When “ You ” and “ i ” mess around with the hierarchy : a comparative study of Tupi-Guarani hierarchical indexing systems

this paper deals with the person indexing system of tupi-Guarani languages. Past literature has claimed that the relative position of the arguments of a transitive verb on a supposed person hierarchy 1 > 2 > 3 determines what argument is marked on the verb and how. it is also commonly believed that the morphosyntax of individual tupi-Guarani languages is very comparable. this paper surveys in detail the encoding of arguments on transitive verbs in 28 tupi-Guarani languages. it shows that the prior assumptions about indexing in tupi-Guarani languages either do not hold strongly, or need to be stated in more nuanced ways. the study also shows that these languages are not as similar morphosyntactically as is often assumed. importantly, they display a great variation in the domain of local configurations (i.e., when the two speech act participants interact), the arguments of which are often encoded in a non-transparent manner. this leads us to reject the 1 > 2 hierarchy as operative in governing indexing in all languages of the group.

iNTRoDUCTioN the tupi-Guarani branch of the tupi family is a major language group of south America.it comprises around forty languages that are considered very similar morphosyntactically while very dispersed geographically (Jensen, 1999).their typologically most frequently discussed feature is a hierarchical person indexing system, where the relative position of the arguments of an independent transitive verb on a supposed person hierarchy 1 > 2 > 3 determines what argument is marked on the verb and how 1 .the seminal analysis of hierarchical systems in tupi languages by Monserrat and soares (1983) was used as a model in subsequent works, among others the often cited reconstruction by Jensen (1990).A simplified version of the description of the hierarchical system was then repeated over time in individual descriptions and typological studies (such as Payne, 1994).While the main goal of the original work by Monserrat and soares was to stress the variation found in the 'local' configurations (when first and second person interact) within the family, this variationist perspective has long been forgotten and the differences from the idealized system are often underemphasized in both individual language descriptions and comparative works 2 .the fair number of recent descriptions of tupi-Guarani languages and the typological interest on local configurations (Heath, 1998;Zuniga, 2008;Junker, 2011) led us to undergo a new comparative study on that topic.
the terminology in table 1 follows typological practice when studying the effect of the person value of the arguments of a transitive clause on their encoding.
Most typological studies on person indexing present a hierarchy 1 > 2 > 3 (dixon, 1994;Givon, 2001).the basis for this hierarchy is the assumption that speakers are optimally interested in themselves, then in their interlocutors, then in any other person or object.it is nevertheless typologically common that the 1 > 2 > 3 hierarchy works clearly with respect to mixed configuration (3 ↔ 1, 2) but less so for local configurations when the two speech act participants (sAPs) are involved (1 ↔ 2) (Zúniga, 2006).As a consequence, the hierarchy between the sAPs and third person is universally accepted (1, 2 > 3) while the hierarchy between the two sAPs is debatable.some authors consider that first and second persons are not universally hierarchized, their relative order fluctuating from one language to the other (silverstein, 1976; delancey, 1981).More rarely, other authors claim that the universal hierarchy is 2 > 1 (Junker, 2011).there is no universally-valid functional motivation for the ranking of the sAPs on the hierarchy.Local configurations indeed constitute a domain where pragmatics play a major role. in many languages, therefore, first or second person pronominals are replaced in discourse by impersonal, third-person or plural forms such as the french vous, the spanish usted or the German Sie instead of a transparent second person singular pronominal tu/tú/du.throughout this paper, i will refer to a study by Heath (1998).Heath argues that transparent person indexation combining markers for both first and second persons in transitive configurations are avoided in many languages.A transparent person indexation for a transitive configuration is when the person markers for the two arguments of the transitive clause mixed configuration 3 ↔ 1, 2 local configuration 1 ↔ 2 non-local configuration 3 ↔ 3 table 1. transitive configurations according to the person value of the arguments 3 .
1 1 > 2 reads '1 st person is higher than 2 nd person on a grammaticalized person hierarchy'.When the number is not specified, both singular and plural are implied. 2the idealized system is often just used as a starting point for the description of individual systems, and then the divergences with it are merely stated but rarely discussed. 31 ↔ 2 reads '1 st and 2 nd person are interacting'.are overt, occur in separate morphological slots and do not interact (Heath, 1998, p. 84)."in other words, maximally transparent 'i saw you', 'you saw me ', etc. […] are often replaced by some opaque surface forms" (Heath, 1998(Heath, , p. 84). siewierska (2004) ) relates the pragmatically sensitive transitive situation involving two speech-act participants to the notion of 'face threatening act' in the politeness theory (Brown;Levinson, 1987).Heath provides a list of twelve different strategies to avoid transparent combinations of first and second persons, based on data from diverse Australian and native American languages (table 2).He briefly discusses some Guarani data, identifying strategies 3, 4 and 104 .non-transparent person indexing in tupi-Guarani languages was also previously explicitly linked to politeness rules, such as the avoidance of face threatening acts, in studies on Emerillon (rose, 2002; 2003b; 2011).the present study identifies in the tupi-Guarani language group seven out of Heath's twelve strategies (in bold in table 2) used to avoid maximal transparency in the encoding of 1 ↔ 2. it further suggests an additional 13 th strategy.the specific operation of these strategies in the tupi-Guarani languages is discussed from the third to the sixth sections of this article, which challenge the various assumptions.
Based on empirical examination of situations involving the two speech-act participants, this paper will conclude that we need to revise the analysis of hierarchical systems within the tupi-Guarani group, and more specifically the earlier-proposed 1 > 2 > 3 hierarchy.this study is based on data from 28 tupi-Guarani languages representative of the eight subgroups.they are in bold in table 3 with their reference sources. in the paper, language names will be cited with their subgroup number in roman numerals.the comparative data will be used to question the grammaticalized relative ranking of sAPs on the person hierarchy for the argument indexing system of the tupi-Guarani languages.After briefly presenting the 'idealized' tupi-Guarani hierarchical system in the next section, this paper will underline the variation among individual languages and the many discrepancies with the 'idealized' model (third to sixth sections of this article, challenging the various assumptions).it will more precisely focus on the local configurations.At the end of the article, we will develop two conclusive ideas: first, there is no unique hierarchical system within the tupi-Guarani language group, but much variation (section: Conclusion 1); second, the person hierarchy really involved in these systems is reduced to a 1, 2 > 3 hierarchy (section: Conclusion 2).
THE 'iDEALiZED' TUPi-GUARANi HiERARCHiCAL SYSTEM since silverstein's pioneering work, it is known that hierarchies of features can play a major role in argument encoding systems (silverstein, 1976).this author highlighted the role of semantic properties of nominals on case-marking and agreement (more specifically in the domain of ergative or split-ergative systems).the term 'person hierarchy' used in the present paper corresponds roughly to designations for such hierarchies used by others, such as 'empathy hierarchy' (delancey, 1981), 'referential or inherent topicality hierarchy' (Givón, 1994), and 'indexability hierarchy' (Bickel;nichols, 2007).• Tembé / Guajajára / turiwára (tenetehara) (Bendor-samuel, 1972;Harrison, 1986;duarte, 2005) V • the fundamental idea behind the invocation of such hierarchies is iconicity: the more referential/topical/animate or semantically salient a participant is, the more likely it will have access to morphosyntactic slots.A first explicit definition of indexing systems entirely based on such hierarchies is nichols (1992, p. 66), who says that in hierarchical systems: "Access to inflectional slots for subject and/or object is based on person, number, and/or animacy rather than (or no less than) on syntactic relations". in practice, this means that the participant that is higher on the hierarchy is favored over the lower one8 .the person hierarchy that is often considered relevant for hierarchical systems is 1 > 2 > 3 (siewierska, 2004).However, the literature on several language families rather posit a 1, 2 > 3 hierarchy, because 1 st and 2 nd persons cannot be hierarchized in a simple manner (see for example Macaulay, 2009, on Algonquian;Gildea, 2012, on Cariban).the argument indexing system on independent transitive verbs in tupi languages (including the tupi-Guarani branch) constitutes a telling example of the role of a person hierarchy (Monserrat;soares, 1983;Jensen, 1998).this section summarizes the usual presentation of this system as found in Proto-tupi-Guarani reconstructions (Jensen, 1990;schleicher, 1998) or in most individual descriptions.this analysis is repeated (often in a simplified version) in typological studies (dixon, 1994, p. 107;Payne, 1994;Payne, 1997, p. 214).the debate on the definition of such systems as inverse is left aside here (for a discussion see Payne, 1994;rose, 2009).table 4 lists four assumptions found in the above-mentioned sources with respect to the hierarchical argument encoding system.
the tupi-Guarani hierarchical system claimed to exist in independent transitive verbs is described below.A more detailed description can be found in Jensen (1990).there is a sole person slot on the verb; this slot precedes the stem and it is obligatory filled.there are two sets of person markers that qualify for it, called set i and set ii after Jensen's (1990)  Jensen reconstructs set i as prefixes and set ii as pronominal words (realized as pronouns, clitics or prefixes depending on the languages) 11 .their distribution is summarized in table 6. set i marks A (as well as s A on active intransitive verbs).set ii marks P (as well as s P on stative intransitive predicates, possessor of nouns and object of postpositions).the person value of set i and set ii forms on transitive verbs is unambiguous given their person value on other root classes.
the central point of the prior claims about the system is that the participant that is higher in the 1 > 2 > 3 person hierarchy is the one that systematically gets access to the unique index slot on the verb. in mixed configurations (involving a third person and an sAP), the sAP is therefore predicted to always be indexed on the verb.if the participant to be encoded is the A, it is indexed by set i, as shown in example (1).if it is the P, it is indexed by set ii, as in example (2). in non-local configurations (3 → 3), the person hierarchy is irrelevant.the A argument is systematically indexed on the verb (3)12 , whatever the other semantic characteristics of the two arguments might be, and their relative topicality.the encoding of the two local configurations (sAP → sAP) is more complex.When a second person acts on a first person (2 → 1), whatever their number, it is said that the first person patient is indexed by virtue of being higher on the person hierarchy (4) 13 .When a first person (whatever its number) acts on a second person (1 → 2 ), Jensen (1990) says that a special set of markers (set iV) is used.this set iV consists of portmanteau forms indexing the person value of both A and P (whatever the number of A). they are reconstructed as in Proto-tupi-Guarani and a few descendant languages, a third person P is additionally systematically marked by a set ii prefix following the set i prefix for A; see the reconstructed forms (marked by '*') in examples ( 7) and ( 8).Proto-tupi-Guarani (Jensen, 1998, p. 518-522) (7 1Sg.i-3P-like 3.i-3P-like 'i like him/her/them/it.' 'He/she/they/it like(s) him/her/them/it.' the four assumptions presented above are rarely challenged in the literature 14 .Admittedly, Monserrat and soares (1983) do consider Assumption 2 a 'leak' in their proposed hierarchical system.Monserrat and soares identify five types of encoding for the 1 → 2 configuration, on the basis of a small number of languages.the portmanteau analysis presented above for the configuration 1 → 2 also acknowledges that the 1 > 2 hierarchy is not operative in this configuration 15 .the fundamental idea behind a portmanteau analysis is that the form encodes a whole configuration and not one argument over the other.Consequently, portmanteau forms do not support any hierarchy.the hierarchy 1 > 2 therefore applies only to the configuration 2 → 1, to which some authors add the configuration 1Pl → 2Sg (Monserrat;soares, 1983;Payne, 1994;seki, 2000).this will be discussed in the section 'Challenging the General Assumption'.the present paper will now re-consider the suggested 1 > 2 hierarchy by challenging each of the four assumptions of table 4, in the respective sections below.
Before that, it must be noted that 23 languages among the 28 tupi-Guarani languages of the study are described as displaying a hierarchical system, whereas five are not: Achê i, nheengatu iii, Kokama iii and omagua iii, and Urubu-Ka'apor Viii. it is generally considered that these five languages lost the hierarchical indexing system reconstructed for Proto-tupi-Guarani.this is quite plausible due to the specific genesis of each of these languages (Cabral, 1995;Jensen, 1998, p. 497, citing a personal communication from Aryon rodrigues;roessler, 2008;da Cruz, 2011).these five languages will not be discussed in the remaining sections.

CHALLENGiNG ASSUMPTioN 1: MULTiPLE iNDEXiNG
Assumption 1 -There is only one slot for person indexing on Tupi-Guarani verbs the literature on tupi-Guarani languages often stresses the fact that only one argument is marked with an index on the verb, even with transitive verbs.indexes include affixes, clitics and weak pronouns 16 , and are distinct from free pronouns that are "morphologically and syntactically independent expressions of person" (siewierska, 2003).this section reviews three counterexamples to the 'only one index' rule (first three lines of table 7).A fourth possible counterexample is discussed, in the use of some special pronominal forms that could also be considered person indexes (last line of table 7). 14An exception is dietrich (2001, p. 30-31), who considers that whenever the patient is 1 st or 2 nd person, the syntax of the clause is like that of a predicate nominal clause.this means that the word including the verb root and its person prefix should be analyzed as an existential predicate, comparable with expressions of possession.dietrich's analysis precisely does not explain the local configurations on which this paper focuses.nominal predicates indeed do not show the same variation and forms in person encoding when sAPs interact as those on verbs, examined in this paper. 15the hierarchy 1 > 2 could be supported only if the supposed portmanteaus were analyzed as A markers (as partially done in seki, 2000, on Kamaiurá; dooley, 2006, on Mbya). 16see note 20.
A first counterexample to Assumption 1 is the above-mentioned marking of a third person object by a set ii prefix *i-~ *ts-17 between the root and the set i prefix in some languages (for instance in Mbyá i and tupinambá iii) as well as in the reconstruction suggested by Jensen (1990).the verb shows a succession of two obligatory person markers.
(9) 1 → 3 (Mbyá i, dooley, 2006, p. 18) (10) 1 → 3 (tupinambá iii, Jensen, 1999) A possible second counterexample to Assumption 1 lies in the reconsideration of the analysis of the so-called portmanteau forms used in the local configuration 1 → 2Pl as a sequence of two morphemes.the bi-morphemic analysis is developed in the section challenging Assumption 2.
A third potential counterexample to Assumption 1 is very marginal.tapieté i normally follows the regular pattern for 2 → 1 but some speakers use a instead a sequence of set i and set ii markers referring respectively to A and P (11).finally, some specific pronominal forms are used in the local configuration 2 → 1 that could disprove Assumption 1 as well. in the local configuration 2 → 1, the first person P is indexed on the verb as expected, since it is higher on the person hierarchy.But in seven languages (see last line of table 7) a special pronominal form also follows the verb (14).it is unknown to what extent Assumption 1 holds for these languages given the lack of information about these pronominal forms.i suspect that Assumption 1 has the effect of downgrading the importance of the reconstructed pronominal forms *epe and *peyepe, which are unexpectedly almost never presented in the tables with the other Bi-morphemic analysis of so-called "portmanteaus" Emerillon Viii, Guajajara iV, Kaiwá i, Kamaiurá Vii, siriono ii, tapieté i, tapirapé iV, tupinambá iii two person indexes for 2 → 1 tapieté i (variant) special pronominal forms for 2 → 1 Asurini do tocantins iV, Guajajara iV, Kayabi Vi, tapirapé Vii, tupinambá iii, Emerillon Viii, Wayampi Viii pronominal paradigms18 .these special pronominal forms, often neglected in the descriptions, differ from the regular free pronouns presented in table 5. they are probably considered to be "pronouns", i.e. independent words, due to their lack of phonological or morphological interaction with the verb.from the scarce data presented, it nevertheless seems that these pronominal forms could be considered as post-verbal indexes, due to their placement right after the verb and their (apparent) obligatoriness 19 .More information is needed to evaluate whether they could be considered to be 'weak pronouns'20 , i.e. a type of person index.the variation in these special pronominal forms will be discussed in the following section.they could maybe be considered to be post-verbal person indexes used additionally to the pre-verbal ones.
(12) 2Pl → 1Sg (Kayabi Vi, dobson, 1997, p. 53) 1Sg.ii hit 2Pl 'You all hit me.' there are thus at least three (or four) reasons to nuance Assumption 1, claiming that there is only one slot for person indexing on tupi-Guarani verbs.While in most cases there is only one obligatory prefix/clitic on the verb, some configurations call for two morphemes.
Strategy 6-Entire combination expressed by an unanalyzable portmanteau form.the reflexes of *oro-for 1 → 2Sg are strikingly homonymous with the first person exclusive set i markers in fourteen languages, i.e. all the languages of table 8 except the three languages at the bottom of the table.
Economy leads to analyzing *oro-as a transparent A marker in the configuration 1eXcl → 2Sg (13).P is left unexpressed, as it also is in the mixed configuration 1eXcl → 3 where A is higher than P on the person hierarchy (14).indeed, there is no formal distinction between these two configurations in most languages.Another argument for analyzing *oro-for 1eXcl → 2Sg as an A marker is that no r-relational marker is present in (13) as it should be if *orowas a set ii marker (15).the so-called "relational marker" is indeed found on some verbal roots when preceded by a P argument expressed either as a nominal phrase preceding the verb or as a sAP set ii prefix.the use of these two meaning extensions suppose number neutralization.Using a plural form for a singular is a common strategy in the expression of speech act participants, illustrating Heath's (1998) Strategy 4. it is, for example, at work in the polite use of vous '2Pl' for a singular addressee in french, a strategy called 'plurification' in Heine and song's paper on the development of personal pronouns (Heine;song, 2011).Using an inclusive form to refer to two local arguments is another pattern attested cross-linguistically (Heath's Strategy 8).Yuki ii possibly uses a comparable strategy for 1 → 2Pl. in Yuki ii, this configuration is expressed with ya-, a form identical with the inclusive marker (Villafañe, 2004, p. 209).
Strategy 4number neutralization, sometimes including use of Pl for semantic Sg.
Strategy 8inclusive marker replaces 1 st or 2 nd marker, or entire combination.in three languages (among which the two members of subgroup ii), the forms used for 1 → 2Sg are less easily derivable from *oro-; these are siriono ii are-~ ane-, Yuki ii are-and a variant of tapieté i arɨ-~ andɨ-for 1Sg → 2Sg. in these three languages the pronominal encoding can plausibly be analyzed as two morphemes a-'1Sg.i'plus re-~ nethat could be formally related to the second singular set ii index (see table 5).Assuming that *oro-is the correct reconstruction for 1 → 2Sg, the commonalities in 1 → 2Sg marking for these three Bolivian tupi-Guarani languages could be considered an innovative pattern via the reanalysis of *oro-as a more transparent encoding for 1 → 2Sg.tapieté i additionally shows a specific and transparent encoding for the 1eXcl→ 2 configuration.this construction shows multiple indexing, with both a 2 nd person set ii prefix for P (also encoding number) and the -ha eXcl suffix for A (see note 22).'We (eXcl) take you (Sg).' now, regarding the forms for 1 → 2Pl, it is striking that the earlier reconstruction *opo- (Jensen, 1990, p. 120) does not easily account for the synchronic forms of the supposed portmanteaus for 14 languages in the database (table 9).Moreover the forms used for 1 → 2Pl can be transparently parsed as two morphemes in many tupi-Guarani languages: *a-'1Sg.i'and *poro-'generic human' (20).the morpheme *poro-is used in many languages to refer to a generic human referent (21)23 .'He will come one day to raise all men (from the dead).'Cabral (2001) therefore proposes a bi-morphemic reconstruction, consisting of *a-1Sg.iand *poro-~ po-'generic human', for the configuration 1 → 2Pl, in which the morpheme for a generic human referent is used to refer to a second person plural referent.in line with Heath (1998), the use of *a-1Sg.ifor configurations with a first person plural agent can be explained as number neutralization (Strategy 4), while the use of the indefinite expression *poro-to refer to a second person plural argument corresponds to Strategy 5. the use of an indefinite pronoun (derived from a generic noun) as a second person plural marker has already been described as cross-linguistically common (Heine;song, 2011).
Strategy 4number neutralization, sometimes including use of Pl for semantic Sg.
Strategy 5-1 st or 2 nd marker merged with (or replaced by) 3 rd person marker.Cabral's (2001) analysis accounts for the synchronic encodings in 12 of the 14 languages that do not express either a first person A or a second person plural P transparently.table 9 shows that some languages use reflexes of the two morphemes; others use only a reflex of *poro (po-in subgroup i); and the relevant forms in yet two other languages show no link with *poro24 .
Among the languages that use only a reflex of *poro, the languages of subgroup i (Avá, Chiriguano, Guarani correntino, Jopara) show the reflex po-for 1 → 2Sg (22), and table 8 shows that they use the prefix ro-for 1 → 2Sg. the prefixes po-and ro-are considered P markers in the synchronic descriptions of those languages, but without specific arguments supporting the claims (dietrich, 1986;Kallfell, 2010).it is clear only in tupinambá that opo-has been reanalyzed as a set ii marker (Cabral, 2001, p. 135). it indeed shows the regular morphosyntactic distribution of set ii markers (see table 6). it is used as a possessive prefix on nouns for instance.table 9 shows that only Guajajara and Kaiwa have developed a distinction between 1Sg → 2Pl and 1eXcl → 2Pl with a transparent encoding of a first person exclusive A in the expression of 1eXcl → 2Pl. in this configuration, Strategy 4 on number neutralization does not apply.the reflex of *oro-1eXcl.i is used instead of *a-'1Sg.i'before *poro-.
this section offered a new reconstruction for 1 → 2 replacing the previous analysis that involved portmanteau forms (Assumption 2). the expression of 1 → 2Sg is reconstructed with *oro-1eXcl.i.this use of a 1eXcl.imarker can be explained as a replacement of a transparent combination of a first and a second person markers with a single inclusive marker (Strategy 8) and by neutralization of number since the 1eXcl.imarker can express a configuration with a first person singular A (Strategy 4). the expression of 1 → 2Pl is reconstructed with *a-'1Sg.i' and *porogen.huM.P. the use of *a-'1Sg.i' for configurations with a first person plural A can be explained as number neutralization (Strategy 4), while the use of the indefinite expression *poro-to refer to a second plural argument corresponds to Strategy 5.
Strategy 4number neutralization, sometimes including use of Pl for semantic Sg.
Strategy 5-1 st or 2 nd marker merged with (or replaced by) 3 rd person marker.Strategy 8inclusive marker replaces 1 st or 2 nd marker, or entire combination.

CHALLENGiNG ASSUMPTioN 3. oN 2 → 1
Assumption 3 -The 1 > 2 hierarchy is justified by first person indexing when 2 → 1 information on this configuration is lacking for Yuki ii. in all other languages of the study, except for Emerillon Viii and Wayampi Viii, the first person P argument is indexed on the verb with a set ii marker, while the second person A argument is expressed by a free pronoun (23) 26 .this construction is used regardless of the number of A and P.

Yuki ii
ya-25 table 9 does not include the languages which encode 1 → 2Pl with a clear reference to just one of A or P (Ava-Canoiero iV, Kayabi Vi, Anambé V, Araweté V, Guaja Viii and Xeta i), or in a similar way to how they code 1 → 2Sg (Asurini do tocantins iV, Mbyá i).A rare alternate form of a-poro-specifically for 1eXcl → 2Pl is oro-in Emerillon Viii. 26Whether this pronoun is optional or obligatory is not always specified in the sources.the absence of a pronoun here makes the interpretation of the sentence ambiguous, with either a second or third person A. only two languages do not show a set ii index on the verb for 2 → 1: Emerillon and Wayampi.these two close members of subgroup Viii have diverged from the rest of the tupi-Guarani languages.instead of P, A is indexed on the verb, in non-conformity with a 1 > 2 hierarchy.in the two languages, pronominal forms are found after the verb but they are not cognate with each other.the entire constructions are given in table 10. (the double second-person reference in two of the Emerillon examples will be discussed shortly.) in Wayampi Viii, the system is almost transparent, with both A and P explicitly expressed: A as a set i prefix on the verb, and P as a set ii prefix on an auxiliary.this auxiliary shows different forms depending on the number of P 27 .Person indexing on the verb is not conform with a 1 > 2 hierarchy, but the indexing on the auxiliary following the verb is.
in Emerillon Viii, the system is less transparent: both arguments are not always explicitly expressed.A is marked with a set i prefix on the verb, but the post-verbal pronominal forms do not systematically refer to P. the forms ereɲ and peɲ are special second person pronominal forms, and they can only be interpreted here as referring to A as the meanings of the relevant examples in table 10 are not reflexive but express a 2 → 1 configuration.these two pronominal forms are composed of a form similar to the set i prefixes ere-'2Sg' and pe-'2Pl' and a final ɲ formative, a continuous clitic elsewhere in the language.this means that in two of the four 2 → 1 patterns in Emerillon, A is double marked, and P is unexpressed.in previous works (rose, 2003a; rose, 2008; 2009), i hypothesized that this double-marking pattern could be explained by a change in sAP hierarchy from 1 > 2 in Proto-tupi-Guarani to 2 > 1 in Proto-Emerillon.the new hierarchy would have affected the pronominal prefix on the verb, replacing a 1 st person marker by a 2 nd person marker.some of the post-verbal pronominal forms for 2 nd person A would have resisted the change (i.e.been retained as marking A), leading to double-marking of A. it is by no means a transparent pattern but still works as a distinctive and unambiguous way to express a specific configuration: there is no explicit and separate expression of both A and P, but double-marking is unequivocally associated with a 2 → 1 configuration.this constitutes a strategy to avoid transparent marking of local configurations that was not listed in Heath's (1998) work.i am suggesting an additional strategy, that i label Strategy 13.
suggested Strategy 13double marking of one of the arguments.the other one is unexpressed.now going back to the languages that index P on the verb for 2 → 1, it was mentioned earlier that five of these languages belonging to three different subgroups use special pronominal forms in this situation (tupinambá iii, Kayabi Vi, tapirapé iV, Guajajara iV, and Asurini do tocantins iV).these pronominal forms are placed after the verb (24), while free pronouns are clause-initial.they do not all refer transparently to second person.this non-canonical person marking is reminiscent of Heath's (1998) Strategy 2 and Strategy 11 to avoid transparent combinations of first and second persons.
Strategy 2one of the two markers is expressed by isolated suppletive allomorph 28 .
Strategy 11-Co-occurring 1 st and 2 nd markers are widely separated 29 .
A pe formative at the end of the forms can be identified in the five languages.it is noticeably identical in form with the reflex of the dative/locative postposition *-pe in all these languages.in Guajajara iV and Asurini iV, this formative is invariably used for the four 2 → 1 configurations (i.e.2Sg → 1Sg, 2Sg → 1Pl, 2Pl → 1Sg, and 2Pl → 1Pl). in tupinambá iii, Kayabi Vi and tapirapé iV, this element is combined with a preceding formative that 28 see note 5. 29  A2Sg/Pl → P1eXcl arepe looks pronominal both in its function and its form.the pe or peye initial formative found in the configuration with a second person plural A corresponds to the 2Pl in set i and set ii (pe) or set iii (peye-) in the three languages (see table 5) 30 .Altogether, these second person plural special pronominal forms are very likely made of a second person marker and a dative/locative postposition.the initial formative found in configuration with a second person singular A are formally similar to the 1Sg set i prefix in Kayabi and to the 1Sg set ii prefix in tapirapé.they thus index a first singular argument, making a second reference to P in this construction for the 2Sg → 1 configuration.this double marking of P is also attested with the third pronominal form in tapirapé iV. in tapirapé iV, the pronouns corresponding to *jepe and *pejepe are restricted to with a first person singular P. they are illustrated in examples ( 25) and ( 26). in addition, an extra form arepe is used for 2Sg/Pl → 1eXcl (27). it is apparently built on are, the first person exclusive set ii marker plus the possible dative/locative -pe.this etymology can be explained by analogy with an analysis of xe in xepe as the first person singular set ii pronoun visible sentence-initially in example (25).
tapirapé iV (Praça, 2007, p there is consequently double marking of P and no marking of A in the clause when ape, xepe and arepe are used (for 2 → 1Sg, 2Sg → 1Sg and 2Sg/Pl → 1eXcl respectively) 31 .interestingly, tapirapé iV and Kayabi Vi show double marking of P in the same configuration than Emerillon Viii shows double marking of A, namely when P is first person, i.e. in a highly face-threatening situation for the speaker.these three languages illustrate Strategy 13, double marking of either A or P, for expressing local configurations in a non-transparent manner.this strategy can therefore combine with either 1 > 2 or 2 > 1 local hierarchy in synchrony.
suggested Strategy 13double marking of one of the arguments.the other one is unexpressed.in summary, this section has shown that Assumption 3, which states that the marking in 2 → 1 configurations supports the 1 > 2 hierarchy, is challenged by features of two languages: by the marking on the lexical verb in Wayampi (though not by the auxiliary verb marking), and by Emerillon.the section also showed much variation in special pronominal forms, illustrating three different non-transparent strategies in the encoding of local configurations. 30set ii 2Pl in Kayabi is pẽ. 31since the tapirapé three special pronominal forms do not unequivocally all refer to just A or just P, Praça (2007, p. 103-104) considers them as portmanteau pronouns.Praça's gloss for the final pronouns of examples ( 25) to ( 27) is 2Sg → 1Sg, 2Pl → 1Sg, 2 → 1eXcl.

CHALLENGiNG THE GENERAL ASSUMPTioN. DECoNSTRUCTiNG THE 1 > 2 HiERARCHY
General Assumption -Access to the sole person slot follows from a 1 > 2 > 3 hierarchy the present study underscores five arguments against a 1 > 2 hierarchy applicable to all tupi-Guarani hierarchical indexing systems, based on data for local configurations.they are summarized in table 12 and detailed below.the first argument against a 1 > 2 hierarchy for all tupi-Guarani languages considered as showing a hierarchical indexing system is that, among the languages that do not show reflexes of *oro-and *a-poro-, four languages in fact show no sAP hierarchy.these four languages, belonging to three different branches (Anambé V, Araweté V, Guajá Viii, Xeta i) index the second person P on the verb when 1 → 2 (28).this should fit a 2 > 1 hierarchy.However, P is also indexed in the other local configuration 2 → 1 (29).this supports the traditional 1 > 2 hierarchy as in the common system presented in the section on the 'idealized' tG hierarchical system.thus no over-arching person hierarchy can be posited between the speech act participants in these four languages since each local configuration supports a different hierarchy.to summarize, P is indexed on the verb and A can only be expressed by a regular free pronoun, in the two local configurations of these languages.their system thus favors P over A in terms of marking on the verb.
(28) 1Sg → 2Sg (Guajá Viii, Magalhães, 2007, p. 195 second, this analysis of P-marking outranking A-marking can be replicated for languages whose descriptions view the forms used for 1 → 2 simply as P markers (see 'Challenging Assumption 2'). it is unclear how the hierarchy 1 > 2 can be supported in these cases, since the second person is the one that gets encoded on the verb, not the first person.
third, following the reconstruction of 1 → 2 as *oro-and *a-poro offered in the section 'Challenging Assumption 2', 1 st person marking is favored only for 1 → 2Sg. the two arguments are treated equally for 1 → 2Pl. the 1 > 2 hierarchy thus applies only partially as far as the expression of 1 → 2 is concerned.Anambé do Cairari V, Araweté V, Asuriní do tocantins iV, Ava i, Chiriguano i, Emérillon Viii, Guajá Viii, Guarani correntino i, Jopara i, Kaiwá i, Kamaiurá Vii, Mbya-Guarani i, siriono ii, tapieté i, tapirapé iV, Guajajára iV, tupinambá iii, Wayampi of french Guiana Viii, Xeta i, Yuki ii, Zo'é Viii fourth, Emerillon Viii and Wayampi Viii do not even favor the 1 st person in the configuration 2 → 1 (Challenging Assumption 2), leaving no argument at all for operation of a 1 > 2 hierarchy in these two languages.
fifth, the possible analysis of local configurations in terms of a hierarchical indexing system is obviously blurred by the non-transparent use of morphology. in our study, only two languages out of the 23 with a hierarchical system show highly transparent marking of the 1 st person and the 2 nd person arguments for both 1 → 2 (30) and 2 → 1 (31) configurations.Ava-Canoeiro iV and Kayabi Vi systematically encode the 1 st person in the verb prefix slot in local configurations, while the 2 nd person is encoded by a free pronoun (apparently optional in Ava-Canoeiro; post-verbal in Kayabi).only in these two languages it is completely relevant and useful to introduce an explanation of marking in terms of the person hierarchy 1 > 2 in the local configurations.
(30) 1Sg → 2Sg (Ava-Canoeiro iV, Borges, 2006, p. 158 All other languages show a non-transparent marking when 1 → 2, which has led to widespread use of the term 'portmanteau' in the tupi-Guarani literature.tupi-Guarani data perfectly illustrate Heath's claim that languages avoid maximal transparency in the encoding of pragmatically sensitive combinations like the local configurations (Heath, 1998), and more specifically 1 → 2. this study has identified seven out of Heath's twelve strategies (in bold in table 2) used to avoid maximal transparency in the encoding of 1 ↔ 2 .the study has further suggested an additional 13 th strategy.this shows that the 1 > 2 hierarchy does not apply straightforwardly to all local configurations in all tupi-Guarani languages with a hierarchical indexing system.
the present section precisely listed five arguments against a straightforward hierarchy 1 > 2 in tupi-Guarani indexing systems (General Assumption).it can at best be stated that there is a partial preference for the first person over the second in most languages of the tupi-Guarani group on the basis of marking in the configuration 2 → 1. if the 1 > 2 hierarchy was considered to be at work, one should explain why it is inactive in some of the configurations it is relevant for.if the effect of this hierarchy must be limited to specific configurations, the explanation it provides is first of all not an economic analysis and second, not a powerful functional explanation for the overall indexing system32 .
CoNCLUSioN 1-VARiATioN this comparative study has shown that tupi-Guarani languages are not so similar morphosyntactically as is commonly asserted.Means of encoding local configurations vary a lot within the tupi-Guarani group.this paper does not postulate a strong correlation between particular formal means and particular language subgroups (either subgroups already identified or hypothetical alternative groupings).for example, the four languages systematically favoring P over A marking for local configurations belong to various subgroups: Anambé V, Araweté V, Guajá Viii, Xeta i. these four languages behave in a rather simple and homogenous way while other languages show great variation in their non-transparent encoding of the local configurations.on the one hand, the homogeneity in the four languages could well be due to common inheritance, and the variation in encoding in other languages would be explained as distinct innovations that each blur the transparent encoding of the local configurations.on the other hand, the homogeneity of the four languages could also be explained as innovations seeking more transparency in local configuration encoding.data on languages from other tupi branches should be examined to resolve what is an innovation versus a retention.
it is nevertheless interesting to note some correlations between some formal patterns and the groupings: • In the expression of 2 → 1, three of the five languages using special pronominal forms belong to subgroup iV.

•
In the expression of 2 → 1, the two languages with second person prefixes for A on the verb root belong to subgroup Viii.

•
In the expression of 1 → 2, the reduced forms roand poare shared by at least four of the 9 members of subgroup i.

CoNCLUSioN 2 -WHAT iS LEFT oF THE HiERARCHY: SAP > 3
At this stage, an important conclusion can be drawn: the literature has offered an overgeneralized and simplified description of the tupi-Guarani indexing systems.Previous sections have presented many counter-arguments to the supposed 1 > 2 hierarchy.table 13 recapitulates how the four assumptions of table 4 have been challenged in the above sections.the idealized hierarchy is obviously a fallacy, if it is taken as a rule supposedly governing all verb forms in all situations. in that idealized system, it applies only partially to the local configurations.
Given the challenges summarized in table 13, what kind of hierarchy must more accurately be posited for the tupi-Guarani indexing systems?table 14 summarizes the findings regarding the hierarchies at play in the 28 languages of the study.Asuriní do tocantins iV, Ava i, Chiriguano i, Guarani correntino i, Jopara i, Kaiwá i, Kamaiurá Vii, Mbya-Guarani i, siriono ii, tapieté i, tapirapé iV, Guajajára iV, tupinambá iii, Yuki ii, Zo'é Viii five languages do not show any influence of any person hierarchy in their argument indexing system (cf.end of the section: the 'idealized' tupi-Guarani hierarchical system).only two (Ava-Canoeiro iV and Kayabi Vi) show a transparent hierarchical system of person indexing based on a 1 > 2 > 3 hierarchy (section: Challenging the General Assumption).six languages encode the local configurations independently from the person value of the two arguments, but do code with reference to the grammatical role of the arguments.Anambé V, Araweté V, Guajá Viii and Xeta i systematically encode P, while Emerillon Viii and Wayampi Viii systematically encode A in a prefix on the verb.All the other languages use some strategy to avoid transparent marking for 1 → 2, such as double-marking, the use of a single 1eXcl A marker for the 1 → 2Sg configuration, or the use of generic *poro-for a 2 nd plural argument.there is no strong argument for stating that there is an underlying person hierarchy subsequently blurred by some pragmatic strategy.i therefore consider that no clear tupi-Guarani-wide hierarchy between the sAPs determines the encoding of 1 → 2. the preferred encoding of 1 st person P in 2 → 1 could just as well be explained as a sub-system of another type of indexing system (absolutive), or as obeying a hierarchy P > A. so that this configuration alone is not enough to strongly argue for a 1 > 2 hierarchy.only the hierarchy sAP > 3 can be strongly posited for the common (i.e.widespread) tupi-Guarani hierarchical system. it is very clearly operative in 23 languages out of the 28 investigated in the study.it is operative in a straightforward way: the participant that is higher-ranked is the one to be indexed on the verb.it determines almost exactly the encoding of the relevant participants33 .the general lesson that must be drawn from the diversity of tupi-Guarani indexing systems described in this paper is that linguists should pay more attention to how languages encode local configurations.

Strategy 8 inclusive marker replaces 1 st or 2 nd marker, or entire combination Strategy
9 Merged 1 st / 2 nd person marker is part of both 1 → 2 and 2 → 1 combinations

table 6 .
distribution of person markers in Proto-tupi-Guarani.

table 7 .
Counterexamples to the 'only one index' rule.

table 10
27the singular form jpa of the auxiliary is used elsewhere as a progressive auxiliary, and the plural form kupa as a collective marker.neither form can be used as a verb in an independent clause.this suggests that the pattern illustrated in table 10 grammaticalized long ago.

table 13 .
How the four assumptions about the hierarchical indexing system are challenged.