Mojeño Trinitario, an Arawak language of Bolivia, has a set of 27 classifiers that express an abstract meaning, generally physical properties like shape. They are found in multiple environments (on numerals, nouns, adjectives and in verbs), as is typical of “multiple classifier systems” commonly found in the Arawak family (Aikhenvald 2000). The focus here is on verbal classifiers. They are not obligatory but frequent: searching four narrative texts, classifiers were present in 38% of the sentences, and half of them were verbal classifiers. These verbal classifiers categorize either the S of an intransitive verb or the P of a transitive verb, as is typical for verbal classifiers, or more surprisingly, an adjunct. Following Contini-Morava and Kilarski’s (2013) review of the functions of classifiers, the paper examines the following three discourse functions: reference management, referent identification, and re-presentation of referents. The data examined consist of 8 hours of natural, first-hand textual data. Reference management Payne (1987) suggests that "the primary function of a noun classification system may be related to discourse level participant reference." A recurrent pattern presented in the literature is the following: a noun phrase is used to introduce a referent for the first time, and then a classifier is enough to mention the old information again (Mithun 1986, Seifart 2005). The pattern observed in Mojeño Trinitario narratives differs. First, to foreground a referent, verbal classifiers co-occur with a coreferential noun phrase. This can be done either to introduce a new topic as in (1) or to mention again as a topic a participant that was ‘given’ earlier in the text as in (2). This foregrounding effect led to the reanalysis as an applicative construction of some classifier construction involving an adjunct (Author 2018). (1) ta-ni-k-'o to 'santi 3NH-burn-CLF:rect.vol-ACT ART.NH field ‘It (the fire) burns the field (it burns the forest to create a space for a field in it).’ 12.009 [‘field-to-be’ never mentioned before] (2) te to une t-ewts-e-k-wo-m=po eto to uupi-ono PREP ART.NH water 3-throw-CLF:liquid-ACT-MID-PL=PFV 3NH ART.NH frog-PL ‘The frogs jumped into the water.’ 18.054 [ten sentences after previous mention of the river] Second, because noun phrases can be omitted for discourse reasons, a verbal classifier can occur without a coreferential noun phrase. This is very common in Mojeño Trinitario: out of the 70 verbal classifiers to occur in four narrative texts, 49 occur without a noun phrase. This pattern has the function discussed by Mithun of mentioning a participant that is ‘old information’ (‘anaphoric use’) as in (3), but it also often serves to introduce of a new non-topical participant as in (4). (3) na-nok=po to muiji, n(a)-oktaya-ji-k=po 3PL-put-ACT=PFV ART.NH straw 3PL-step_on-CLF:shapeless-ACT-PFV ‘They put the straw, they step on it.’ 20.026 (4) t-mopku-mo 3-be.dark-CLF:fabric ‘It (the sky) was dark.’ 19.53 Referent identification When the referent is not overtly expressed, the classifier participates in referent retrieval in different manners. First, the referent is most often inferred from the background of the general situation rendered by the preceding stretch of discourse (anaphora). In some cases, the interpretation of the referent is based on metonymy (often meronymy): the referent is retrieved through bridging, or indirect anaphora (for instance, -si ‘round’ is interpreted as ‘the head’ when an animate participant is ‘given’). Second, the referent can be retrieved either as a ‘kind’ of entity, defined by the characteristics typically associated with the classifier (e.g. -e ‘liquid’ is often interpreted as ‘a body of water’), or as a prototypical member of the category. In a translation task of classifiers without coreferential noun phrase, consultants generally suggest a prototypical member of the class that fits well the verb event. Third, the retrieval of the referent can be exophoric, i.e. located deictically in the speech event, as in (5). (5) v-eja-pue-gi-a 1pl-sit-CLF:ground-ACT-IRR ‘Let’s sit on the ground.’ 24.099 [‘ground’ identified deictically as the space under the feet of the speech act participants] Re-presentation of referents In a stretch of discourse, different classifiers can track the same referent, showing either a change of properties of the referent, or a change in perspective by the speaker. Excerpts of texts will illustrate each sub-function. This paper provides a rare comprehensive account of the discourse functions of verbal classifiers in an individual language. The discourse functions of Mojeño Trinitario verbal classifiers will be discussed and illustrated with a detailed description of examples in their textual context.