. Paper presented at the 25th Annual 1567 Conference on African Linguistics. Online document. 1568 URL: 1569 Zeller, Jochen & Paul Ngoboka. 2006. Kinyarwanda locative applicatives and the minimal link 1570 condition. Southern African Linguistics and Applied Language Studies 24, 101-124.

41

Draft for comments (2015) - Do not cite without author’s permission

1571 Appendix: Language sample 1572 Baati 1610 Karanga 1648 Nyoro 1573 Bafut 1611 Kete 1649 Nyungwe-Tete 1574 Bangubangu 1612 Kiga 1650 Nzadi 1575 Bankon 1613 Kikuyu 1651 Ombo 1576 Basaá 1614 Kiluba 1652 Rangi 1577 Bemba 1615 Kituba 1653 Ronga 1578 Bembe 1616 Kpāʔ 1654 Rundi 1579 Bena 1617 Kwezo 1655 Rutooro 1580 Béo 1618 Lega 1656 Rwanda 1581 Boa 1619 1657 Sanga 1582 Bobangi 1620 Lokelé 1658 Senga 1583 Bolia 1621 Lomongo (Nkundo) 1659 Sesotho 1584 Bongili 1622 Londo 1660 Shambala 1585 Bukusu 1623 Luba 1661 Shi 1586 Bulu 1624 Lulua 1662 Shona 1587 Bushoong 1625 Lwena 1663 Sikongo 1588 Chaga (Mochi-dialect) 1626 Mabiha 1664 So 1589 Changana-Tsonga 1627 Makaa 1665 Songe 1590 Chasu 1628 Makhuwa 1666 Sotho 1591 Chewa 1629 Mambwe 1667 Sukuma 1592 Chokwe 1630 Mbesa 1668 Swahili 1593 Doko 1631 Mbomitaba 1669 Swazi 1594 Duala 1632 Mfinu 1670 Talinga 1595 Ekoti 1633 Mituku 1671 Tetela 1596 Enga 1634 Mòkpè (Bakweri) 1672 Tofoké 1597 Eton 1635 Mwera 1673 Tonga 1598 Ewondo 1636 Ndebele 1674 Tsogo 1599 Fang 1637 Ndibu 1675 Tswana 1600 Ganda 1638 Nkengo 1676 Túnen 1601 Gusii 1639 Ǹƞwɛ̄ 1677 1602 Haya 1640 Nsongo 1678 Venda 1603 Herero 1641 Ntomba 1679 Wisa 1604 Holoholo 1642 Nyakyusa 1680 Woyo 1605 Hungu 1643 Nyamwezi 1681 Xhosa 1606 Isangu 1644 Nyanja 1682 Yao 1607 Iyembe 1645 Nyankore 1683 Yombe 1608 Kamba 1646 Nyanyembe 1684 Zombo 1609 Kami 1647 Nyiha 1685 Zulu

42