The DiscourseFunction ofObject MarkinginSwahili AmandaSeidl andAlexisDimitriadis UniversityofPennsylvania 1. Introduction TheSwahili objectmarker(OM)is averbal prefixthat agrees withan objectofthe verb. In Swahili there is no semantic or lexical class of objects for which object marking is obligatory, nor is there any class for which it is impossible. The nu- merous earlier studies of the object marker have discovered no hard and fast rules for its distribution. Its usage has been found to correlate with a number of fac- tors, which we can divide into semantically/grammatically oriented factors, espe- ciallyanimacy(Wald1979,HymanandDuranti1982,Keach1995),anddiscourse- oriented factors such as topichood, specificity or “salience” (Givon 1976, Hyman andDuranti1982,Allan1983). Butsuchcorrelationsareinvariablytendencies,not categoricalfacts. Colloquiallyspeaking,then, useof theSwahili OM is “optional”. But such a characterization must be forever suspect: are we really observing unpredictable variation, or are we simply failing to take into account the right variable, or vari- ables,thatcontroltheuseoftheOM?Untilandunlesswecanhaveanunderstand- ing of the role of genuine optionality in language, we can only hope to settle the questionbyidentifyingacategory(orseveralcategories),grammaticalordiscourse- oriented,that theOM canbesaidtoconsistentlydenote. OurstudyoftheSwahili OM isundertakeninthis spirit. Webase ouranal- ysisonacorpusstudyofwrittenSwahilitextsselectedfromtheHelsinkiCorpus of Swahili. The extent of variability found in the use of the OM, and the fact that in- tuitions about when to object mark an NP are highly variable across speakers, led us to believethat the conditions determining the use of the OM must be pragmatic in nature. As we show in section 4.1, Prince’s (1992) notion of Hearer Status is successfulindefiningconditionsunderwhichtheOMisnot used;buttheconverse doesnothold,andtheOMisunaccountablyabsentinmanycasesthatmeetthecon- ditionswe have identified. The relationshipof animacyto object markingcan then beconsideredanindirecteffect ofotherfactorsthattendtocorrelatewithanimacy, suchassalienceindiscourse. The optionality of the OM also calls into question its typological status: is the Swahili OM an inflectional morpheme, or a separate, cliticized element? Mor- phologically,it appears between the verb root and the tense morpheme,suggesting that it must be inflectional; but we do not expect an inflectional category to be optional, and the Swahili OM has other properties that are more reminiscent of a pronoun than of an inflectional element. In section 5 we argue for a reanalysis of theSwahiliverbthatreconciles themorphologicalevidencewithananalysisofthe OMas apronoun. 1.1. Methodology Ourcorpus study was performed on material from the Helsinki Corpus of Swahili, an online corpus located on the Helsinki University Language Corpus Server, and administeredbythe Department of Asian and African Studies. The corpus is com- prisedofacollectionofbookswrittenbyauthorswhospeakoneofseveraldifferent dialects of Swahili and come from different Swahili speaking countries. Swahili is notafirst languageforallofthesewriters,but allacquireditat anearlyage. CLS33:TheMainSession(1997):373-389 c 1997bytheChicagoLinguisticSociety.Allrightsreserved. (cid:13) 374 AMANDASEIDL &ALEXISDIMITRIADIS Wecodedatotalof23randomlyselectedpassages,eachcontaining31sen- tences. Sentences with multiple objects were excluded from the data presented here. We coded for the following variables: presence of the object marker; an- imacy; whether a full-NP object is present; Information Status (in the sense of Prince1992);andwhethertheobjectNPis referential orpredicative. Objects that were coded as animate included animals and humans, but ex- cluded anything human that could not move; corpses, body parts, etc... The noun class of the object was ignored, forexampleif an animate was in a typicallyinani- mate noun class, e.g., ki-toto ‘class7-small child’ (cf. the canonical m-toto‘class1- child’), the object was still coded as animate. The remainder of the objects were codedas inanimate. Codingwasbasedonourinterpretation(asreaders)ofthepassagesinques- tion. In particular, the objects of verbs without overtfull-NP objects were deduced on thebasis of ourunderstandingof the text; andthe hearer-status of verbs usedin conversations, which is defined in terms of the hearer’s knowledge, was assigned onthebasisofareader’sunderstandingofwhatthehearerintheconversationknew ordidnotknow. Only sentences with transitive main verbs were coded. Verbs were consid- ered transitive if they met any of the following criteria: first, if there was an overt objecteitherfollowingorprecedingtheverb;second,iftherewasanobjectmarker in the verb; or third, if in English, the verb could not appear without an object (which would show that an object is not semantically required). For example the verb eat can be used intransitivelyin English,hence object-less uses of it were not coded. Thisselectionmethodwasintendedtoincludeinourstudynullobjectsthat arenotindicatedbyanagreementmarker. Itturnsoutthatsuchcasesarepractically nonexistentinwrittenSwahili (seesection2). Sentences at discourse segment boundaries or the beginning of new dis- coursesegmentswerenotcoded,sincereferenceacrossdiscourseboundariesisnot well-understood and creates additional difficulties. Because we were only inter- estedinunmarkeduses ofthe object marker, all syntacticallyspecializedconstruc- tions; Object Relatives, Reflexives, Left-Dislocations, Topicalizations, and Right- Dislocations were excluded from the final analysis. Cases of Discourse Deixis (in thesenseofWebber(1990),i.e.,event-levelreference) werealsoexcluded. The statistics we discuss are based on analysis of 312 coded sentences that metalloftheabovecriteria. Thiscountbreaks downfairlyevenlyforanimacy,ob- ject marking, and information status: 124 sentences were object marked, vs. 188 whichwerenot;144objectswereanimate,and168werenot;and175objectswere codedas HearerOld,vs.137thatwereHearerNew. 2. The grammarofSwahili objectmarking Swahili or KiSwahili is a Bantu language spoken in much of East Africa as a lin- guafrancaandistheMotherTonguealongmuchofcoastalTanzaniaandKenyaas wellas onseveralislands(Zanzibar,Pemba,Mafia,Lamu,etc.). AccordingtoEth- nologue (Grimes 1996), Swahili is spoken by some five million speakers as a first language,andbyanotherthirtymillionas asecondlanguage. Swahili verbs carry morphemes that agree with their subject and object in noun class. The agreement markers include informationabout number, person and noun class. The noun classes correspond to some degree to semantic classes; an- imate (class 1/2), tall-thin things (3/4), man-made things (7/8), loan words (9/10), etc... ThebasicstructureoftheSwahili verbisas follows: THEDISCOURSE FUNCTION OFOBJECTMARKINGINSWAHILI 375 (1) SM Tense(RM)(OM)VerbRoot (DS )FinalVowel (cid:3) Ni–li– ku– on– a 1sg Past 2sg see Indic (SM =subjectmarker; RM =relativemarker;DS =derivationalsuffixes) ‘I sawyou.’ The “final vowel” can carry mood information: in the imperative and the subjunc- tive, a special final vowel appears and there is no tense morpheme. Reflexives are expressedthrougha reflexivemorpheme,which appears inthe Object Markerslot. Verbs can carry a number of derivational suffixes which must appear in a specific order after the verb root. The derivational suffixes include morphemes expressing causative,applicative,passive,reciprocal,and“stative”relations. Unlike object marking, subject agreement is almost always mandatory in Swahiliforfiniteverbs,i.e.,allfiniteverbscarry asubjectagreement morpheme. 1 The object marker, on the other hand, is not always present. A transitive verbwithalexicalobjectNPmayormaynotcarryobjectagreement. Object mark- ingispossiblewitheverylexicalclassofobjects, butisnotrequiredbyanylexical 2 classofobjects;sometypesofobjects(suchaspropernamesandanimates)areob- jectmarkedmorefrequentlythanothers,butexceptionsineitherdirectionarequite common. The examples in (2) show overtly expressed animate objects with and withoutobject marking;thosein(3) showthesamefrom inanimates. (Inexamples with more than one verb, the verb whose object marking is of interest is shown in boldface). (2) a. +animate+OM +FullNP Ni-li-mw-uliza Helenakamamtu a-li-bisha hodi. SM-Pst-OM-askHelenaif person SM-Pst-knockknock.knock ‘IaskedHelenaifsomeonehadknockedonthedoor.’ b. +animate–OM+FullNP WakatihuuRosaa-li-hitaji watu wa ku-m-tuliza. time this RosaSM-Pst-needpeople PREP to-OM-comfort ‘Atthat timeRosaneeded someonetocare forher.’ c. +animate–OM+FullNP A-li-ulizaBosibaadayaku-ingiandani,... SM-T-ask boss after of to-enter inside ‘Aftergoinginside,heasked theboss...’ (3) a. –animate–OM +FullNP A-li-kata kambana ku-kimbia. SM-Pst-cutrope and Inf-run ‘It [thedog]cut theropeandranfree.’ b. –animate+OM +FullNP Tulia-li-ya-amini manenohayo. TuliSM-Pst-OM-believewords these ‘Tulibelievedthesewords.’ Theobjectmarkercanalsodenoteanargument thatisnot expressedbya fullNP: (4) a. +animate+OM –FullNP Yulemtua-li-kataa ku-funuliwa. Albert a-li-m-sikia a-ki-lia. that manSM-Pst-refusebe-uncovered Albert SM-Pst-OM-hearSM-Ipf-cry ‘Thatmanrefusedtoshowhisface. Albert heard himcrying.’ 376 AMANDASEIDL &ALEXISDIMITRIADIS b. –animate+OM –FullNP Manenohaya a-li-ya-sema kwa sauti kubwa.Rosaa-li-ya-sikia. words theseSM-Pst-OM-saywithvoicebig RosaSM-Pst-OM-hear ‘Hesaidthesewordsveryloudly. Rosaheardthem.’ Finally,itispossiblefortheobjectofatransitiveverbnottobeexpressedovertlyat all. Wefoundthisconstructiontobeveryrareinourcorpusofwrittentexts. In our data,we countedonlytwosentences as containingbotha nullobject andnoagree- ment (thebreakdownof object markingand overtobjects is giveninTable1); they both involveduse of the verb kushukuru ‘thank’. We believethat the object of this verbissimplyoptional(beingusedlike“givethanks”),asitisinEnglishverbslike eat orask,whichalsooccurwithneitherovertobjectnoragreementinSwahili. As Nicolle (1996) notes, however, the use of transitive verbs with neither OM nor an NPobject iscommoninspokenSwahili. We reproducetwoofhisexamples: (5) a. A:Je,unawatoto? ‘So,doyouhavechildren?’ B: Bado. ‘Not yet.’ A:U-ta-pata. you-Fut-get ‘Youwillget[them].’ b. A:U-me-letachakula? ‘Haveyoubrought(the)food?’ B:Ni-me-leta Bwana. I-Prf-bring sir ‘I havebrought[it],sir.’ A verb can agree either with its direct or its indirect object. Swahili, like all Bantu languages, does not grammatically mark the distinction between direct and indirect objects in any way. Swahili allows only one object marker per verb, while some Bantu languages, like Haya and Kivungo-Chaga, allow more. When there is more than one object, the OM, if present, is generally restricted to refer- ringtoaspecificchoiceamongtheavailableobjects,accordingtoacomplexset of considerations(cf. BresnanandMoshi1990,Rugemalira1993). It shouldbe addedthat therearesomeconstructionsthatrequire orprohibit theuseoftheobject marker. TheOMis prohibitedwithpassiveverbs,as shown in example(6a). In writtenSwahili, the OM is obligatorywhen therelativemarker is present, as inexample(6b). Since ourintent was tofocus on theproperties ofsyn- tacticallyoptionalusesoftheOM,weexcludedsuchconstructionsfromourcorpus study. (6) a. Rosaa-li-(*ki)-p-ew-a kitabu. RosaSM-Pst-OM-give-Pass-Indicbook ‘Rosawas giventhebook.’ Table1: Object Marking Presence ofFull-NPObject (cid:2) +FullNP –FullNP OM 65 59 +OM 186 2 (cid:0) THEDISCOURSE FUNCTION OFOBJECTMARKINGINSWAHILI 377 b. Kitabua-li-cho-ki-nunua ki-me-potea book SM-P-REL-OM-buySM-Perf-be.lost ‘Thebookthatheboughthasbeenlost.’ Proper names in Swahili tend to be object marked, even when inanimate. Keach (1995) claims that object-marking of proper names is obligatory, and provides ex- ample (7), which involves the mandatory object-marking of an inanimate object. (Sentence (7b)showstheunacceptabilityofanimatemarkingfor Lucille). (7) a. Ni-li-li-kamata“Lucille”(BB King’sguitar). I-Pst-OM-grab Lucille ‘I grabbedLucille.’ b. *Ni-li-m-kamata“Lucille” c. *Ni-li-kamata“Lucille” (from Keach1995) However, we found that our corpus contained uses of proper names that werenotobject marked,suchasexample(8a). Weconcludethatobjectmarkingof propernamesis,infact,optional. (8) a. Rosaa-li-sikia Stellaa-ki-zungumza na chakulamdomo-ni. RosaSM-Pst-hearStellaSM-Ipf-talkingwithfood mouth-in ‘RosaheardStellatalkingwithhermouthfull offood’. Investigations into object marking have long noticed several frequency ef- fects, in particular, the high frequency of co-occurrence of the object marker with certain semantic classes of nouns. Animate objects in Swahili are object marked more often than inanimates, a fact that has motivated numerous analyses of the Swahili OM as a marker of animacy (Hinnebusch and Kirsner 1980, Vitale 1981, HymanandDuranti1982,Allan1983,Keach1995). Objectmarkinghasalsobeen claimed to denote definiteness (Givon 1976, Hinnebusch and Kirsner 1980), stress (Vitale 1981), givenness or distinctiveness (Allan 1983), or a hierarchical combi- nation of factors including theta role, combined person-animacy, and specificity (HymanandDuranti1982). 3. Informationstatus While discourse considerations are of interest in characterizing the uses of the OM, the notions we discussed above—definiteness, specificity, determinedness— correspondonlylooselywiththepresenceorabsenceofobjectmarking. Ourstudy found that Prince’s (1992) notion of Information status, which we summarize in thissection,ismoredirectlycorrelatedwiththeuseornon-useoftheOM. Informationstatusisaclassificationofwhatisknownaboutanentityatthe point of a particular mentionof it in discourse. Prince (1992) defines the Informa- tionStatus of an entityaccording totwocriteria: whethertheentityis Hearer-Old, that is,already knowntothehearer, andwhetheritis DiscourseOld,i.e.,whether 3 it has been previously mentioned in the discourse. The possible combinations of these factors are shown in Table 2. (Since a hearer is assumed to remember what Table2: Informationstatuses,afterPrince (1992) Discourse-new Discourse-old Hearer-new Brand-New — Hearer-old Unused Evoked 378 AMANDASEIDL &ALEXISDIMITRIADIS has beensaid,thereare noDiscourseOld,HearerNewentities). In the following examples, the indefinite a kid is understood as referring to an entity unfamiliar to the hearer before this point (a Brand-New entity), while the definite in (b) is understood as referring to a kid that the hearer already knows about, and is able to identify. (An Unused, or perhaps Evoked entity). Sentence (10)isanotherexampleinvolvinganUnusedentity. (9) a. Intheparkyesterday, akid threwuponme. (Brand New) b. Intheparkyesterday, thekid threwuponme. (EvokedorUnused) (10) ThePresident returnedtoWashingtontoday. (Unused) PrincedefinestheadditionalcategoryofInferrables,whichareentitieswhose existence is inferable, on the basis of shared knowledge and beliefs, from Hearer- Old entities. Inferrables are technically Hearer-New and Discourse-New, but they have special properties, for example, cross-linguistically they are marked as defi- nite(Gundeletal.1993). Example(11b)containstheInferrable thedoor: sincewe know that the Bastille is a building, we understand that there is an associated door even though no door has been mentioned. Example (11c) is ill-formed, since we havenoreasontoinfera trunkwhentheBastilleis mentioned. (11) a. HepassedbythedooroftheBastilleandthedoor was paintedpurple. b. HepassedbytheBastilleandthedoor was paintedpurple. c. #HepassedbytheBastilleandthetrunk was paintedpurple. 4. The corpus study 4.1. Informationstatus Table 3 shows our findings concerning the relationship of information status with object marking. It can be seen that Brand New and Inferrable entities are rarely or never object marked, while Unused and Evoked entities are object marked in 50 per cent or more of the sentences studied. The categories Unused and Evoked have in common the property of being Hearer Old, in contrast to Brand New and Inferrable, which are Hearer New. In other words (abstracting away from the few exceptions),theOMcan beusedwithHearerOld,butnotHearerNewentities;but it would be inappropriate to say that the OM is a marker of hearer-oldness, since onlyabout62%oftheHearerOldobjectsinoursamplewereobject-markedbythe verb. Rather, we have a negativeconstraint: unfamiliar objects may not be object- marked. 4.2. Animacy Space does not permit a review of the voluminous literature on the relationship of animacy and object marking. We mention Wald (1979) and Bentley (1994), who are among several authors that identify animacy as the primary motivationfor ob- ject marking. Wald also notes that object marking an inanimate draws attention to it, and concludes that “discourse distinctiveness” as well as animacy and defi- niteness cause a speaker to use the OM. Bentley appeals to a generalization of the Table3: Object Marking InformationStatus(all objects) (cid:2) (counts) (percentages) HO HN HO HN UN E BN IN UN E BN IN OM 7 109 2 6 % OM 47 68 2 12 +OM 8 51 84 45 % +OM 53 32 98 88 (cid:0) (cid:0) THEDISCOURSE FUNCTION OFOBJECTMARKINGINSWAHILI 379 OM from pure marker of animate agreement (“cross-reference”) to contrastive or emphaticuses. Our study examined the correlation of animacy with object marking. The followingpassagefromourcorpusillustratestherelevanceofinformationstatusto theobjectmarkingofanimates: theindefiniteanimatembuzi‘goat’isBrandNewin sentence (a), andis not object marked, but is Evokedinsentence (c) andis accord- inglyobject-marked. The correlation is not perfect: the inanimate kamba ‘rope’ in (b),whichhappens tobeEvoked,is notobjectmarked. (12) a. Mbwaa-li-ona mbuzi. Dog SM-Pst-seegoat ‘Thedogsawagoat.’ b. A-li-kata kamba na ku-kimbia. SM-Pst-cutrope andto-run ‘It cuttheropeandranfree.’ c. A-li-m-rarua mbuzi vipandevipande. SM-Pst-OM-tear.apartgoat part part ‘It torethegoattopieces.’ Table 4 shows the correlation of object marking with animacy; to facilitate comparison,Table5presentsobjectmarkingvs.HearerStatusintheformofatwo bytwotable. ItcanbeseenthatHearerStatusprovidesastronger,thoughnotover- whelmingly stronger, correlation. Neither animacy nor Hearer-Oldness guarantee object marking. Object marking of inanimates is rare, but more likely (12%) than thatofHearerNewentities(6%). Wehavesuggestedthatthecorrelationofanimacywithobjectmarkingmay beanindirecteffectofthedifferentdiscourserolestypicallytakenbyanimatesand inanimates. (Cf. Bentley1994:47fffora discussionoftheconnectionbetweenani- macy and discourse role). If object marking was solely determinedby information status, then animates and inanimates with the same information status would be equallylikelytobeobjectmarked. Suchisnotthecase,however: restrictingourat- tentiontoFull-NPobjects (inordertoabstract from theeffects offull-NPpresence or absence), we found that Evoked animate objects are significantly more likelyto beobject markedthaninanimates,as showninTable6. (The othercategorieswere pretty close). It is clear that we are still short of a complete characterization of the impetusbehindobjectmarking. 4.3. Definiteness and theOM A nearly universal claim about the object marker (adopted in some form by every source we reviewed) is that it is a marker of definiteness, or specificity. But on examination, characterizations along these lines turn out to be untenable. First of Table4: Object Marking Animacy(allobjects) (cid:2) animate inanimate OM 104 20 +OM 40 148 (cid:0) Table5: Object Marking HearerStatus(all objects) (cid:2) Hearer-old Hearer-new OM 116 8 +OM 59 129 (cid:0) 380 AMANDASEIDL &ALEXISDIMITRIADIS Table6: Object Marking Animacy,Evokedfull-NPobjects (cid:2) Animate Inanimate OM 43 7 +OM 13 36 (cid:0) all, only half the Evoked (and therefore definite) entities in our survey were actu- ally marked (see Table 3). But it is not even correct to say that object agreement 4 is an optional marker of definiteness. “Definiteness” in this context is a semantic notion based on a generalization of syntactic definiteness. Since Swahili does not have determiners, we must characterize definiteness on the basis of semantic crite- ria alone, by analogy with other languages. Cross-linguistically, definites may be Evoked, Unused, of Inferrable in their Information Status (cf. Gundel et al. 1993, and section 3, and examples (9)). In particular, if the OM is a marker of definite- ness it should be freely used with Inferrables; but as can be seen from Table 3, Inferrables are only rarely object marked. (12 per cent of Inferrables were object marked,comparedto66percent ofall Hearer-Oldentities). Since object marking distinguishes Evoked from Inferrable entities but the category “definite” does not, we must conclude that the OM cannot be a markerof definiteness. Insection5wewillargue that thedistributionoftheOM is similarto thetypicalcross-linguisticdistributionofpronouns. 4.4. Full NPobjects As we showed earlierin Table 1, the OM is virtuallyobligatorywhen no overt full NPobject ispresent. SincetheOMitselfcanbeconsideredtobetheobject,we do not talk of object drop, but of the presence or absence of an overt nominal object. Thisphenomenonissomewhatclarifiedwhenweconsidertheinformationstatusof the objects. As Table 7 shows, all non-NP sentences involve Evoked objects. (All but 2 of the –FullNP objects were object marked, as shown in Table 1). Although it is not surprising that Brand New objects must be explicitlynamed, the complete lackofnon-FullNPInferrableorUnusedobjectsisunexpected. Clearly,thispattern ismorerestrictedthanwouldbeexpectedbyfunctionalconsiderationsalone. If we accept that null objects (with no object marking) are nonexistent in writtenSwahili,wereachthefollowingconclusions: theOMisrestrictedtotheen- vironments that allow pronominalization cross-linguistically, and in particular, In- ferrableobjectscannotbeexpressedthroughtheOM.ButtheOMdoesnot replace thefullNPobject,thatit,thepresenceoftheOMisnotsufficienttolicenselackof a full NP object; for this more stringent conditions apply, namely, a non-overt NP objectseems toberestrictedtoEvokedobjects. SincewewishtolimitourattentiontotheoptionalusesoftheOM,wepro- videinTable8abreakdownofinformationstatusbyobjectmarkingforonlythose sentences that have overt full-NP objects. It can be seen that the results reported abovedonotchangequalitatively. TheproportionofobjectmarkedHearerOlden- tities (Unused and Evoked) nowbecomes about fifty per cent forboth types, while HearerNewentitiesare notnormallyobjectmarked. Table7: Full NPobjects InformationStatus (cid:2) UN E BN IN +FullNP 15 99 86 51 –FullNP 0 61 0 0 THEDISCOURSE FUNCTION OFOBJECTMARKINGINSWAHILI 381 Table8: Object Marking InformationStatus(full-NPobjects) (cid:2) (counts) (percentages) HO HN HO HN UN E BN IN UN E BN IN OM 7 50 2 6 % OM 47 51 2 12 +OM 8 49 84 45 % +OM 53 49 98 88 (cid:0) (cid:0) 5. Inflection orpronoun? Object agreement can, but need not, occur along with a full nominal object. For this reason the literature frequently refers to it as being a “pronoun” or “clitic” when no overt object is present, and an “agreement marker” when an object is also present (e.g., Ashton (1944), and Allan (1983) both characterize the OM in these terms). Other accounts variously argue that the OM is always a pronoun (Givon1976,HymanandDuranti1982,Wald1979);thatitisalwaysanagreement marker (Vitale 1981); or that it is one of the above for animate and yet another for inanimate objects (Keach 1995). It appears, however, that the duality implied by thisnomenclatureisterminologicalratherthanreal. Itmustbestressedthatthereis no morphologicaldifference between object “pronouns” and “agreement markers” attachedtotheverb. Inallcases,theobjectmarkerappearsbeforetheverbroot,but following the subject and tense prefixes; and its form is always dependent on the nounclass oftheobject,andneveronwhetherafullNP objectispresent. The distinction, then, is based on function, not form. The terms “pro- noun” and “agreement” are used in a functional sense by a number of authors (Givon 1976, Wald 1979, Allan 1983), so that characterization of the OM as a pronounisbydefinitiondependentonthepresenceorabsenceofafullnominalob- ject. Given such a definition, the statement that the OM is a “pronoun” when used alone and an “agreement marker” when used with a lexical object is tautological; thetheoretical issueiswhethertheOM canbeusedas an“agreement marker,”i.e., whetherobjectmarked“objects”arereallyinobjectposition,ratherthanbeingina dislocatedposition(“topics”). In this paper we use a morphological criterion. We are concerned with the distinctionbetweeninflectionalmorphemes,whicharemorphologicallypart ofthe verb (and which we consider as not being arguments of the verb), and pronouns, whichwetaketobeseparatelexicalentitiesthatmaybecomecliticizedontoaverb. Is the Swahili OM an inflectional morpheme, an incorporated pronoun, or some- timesoneandsometimestheother? Previousstudiesofobjectmarkinghavereachedvaryingconclusions. Givon (1976)considerstheSwahiliOMtohaveoriginatedasapronoun,itsuselatergen- eralized to the agreement function, while Bentley (1994) argues that the Swahili OM is an inflectional element (a marker of animacy)that has acquired pronominal functions. In this paper we will steer clear of issues of diachronic derivation; syn- chronically, however, there are several good reasons to treat the Swahili OM as an incorporated pronoun, rather than as an inflectional morpheme. From a syntactic pointofviewtheOMis“optional,”whilemorphosyntacticfeaturesareinprinciple marked obligatorily. (E.g., English nouns must be marked for number). Second, theOMcancarrycontrastivestress,asinexample(13). Again,thisisapropertyof words,not ofpieces ofinflection. (13) Ni-li-k´ı-nunua kitabu! SM-PAST-OM-buy book ‘Idid buyit!’ 382 AMANDASEIDL &ALEXISDIMITRIADIS Finally, in section 4.3 we showed that the OM, unlike markers of definiteness cross-linguistically, does not tend to be used with Inferrable entities. Pronouns also have certain cross-linguistic requirements on their use: they refer to entities that are already highly salient in the discourse. Heim’s (1982:386) Prominence Condition notes that anaphoric pronouns should have antecedents that are “promi- nent” in status at that point in the discourse, generally by having been explicitly mentioned. Inferrableentities,forthemostpart,cannotbe pronominalized,as the 5 ill-formednessofEnglishexample(14)shows. 6 (14) I went toMary’shouse. I knockedonthedoor/#it. The Swahili OM is used only rarely with Inferrables, as Table 3 showed. If we interpret the data to mean that the OM is not typically used with Inferrables, we can summarizethe properties of pronouns,definites, and the OM as inTable 9. It can be seen that the distribution of the Swahili OM resembles the typical cross- linguisticdistributionofpronouns. Table9: SummaryofthedistributionoftheOM BN E IN Definites – + + Pronouns – + – OM – + – Considernowthepossibility(suggestedbythefunctionaldefinitionof“pro- noun”) that the OM is a pronoun when it occurs alone, and an inflectional affix when it co-occurs with an NP object. In its pronominal incarnation, the OM might be expected to be subject to the Prominence Condition as discussed above. As an agreement marker, on the other hand, the OM ought to be sensitive to syntactic andsemanticproperties of theagreed-withobject,suchas definiteness oranimacy, but the object’s place in the discourse should be irrelevant. But our data shows no evidence of such dual nature in the object marker. Table 8 shows that the OM is unlikelytooccurwithInferrable entitieseveninthepresenceofanovertobject. Ontheotherhand,thepositionoftheOMargues infavorofanalyzingitas inflection: The OM appears between the verb root and the tense and subject mark- ers. If the OM is a cliticizedpronoun, the principleof lexical integrityimplies that the tense and subject agreement morphemes would have to be clitics as well. But sincetenseandsubjectmarkingareobligatoryandhenceinvariablytakentobe part oftheverb,thepositionoftheOMleadsustoconcludethattheobjectmarkerisan inflectionalaffix. Although the OM behaves like a pronoun, the fact of its morphological lo- cation is too conspicuous to disregard. We were able to resolve this impasse by takingintoaccount the workof Barrett-Keach (1986),who argued that the Swahili verb has a bipartite internal structure, with a break between the tense marker and the OM. Barrett-Keach’s interest was in arguing for the existence of an Aux (or Infl) projection, but for our purposes the important point is that the posited struc- tureimpliesthattheOMisattachedperipherallytothesecondpartoftheverb,and hencecanbeanalyzedasacliticizedpronounwithoutviolatingtheprincipleoflex- icalintegrity. AccordingtoBarrett-Keach(1986),theclusterofsubjectmarkerplus tenseformsanincorporatedauxiliary,withawordbreakseparatingitfrom theOM, whichis attachedtotheverbproperas illustratedhere: (15) Ni-li + zi-andika. SM-Pst+ OM-write ‘I wrotethem(theletters).’