THE DEVELOPMENT OF PERSON AGREEMENT MARKERS: FROM PRONOUNS TO HIGHER ACCESSIBILITY MARKERS* Mira Ariel (1999) Tel-Aviv University 1 . From free pronouns to verbal agreement: Introduction Grammatical items often begin their linguistic life as regular lexical items (Meillet 1912) . Grammaticization is said to have occurred when the position of such lexemes (or even phrases) becomes fixed , their meaning is generalized/bleached , their domain of applicability enlarged , significantly raising their frequency , their form modified (usually phonetically reduced) , and their occurrence made obligatory , even when informationally redundant (see Bybee et al . 1994) . Bybee (1985: 38) described the process , saying that "frequent items are gradually reduced both phonologically and semantically , and are simultaneously gradually fused , again both phonologically and semantically , with lexical matter contiguous in the syntactic string" . High frequency is then a prerequisite for grammaticization (see also Hopper and Traugott 1993) . A similar process of grammaticization may also turn less grammatical items into more grammatical ones . This is what happens to pronouns when they gradually become bound inflections . Graduality is characteristic of grammaticization . Full forms first reduce (usually as a result of de-stressing) and become dependent and cliticized , and only then do they become (obligatory) bound morphemes , sometimes merging with other morphemes to such an extent that they become "portmanteau" morphemes (see Heine and Reh 1984 , Bybee et al . 1990 , Croft 1990 , Hopper 1991) . The old , alternative forms may continue to be used alongside the innovative forms . Bybee (1985) , however , added to the characterization of such historical changes that this fusion is not merely the result of frequent phonological adjacency , but is , rather , dependent on the coherent relation between the fusing elements . The more relevant the items are to each other , the more likely they are to fuse . Person agreement , argued Bybee , is not the best candidate for morphological fusion (as opposed to aspect , for example) . But she did find that 56% of the languages in her sample had person inflections on the verbs nonetheless . I will suggest that this "surprising" fact results from an independently motivated though unrelated function , namely the tendency for highly accessible referents to be linguistically coded by highly reduced forms . In other words , agreement markers are seen as the diachronic outcome of speakers' synchronic preference for reduced pronouns , forms which well match the high degree of accessibility usually associated with the speaker and the addressee . Indeed , typologists are in agreement that most verbal inflections are developments of free pronouns (See Greenberg 1966 , Dixon 1979 , Lehmann 1987 , Hopper and Traugott 1993) . Givón (1976: 180) believes that "verb agreement paradigms always arise from anaphoric pronoun paradigms" (original emphasis) . Comrie (1978 , 1981) agrees with Givón in general , and offers as example the Mongolian languages , currently undergoing this very change . Givón (1971) had originally argued for such a development with respect to Bantu languages . Moscati et al . (1969) in effect show this for many verbal forms in all Semitic languages (and see Gesenius- Kautzsch - Cowley 1910 , who explicitly make this claim) . Steele (1977) discusses the cliticization of pronouns to verbs in a few Uto-Aztecan languages , and Mithun (1991) describes the development of bound pronouns in other North American languages . Haiman (1991) argues for such a development for Northern Italian dialects . Bosch (1983) , who incorporates this insight into his synchronic theory about pronoun uses , cites 18th and 19th century grammarians who held this view too (Brugmann and Delbr?ck 1911 , Windisch 1869 , Horne Tooke 1798/1968 , Bopp 1816 , Miklosich 1868 and Grimm 1812) . Later we can cite Meillet (1912) , Jespersen (1924/1965) and Bally (1932) . Dixon (1980) argues the same point for Australian languages , suggesting a gradual development of pronouns into agreement clitics/inflection . Helmbrecht (1995a) mentions Tabasaran , Batsbi and Udi (all East-Caucasian languages) . Even American Sign Language , as described by Newport and Meier (1985) , seems to inflect for agreement in a manner which is unmistakably pronominal in origin . Such an origin explains not only the formal affinity between free and bound pronouns , but also why universally , inflections do not mark more distinctions than free pronouns , and sometimes less (see Barlow 1992) . The development of pronouns into inflections can take place at different times for different persons/numbers etc . , and each stage in the pronoun-agreement development constitutes a potentially stable , functional system (see Mithun 1991) . This graduality explains why there is disagreement in the field as to the status of verbal person markers . Although formal grammarians tend to view agreement as a morphosyntactic sentential phenomenon involving the copying of some features from one argument onto a second "agreeing" source , thereby denying agreement any referential or anaphoric function , agreement markers are often referential and/or anaphoric (see Ariel 1990 , 1991 , Helmbrecht 1995a , and especially Barlow 1992 for persuasive arguments regarding the discoursal , rather than purely formal nature of agreement) . In fact , as Anderson (1985) claims , it is no easy matter to determine which are inflectional agreement markers and which are clitics attached to the verb . This is so because they often exhibit the properties of both bound affixes and independent words (see Haiman 1991 for clitics) . Intermediate status is characteristic of change in progress (see Steele 1977 re Mono , for example , though , as Mithun p . c notes , pronominal affixes can persist over thousands of years) . In fact , Bybee et al . (1994) argue that the original semantic meaning of a grammaticized morpheme lingers on , or at least traces of it do (see also Hopper 1991) , which would account for the referential power of even fully grammaticized person agreement markers which had developed out of independent pronouns (as opposed to other sources) . Dixon (1979,) Jelinek (1984) , Anderson and Keenan (1985) , Ariel (1985 , 1990) , Mithun (1986a , 1991) , Du Bois (1987a,b) , Van Valin (1987) , Anderson (1988) and Barlow (1992) assume that at least some verbal person markers are nominal and/or referential arguments . Moravcsik (1987)and Haiman (1991) see them as mere agreement features copied into the verb . Jespersen (1922 , 1924/1965) even considers them superfluous , counter-logical markers . I suggest these opposing views actually reflect the analysts' observations of the two opposite ends of the historical process which may (but need not) turn fully referential pronouns (free lexemes) into pure "agreeing" inflections (bound morphemes) . In fact , for the most part , it is various intermediate stages which are found , best leading to intermediate and variable statuses assigned to such markers (see Doron 1987 , Fassi Fehri 1987 , Gilligan 1987 and Haiman 1991 , where a few tests are implemented) . I therefore agree with Zipf (1935: 246-9) and Croft (1990: 232) that pronouns and inflections form a continuum from independent words to complete fusion , although for specific paradigms/languages the case may be quite unequivocal . Similarly , but not necessarily in exact tandem with this formal continuum , person agreement markers manifest anything between full referentiality and no referentiality . But this scale is not so different from what can be observed in full NPs . Du Bois 1980 , Ms . suggests that even full NPs with determiners exhibit different degrees of referentiality , with the less referential NPs often being absorbed under a predominantly verbal concept , in a verb-object conflation . The changes from full pronouns to bound , inflectional markers , like many other morphological formations , are unidirectional , but still the changes to the verbal forms are sometimes cyclical , repeating themselves in "re-grammaticization" processes (Dixon 1994: 182-5 , see Kemmer 1993 for syntagmatic versus systemic change) . As Givón (1976: 172) said , "Agreement systems meet their predictable demise via phonological attrition much like other bound affixes" . And then , there can be re-generation of the very same system , gradually leading to new bound affixes (see Watkins 1962 , Wartburg 1969: 238 and Steele 1977) . I will argue below that in colloquial Hebrew , future tense verbal forms may now be undergoing such a renewal , with pronouns obligatorily cooccurring with the verbal forms , even though future verbs are inflected for person (number and gender) . Given that universally , agreement markers tend to develop from pronouns , the first question to be addressed is why and how agreement markers develop out of free pronouns . I will mainly discuss two potential explanations: NP detachment (2 . 1) and accessibility theory (2 . 2) . I will suggest that both historical paths are plausible , and that synchronic distributional restrictions may testify to one or the other alternative in specific cases (2 . 3) . I will then address another question pertaining to this grammaticization process: the motivation for the consistent pattern of agreement marking found , according to which it is mostly first/second persons which are overtly marked on the verb , whereas third person is not . The two questions are obviously related , and a good explanation for the process should also motivate the resulting marking pattern . My argument will be that Accessibility Theory can resolve both questions in a unified way (3 . 1) . Briefly , I will argue that only first and second person referents are consistently very highly accessible , and hence only they merit a grammaticization process of reduction which ultimately results in the obligatory coding of first/second persons by highly accessible markers (the agreement morphemes) . Third person referents are not consistently highly accessible , and hence lower accessibility markers (pronouns , full NPs) are more suitable for coding them . 3 . 2 adduces empirical evidence for the accessibility account . In 4 I consider other potential explanations for the prevalent marking pattern . I will discuss typological markedness (4 . 1) , as well as Bybee's frequency-driven morphologization (4 . 2) . I will argue against using these theories to explain the predominant person verbal agreement pattern by relying on textual distributions (in 4 . 3 and 4 . 4 respectively) . The current grammaticization process of verbal person agreement in Hebrew future tense serves as a test case for the competing theories discussed . I argue that only Accessibility Theory can account for this new development (5) . I then examine the possibility of extending the accessibility account provided for the nominative verbal agreement pattern to the ergative verbal agreement pattern of Mayan in 6 . I conclude that both patterns show sensitivity to the degree of accessibility associated with the relevant argument . In 7 , I mention cases where third person verbal agreement is marked . I argue that such attested cases only form counter-examples to the typological markedness account , but not to Accessibility Theory . I also specify what would count as a true counter-example to the accessibility account . No such cases have been reported in the literature to the best of my knowledge . I conclude with reinterpreting the predominant pattern to be overt marking for first/second persons versus no agreement for third persons (8) . No agreement is distinguished from zero agreement , and hence , although typological markedness cannot account for the predominant verbal agreement pattern , it is not after all challenged by those cases where 3rd person verbs are overtly marked . 2 . The mechanism responsible for the pronoun --> bound agreement change If one accepts that many person inflections develop out of pronouns , then one needs to argue that prior to the creation of inflection , an overt pronoun cooccurred with the verb . Once it was there , a cliticization process could plausibly be argued for . Most researchers have been interested in explaining the surprising fact that some languages at some point started inserting a superfluous pronoun into certain clauses , although it was not called for grammatically (2 . 1) . While some agreement systems may have indeed developed in such seemingly double-subject clauses , I would like to argue that person markings on verbs mainly arise in single subject sentences , the subjects of which are free pronouns . I would then have to account for the superfluous NP subject , rather than pronoun , later inserted into the clause (2 . 2) . 2 . 1 . A superfluous pronoun appears: Sentential rhythm and NP detachment Whereas the change from full pronominal forms to bound inflections (1b to c) seems unproblematic , resulting from phonological processes , the syntactic change involved in the insertion of a superfluous pronoun (the change from 1a to b) has been seen as the main problem to be accounted for . That is , why some languages manifest(ed) a double marking of the subject (or absolutive) , at least for a period , or alternatively , why a perfectly grammatical zero-subject sentence starts containing an overt pronoun: (1) a . NP/zero # V --> b . NP/zero # Pronoun # V --> c . NP/zero # clitic+V/V[+inflection] One path of change argues that pronoun insertion is brought about for sentence rhythm . Paul (1880/1995) , Jespersen (1924/1965) , Vendryes (1925/1952) , Bally (1932) , Wartburg (1962) , Lambrecht (1980 , 1981) and Auger (1993 , 1994) consider French to have evolved a verbal person marking system out of the nominative pronouns , the accusative pronouns (moi , toi , etc . ) now substituting for the nominative ones . Old French , argues Wartburg (1962,) was a verb-second language . When the first word of the sentence was unstressed (e . g . , que) , and hence did not quite count as a first element , the nominative pronouns were added for the rhythm of the sentence . Hence , for Wartburg the change schematically looks as in 2 (see also Kuen 1957 , who refers to Thurneysen 1892 , and Harris 1978): (2) a . Zero # Verb[+inflection] b . Unstressed word # je/tu . . . # Verb[inflection disappearing] c . (NP) # je/tu . . . [pronoun reduced to prefixal conjugation] + Verb Haiman (1991) argues for a similar , and even more advanced , development in Northern Italian dialects , which , due to Germanic influence , adopted a V/2 constraint . This primarily affected inverted constructions , and especially second person verbal forms , a fact which leads Haiman to hypothesize that the addition of the "superfluous" pronoun initially occurred in interrogatives , where second person references are most natural . Indeed , in some related Romantsch dialects (Sumeiran , Puter and Vallader) , bound verbal person markers are still restricted to inverted word order . Givón (1976) proposes a different motivation for inserting the seemingly superfluous pronoun-later-turned-affix into the sentence . Agreement is seen as originally a topic marker rather than as a subject marker . It develops in left and right dislocated NP constructions , where topics are overtly mentioned twice , once in full NP form and once in pronominal form , as in: (3) a . Mayai , shei kissed me . b . Shei kissed me , Mayai Since the topic representations in such sentences are obviously prominent , such syntactic constructions are naturally used in contexts where a shift from the current topic to a new one is proposed by the speaker . The connection of verbal agreement to topicality can further account for the fact that agreement is most common for subjects , but less so for objects , more so for definite NPs (see Givón 1976 on Swahili direct objects) and for animate NPs (see Moravcsik 1978 , Comrie 1981)) . Frequent usage of such constructions then leads to a reanalysis of the marked construction as unmarked , and thus as no longer specifically restricted functionally (see also Bosch 1983) . At the very end of the process , the anaphoric relation between the full NP topic and the pronoun may be reanalyzed as a grammatical relation of subject-predicate . Concomitant with this development , phonological changes are expected , such as de-stressing , cliticization and even fusion . Thus , what started as a superfluous pronoun , inserted in order to highlight a topic , may end up as an agreement marker . Givón's analysis is claimed to account for a variety of languages: non-standard French (as above) , Bantu languages , as well as English and French-based Pidgins and Creoles . Fassi Fehri (1987) has claimed a similar origin for agreement in Arabic . Lambrecht (1980) is a study of topic marking in non-standard French , which indeed testifies to the correlation between the frequency of topic shifting sentences with the use of cliticized/bound pronouns . This is also how I interpret Nadasdi's (1995) findings that subjects that are [+definite] and [+specific] (i . e . , more plausibly topics) are likely to cooccur with clitics 3 . 5 times more often than subjects that are [-definite] and [-specific] . Indeed , as befits a grammatical change , it is gradual , so that even in the 1980s Lambrecht claims that the "autonomous" personal pronouns (moi , toi etc . ) as well as the full NPs in such constructions do not function as full-fledged grammatical subjects . 2 . 2 . A superfluous NP subject appears: Accessibility and inflection In effect in line with Sanford and Garrod (1981) , Givón (1983) and Chafe (1987) , I have argued (Ariel 1985 , 1988 , 1990 , 1991 , 1996) that when reference to mental entities is made , the speaker chooses her referring expression according to how she assesses the accessibility of the specific entity for her addressee at the current stage of the discourse . The higher the mental accessibility , the higher the accessibility marker chosen . Accessibility markers are specialized for a variety of degrees of accessibility , and non-arbitrarily so . Lower accessibility markers are more informative (calculated according to amount of lexical material) , more rigid (i . e . , identify a mental entity relatively uniquely) and less attenuated (lengthier or accented , regardless of informativeness) . Higher accessibility markers are on the whole less informative , less rigid and more attenuated (regardless of informativeness) . Such a form-function correlation seems only reasonable , given that mental entities currently not highly accessible are better retrieved by supplying more information , for the addressee has to choose among very many mental entities he stores in his long-term memory . Highly accessible entities , on the other hand , cannot be numerous , because we can only keep a very limited number of mental entities highly accessible at any given moment . However , there are indications that accessible information is de-accentuated , even pronounced less intelligibly as an unambiguous signal that the information concerned is Given rather than New . In other words , it is not only that the speaker invests more energy in coding New information , because it is harder to process . She also intentionally aims at reducing the form expressing accessible information , since it is a useful linguistic cue for the addressee to search for an already available entity . Now , the form-function correlation between accessibility markers and degree of mental accessibility is not perfectly transparent , but the following scale of accessibility markers is quite representative of many languages (though not all languages have all the options listed , see Ariel (1990) for discussion of accessibility and universality) . The differences in the degrees of accessibility coded by (any) full NPs , free pronouns , bound pronouns , agreement and true zeroes prove to be crucial in the development of verbal person agreement markers: (4) The accessibility marking scale: zero < reflexives < poor agreement markers < rich agreement markers < reduced/cliticized pronouns < unstressed pronouns < stressed pronouns < stressed pronouns + gesture < proximal demonstrative (+NP) < distal demonstrative (+NP) < proximal demonstrative (+NP) + modifier < distal demonstrative (+NP) + modifier < first name < last name < short definite description < long definite description < full name < full name + modifier . I have focused on 2 types of considerations taken into account by the speaker when assessing the degree of accessibility associated with specific mental entities in the addressee's memory (although other factors are no doubt involved): entity salience and unity . The former criterion refers to the degree of salience of the potential antecedent (linguistic or non-linguistic) ; the latter refers to the strength of the connection between the referring expression/accessibility marker and the potential antecedent . The Unity criterion is mainly relevant for anaphoric references , and pertains to the distance and the degree of cohesion between the units (e . g . , clauses) containing the two expressions . It is the first criterion , antecedent salience , which will ultimately prove crucial for the formation of agreement inflections . All things being equal , the entities mentioned on the left in 5 are more salient than the ones on their right: (5) Antecedent salience a . Speaker > addressee > nonparticipant (third person) b . High physical salience > low physical salience c . Topic > nontopic d . Grammatical subject > nonsubject e . Human > animate > inanimate f . Repeated references > few previous references > first mention g . No intervening/competing referents > many intervening/competing referents It is agreed upon that inflections which developed out of pronouns must have derived from attenuated pronouns . All we need add via Accessibility Theory is that this reduction is not merely the result of phonological processes characteristic of fast speech pronunciations of frequently mentioned items , but is also the result of speakers' intention to mark some referents as extremely accessible , even more accessible than regular pronoun antecedents are . The following examples from Hebrew , where the speaker oscillates between full pronouns and reduced pronouns , show how sensitive speakers are to properly marking degrees of accessibility of referents for their addressees . (6) exemplifies how the already repeatedly established discourse topic (the press , previously referred to by stressed and destressed they) is either referred to by a full pronoun (hem) or by a cliticized one (h) , depending on the degree of cohesion between the clause containing the anaphoric expression and the previous clause containing a coreferring expression ('but' and 'another thing' signal a shift from current discourse unit , thereby lowering the accessibility of the discourse entity 'the press' and encouraging full pronominal forms -- see originally Li and Thompson 1979): (6) i . . . h [=hem] + mociim et ze kaxa . . . They publish acc . this like-this . . . ii aval hem madgishim . . . h notnim kama . . . But they emphasize . . . they give some . . . iii od davar she+ hem asu . . . Another thing that they did . . . (7) exemplifies how first , when a new entity is introduced into the discourse it is referred to by a proper name (Nubar) , whereas the current topic is coded by a pronoun (Cameron , referred to by stressed and unstressed pronouns) . However , once both discourse entities become highly accessible (both antecedents being close by and repeatedly referred to) , the discourse topic is coded by a reduced pronoun (h) , while the not quite as accessible entity is referred to by a full pronoun (hu): (7) Preceding discourse (translated): Cameroni . . . HEi . . . hei talked to Nubarj . . . Nubarj said . . . Nubarj was still . . . hi [=hu] pashut diber itoj . . . huj xashav . . . He simply talked with-him He thought . . . So , if the reduction (of some occurrences) of pronouns into cliticized , and later into bound morphemes is motivated by the wish to conform to the form-function convention correlating degree of mental accessibility with type of referring expression/accessibility marker , the structure of (8b) , which may contain an overt NP subject , is the final link in a change initiated in structures such as a . i through a . iii: (8) a . i Pronoun # Verb ii . Cliticized Pronoun+Verb iii . zero # V[+inflection] b . NP/Pronoun/zero # V[+inflection] The problematic link for Accessibility Theory is the change from 8aiii to b . In the most extreme case , the addition of a superficially redundant slot for a subject results from a (gradual) reanalysis of the person marker as merely an obligatory agreement feature on the verb , in some cases (such as Hebrew future tense) , as simply nonexistent (i . e . , reinterpreted as a part of the verbal form) . Such reanalysis depends on a high degree of fusion between the person marker and the verb , a fusion which discourages (but does not preclude) perceiving the reduced pronoun as referential . A final potential development may eliminate the referential power of agreeing verbs , which may bring about a ban against zero subjects , so that overt subjects are required , rather than optional . But I suggest that full NPs start occurring alongside person inflections in such constructions even before the referential power of agreement is eliminated (if at all) , because at least in some situations , the degree of accessibility associated with the referent is not deemed high enough to merit reference by agreement only (as opposed to a free or even bound pronoun) . The more fused the former pronoun is (even if still referential) , the higher the degree of accessibility it marks , and hence , the more restricted the contexts for its use are to very high accessibility . Note that speakers cannot even treat the mental representations corresponding to themselves as highly accessible at all times , as is seen in the following examples from Hebrew (both translations from an Alice Walker story) , where the speaker refers to herself by zero or by pronoun (both marked bold) , depending on the criterion of Unity (relation to a previous mention) . Note that all past tense inflections contain a first person agreement marker: (9) ze haya davar shel ma bexax bishvil yalda o It was nothing for (a) girl or isha le+heanes . ani acmi neenasti, kshe ? (a)woman to get raped . I myself was raped-1st . sg, when [I] hayiti bat shtem esre . Ima af paam lo yadaa, u was-1st . sg twelve years old . Mama never (not) knew, and ? meolam lo siparti le ish . (Noga 1985) . [I] never (not) told-1st . sg (to) anybody . (10) Hu pashut himshix le nasot le alec oti la cet He just kept trying to make me (to)go ito, ve lifamim, mi tox hergel, ani xoshevet, with him , and sometimes, out of habit , I guess-fem sg, ? halaxti ito . gufi asa ma she [I] went-1st . sg with him . My body did what (that) [it] shulam she yaase . ve ima meta . ve ani was being paid to do . And Mother died . And I haragti et buba . killed-1st . sg acc . Bubba . (Noga 1985) . While in both examples the speaker is the continuing discourse topic and the mother is an intervening local topic , in (10) the return to the speaker as the discourse topic comes after a much less cohesive clause (the mother's death) . Hence the choice of an overt first person pronoun in (10) but zero pronoun in (9) . As noted above , Bybee (1985) argued that it is surprising that although persons are not so crucially relevant to the events depicted by verbs , 56% of the languages she examined marked persons on their verbs . It seems that person marking is more prevalent than tense marking (50% of the languages) , though not more than tense and aspect taken together . According to Accessibility Theory , the motivation behind verbal person markers is not the same as that behind tense and aspect marking . It is possibly accidental that persons are marked on verbs . The motivation is based on high referent accessibility plus phonological processes , encouraging reduced forms to be bound to free forms . Indeed , although most languages cliticize their pronouns to verbs (Dixon 1979) , some attach them to something other than the verb: e . g . , a few Pama-Nyungan languages , according to Mithun (1986b) ; In Ngiyambaa , the bound clitics are attached to the first sentential constituent -- see Dixon (1980) ; In Serbian/Croation , clitic pronouns occupy second sentence position , in some Iranian languages cliticized pronouns attach to various sentential elements (Croft 1990 , p . c . ) , and in my Hebrew data , innovative reduced pronouns were not always adjacent to the verb (see (7) above) . In other words , reduction , as well as cliticization , is independent of the creation of an inflectional paradigm , although it is a necessary pre-condition for it . The next step which must be satisfied for a specifically verbal inflectional paradigm to develop is for the verb and the controlling argument to be adjacent (so they can fuse) . This can be due to the prototypical subject (or absolutive) verb ordering , or to the special position allocated for clitics (see the mechanism of verbal attraction , as discussed by Heine and Reh 1984 and Lambrecht 1994) . That verbal agreement typically develops for core arguments (subjects , objects ; agents , patients ; ergatives , absolutives) may in fact result from the fact that such core arguments tend to be adjacent to their verb . 2 . 3 . Accessibility Theory versus NP detachment Logically speaking , it is hard to decide what mechanism better explains the development of person agreement . All the theories in 2 . 1 and 2 . 2 adequately motivate the occurrence of the pronoun which later develops into an agreement marker . In fact , however , I think that the three theories account for different kinds of change mechanisms , all of which may well have occurred , but in different languages . In order to see that different mechanisms are involved , we need to see that the stages of development are at least to some extent different , and/or that synchronic restrictions point to different diachronic sources . For example , those dialects mentioned by Haiman 1991 as showing agreement only in inverted constructions clearly attest to the source of inflection development there (interrogatives) . I would here like to concentrate on the differences between Accessibility Theory and NP detachment as explanations for the development of inflectional paradigms . Note that if person verbal markers develop out of free , referential pronouns , rather than copied arguments (the accessibility explanation versus the NP detachment explanation) , it is to be expected that such bound pronouns bear a referential function . Speakers should then be reluctant , at the initial stages at least , to refer to the subject entity twice (by an overt subject as well as by a person marker -- see Steele 1977 for a similar argument) . In those languages where there is some restriction on the cooccurrence of inflection and overt subjects , it is more plausible to assume a change via accessibility , since it can naturally account for why such restrictions exist . On the NP detachment analysis , such a general ban on overt subjects would be highly surprising , since lexical NPs and pronouns referring to them cooccur in Left and Right-dislocated sentences . Under an NP detachment diachronic development , a more appropriate synchronic restriction would be one where agreement is prohibited for non-topical NPs . If agreement develops in topic shifting constructions (Left and Right-dislocated sentences) , it would only be natural that NPs unsuitable for coding topics should never initially trigger the insertion of the pronoun which later serves as the basis for the cliticization process . Next , where agreement marks all three persons , third person in particular , NP detachment constructions are again the likely source . This derives from the fact that NP detachments are prevalent for third person NPs . Where inflections only mark first/second persons , Accessibility Theory provides the more plausible account . Accessibility Theory predicts that non-participants (third persons) are not as extremely accessible as the speaker and addressee (see (5) above) , and do not merit a reduction in coding as do the speaker and addressee . Hence , third persons are predicted to not often be coded by bound pronominal forms (see the discussion in 3 . 1 below) . We proceed now to examine some synchronic restrictions pointing to either NP detachment or to accessibility historical paths of change . The Colloquial French data (Lambrecht 1980 , 1981) point to an NP detachment source . Third person verbs are marked , and verbs of quantified NPs (unsuitable topics) tend not to carry agreement . But my data for the initiation of a new future person inflection in Hebrew (see section 5 below) shows that left and right dislocated NPs are not necessarily involved in the initiation of the new inflectional paradigm . NP dislocations were extremely rare in my data , and none of the innovative reduced pronouns in future tense occurred in NP dislocated constructions . These new future forms are better motivated by Accessibility Theory (see section 5) . According to Accessibility Theory , although the final stage of inflection formation regularly allows full NPs functioning as the subjects of the inflected verbs , in the intermediate stages (which may last quite long) , such cooccurrences may be prohibited , since the agreement marker is still perceived as a full-fledged referring expression . A few examples for such restrictions follow , corroborating my hypothesis that accessibility considerations motivate the development of verbal agreement . According to McCloskey and Hale (1984) , in Irish , some but not all person-number combinations have inflected forms . But those which do , require zero subjects , disallowing overt NP subjects . So-called analytic forms (non-inflected verbs) require overt pronouns (or full NPs) as subjects . Moreover , as predicted by Accessibility Theory , synthetic forms (inflected) are typically first or first/second person . Analytic forms are typically third persons . When both analytic+pronoun and synthetic+zero forms are allowed (in specific Connacht dialects) , it is restricted to third person . A similar case is Gavi?o (Andean Equatorial) . It too does not use pronominal prefixes when overt NPs occur (Dixon 1994: 47) . Chamorro (based on Doron 1987) reflects a later stage , where full NPs are allowed to cooccur with inflected verbs , but pronouns are not . In other words , it seems that inflections have grammaticized and become obligatory , but pronouns are still seen as marking too similar a degree of accessibility to agreement markers , and are thus prohibited . The same holds for Swahili (see Dixon 1979 , Corbett 1991: 110-11) and Biblical Hebrew (see Gesenius- Kautzsch - Cowley 1910) , where personal pronouns occur only when they are emphatic . In two special present tense paradigms found in Hebrew , which are grammatical borrowings from Aramaic , and hence entered Hebrew relatively late (see Ariel 1998) , inflected forms cannot cooccur with overt subjects for the most part (*ani xoshvatni 'I think-first person') . In Ngiyambaa (and in fact in most Australian languages , see Dixon 1980) a sentence can include either a free pronoun or a bound clitic , but not both . In general , those Australian languages which did not develop bound clitics commonly use free pronouns , whereas those that have developed them hardly ever use pronouns . In Venetian and Padovano Italian , third person verbal markers may be missing when a full NP occurs as subject ; In some Romantsch dialects preverbal tonic pronouns are in complementary distribution with bound atonic pronouns (see Haiman 1991 , who relies on others) . Pemon (Cariban) also exhibits a complementary distribution between full NPs , pronouns and pronominal affixes (Jose R . Alvarez p . c . ) . Allen (1995) claims the same pattern for Inuktitut (an Eskimo-Aleut language): Overt first/second person pronouns occur only for extreme emphasis . Hebrew allows zero with verbs inflected for person , but does not force it . In other words , both full NPs and pronouns can cooccur with inflected verbs . However , where verbs are not inflected for person (as in present tense) an overt subject is (almost) always required (similar to a typical non zero-subject language such as English) . Moreover , as argued in Ariel (1990) , although Modern Colloquial Hebrew does not absolutely ban the cooccurrence of overt subjects with inflected verbs (first and second person past and future tenses) , past tense verbs are routinely accompanied by zero subjects and not by overt pronouns (82 . 5% in natural conversation , 89 . 2% in written discourse) . Similar findings hold for the future tense in written (but not conversational) Hebrew: 76 . 5% of future inflected verbs in my data had no overt subject (for data sources see Ariel 1990) . In other words , although grammatically acceptable , sentences containing person-marked verbs overwhelmingly tend not to have overt subject pronouns , which means that there is at least a discoursal if not grammatical complementary distribution between verbal person markers and overt pronouns (though not with full NPs) . Hebrew is quite typical in this respect (Gilligan 1987: 164-5 also mentions S?o Tom? Creole , Pashto and Ecuadorian Quechua) . In fact , all so-called pro-drop phenomena were initially linked to rich person inflection (Taraldsen 1980 , Rizzi 1982 , Chomsky 1982 , Gilligan 1987 , Comrie 1988 , and see Jespersen 1937 , Benveniste 1971 and Givón 1976 for earlier suggestions in this spirit) . Gilligan (1987) , a cross-linguistic study of 100 languages , found that 93% allow zero subjects . Of these , 81 . 7% (76 languages) have inflected verbs , and only 18 . 3% (17 languages) allow zero subjects without person marking . Only 2 languages show agreement but no zero subjects . Such facts point to an accessibility source for the development of inflection , since only accessibility theory , which treats agreement markers as referential (in varying degrees) , can motivate the restrictions on the cooccurrence of pronouns and agreement markers . However , I wish to emphasize that different inflectional paradigms may result from differently motivated mechanisms . Specifically , NP detachments lead to (overt) third person agreement primarily , and Accessibility Theory motivates (overt) first/second person agreement primarily . The latter , however , is the universally predominant inflectional pattern . 3 . The accessibility account for the prevalent person agreement pattern 3 . 1 . The speaker , the present and the concealed one: Prototypical accessibility differences between first/second and third person referents It is a well-known fact that overt verbal inflections are more common for first and second persons than for third person . In fact , Benveniste (1971) claimed that the Indo-European person inflections for all three persons is the exception (see also Givón 1976) , and that third person agreement should be seen as having developed for symmetry only . He mentions Turkish , and American Indian and Semitic languages as languages marking first and second persons on verbs but not third persons . Huehnergard (1987) and Moscati et al . (1969) in effect show this for many Semitic languages , their analysis moreover showing that although not all person inflections of Semitic languages originated in Proto-Semitic , the same pattern of inflection distinguishing between first/second persons , on the one hand , and third person , on the other , repeats itself . Haiman (1985) mentions Hungarian and Hua in this connection (see also B?tori 1982 re Hungarian and Finn-Ugric languages in general) . Mithun (1986b , 1988 , 1989 , 1991) , discussing Lakhota and North American languages in general , finds extensive evidence for the same pattern of historical development , despite the fact that many North American languages are genetically unrelated . Wichita (Caddoan) also overtly marks only first and second persons (Dixon 1979 , from Rood 1971) . Munro (1974) claims that in Mojave , first and second persons are obligatorily marked on the verbs , but third person marking is only optional ; the normal case for it is not to be marked . Croft (1987) mentions Fula as zero marking third person subjects . Helmbrecht (1995a) refers to Dargva , Batsbi and Tabasaran (Caucasian languages) as marking first and second persons but not third persons . Du Bois (1987a) discusses zero absolutive marking in Mayan third persons . Finally , in a statistical study , Bybee (1985) found that whereas 54% of the languages which manifested agreement did not mark third person on the verb , only 14% did not mark first person . Even if they exist , third person agreement markers may have distinctive marking and change patterns from first/second persons . Moravcsik (1978) mentions the Athapascan languages , where different tones apply to first/second versus third person . Foley and Van Valin (1984) note that whereas first and second person verbal affixes are prefixes in Plains Cree (Algonquian) , third person verbal markers are suffixal . Mithun (1991) argues that historically , not all inflections arise uniformly: "Among persons , first and second person pronouns often become bound before third" (102) . Synchronically they may then occupy different positions within the verb (e . g . , prefixes versus suffixes) , number may be marked differently and/or earlier for first/second persons than for third person , third person may be inflected only for non-specifics , or it may not be marked at all . In fact , as argued by Watkins (1962) for Celtic languages and by Bybee (1985) for Modern Proven?al dialects , even when languages do have third person verbal inflections , these sometimes tend to be reanalyzed as part of the stem or another morphological marking (the preterite in the latter case) , at least synchronically . It seems that there is pressure to view third person verbs as not including a person marker , even if they originally did . Now , a possible explanation for the difference in inflectional patterns for first/second persons and for third persons is the general asymmetry between the three persons , conceptually , and in the way they are coded in the languages of the world . Benviniste (1971) , and following him generative grammarians , distinguished between first and second person pronouns on the one hand , which they classified as deictic , and third person pronouns on the other hand , which were considered anaphoric (see also Bergstr?sser 1928/1983 , Chafe 1976 , and Anderson and Keenan 1985) . Indeed , the Hebrew grammarians' terms for the three persons are quite telling in this respect: first person is the 'speaker' , second person is the one 'present' and third person is the 'concealed' one . The same difference can explain why first/second person reflexives can (marginally) refer extra-linguistically , even in languages where third person reflexives require a linguistic (c-commanding) antecedent (a difference first noted by Ross 1970) . In fact , many of the world's languages do not even have forms that are strictly speaking third person personal pronouns . Instead , they use the general demonstrative pronoun for such references (Steele 1977 , Greenberg 1978 , Dixon 1980 , Comrie 1981 , Mithun 1989 , 1991 , Allen 1995) . Helmbrecht (1995b) mentions Lak in this connection , and Udi (where in fact the third person verbal marker has developed out of a demonstrative pronoun) . Split ergativity systems often distinguish between first/second persons and third persons (Dixon 1979 , DeLancey 1981) . These differences are clearly related to the different discourse roles played by the three persons , with the first and the second referring to actual participants in the speech event , and the third specifically marked as not participating in the discourse (see Hockett 1966 , Benveniste 1971 , Silverstein 1976 and Lyons 1977 , inter alia) . As Lyons (1977: 638-9) says: "That there is a fundamental , and ineradicable , difference between first-person and second-person pronouns , on the one hand , and third-person pronouns , on the other , is a point that cannot be emphasized too strongly" . More recently , researchers have preferred to analyze the differences between the two speech act participants and the non-participant third person in terms of the higher salience/referentiality/animacy of the former . Such researchers were trying to define scales of NP prominence of various sorts , distinguishing between definite and indefinite NPs , subjects and objects , animates and inanimates , etc . , in order to account for case marking , for example . In such frameworks , a person hierarchy emerged , since first and second persons are more prominent in that they are more likely to form the topic of the sentence (Silverstein 1976 , Kuno 1976 , Givón 1976 , Hawkinson and Hyman 1976 , Dixon 1979 , 1980 , Comrie 1981 , 1986 , and Mallinson and Blake 1982 , as cited in Jelinek 1984) . It is also easier to empathize with the speaker , then with the addressee , and last with a third party , argued Kuno (1976) . Givón (1971) noted that whereas both speech act participants are human , third person referents are not necessarily so . They may very well be inanimate . Following Du Bois (1987b) , I suggest that the animacy scales are in this case equivalent to the accessibility scale , and that it is accessibility considerations which provide an adequate account for the prevalent agreement marking , distinguishing between first/second and third persons . 3 . 2 . Empirical evidence for the accessibility account Let us review some empirical evidence for the very high accessibility coding of first/second person references , as opposed to third person references . Harel (1992) is a collection of taped and published long phone conversations between acquaintances in Hebrew . If we concentrate on the high accessibility pronoun/zero subjects in Harel (1992) , we can see how often third person referents are considered accessible enough so as to merit such high accessibility markers . According to Accessibility Theory , it is very high accessibility cases which induce bound person morphemes . Table (1) shows how infrequent third person highly accessible referents are in the data (third persons constitute 22 . 8% of all referents) , as compared with the frequent references to first/second persons by high accessibility markers: 1st person 2nd person 3rd person Total 367=51 . 3% 255=35 . 7% 93=13% 715=100% Table 1 . Zero/pronominal subjects in Harel (1992) To see how much more often first and second person referents are taken as extremely accessible , compare the percentage of zero subjects each of the three persons contributes in this text: 1st person 2nd person 3rd person Total 65=31 . 55% 125=60 . 7% 16=7 . 7% 206=100% Table 2 . Zero subjects in Harel (1992) Note that for Accessibility Theory , it is not really inter-person frequency comparisons that matter . It is the frequency of high accessibility coding per person which is crucial . In other words , we expect many of the first/second person referents , but few of the third person referents , to be extremely accessible , regardless of their absolute frequencies in discourse . Let us now look at a Hebrew single topic journalistic narrative (Levy 1995) , where third person references are by far the most frequent (81% of all references) . As predicted by Accessibility Theory , it is full NPs that form the majority of third person references: 71 . 5% (153 instances) . Table 3 lists the frequency of zero-subject choices for each of the three persons in Levy (1995) , and again there is a gap between first/second persons and third person . Clearly , a much higher proportion of first/second person referents are coded by the highest accessibility marker: 1st person 2nd person 3rd person 14=35 . 9% 9=81 . 8% 29=13 . 5% Table 3 . Zero subjects in Levy (1995) (percentages calculated out of total numbers of occurrences per person) Next , consider the data from Lotan (1990) , a Hebrew face-to-face conversation . Table 4 presents the data for pronoun and zero (calculated together) for the 3 persons . Even though there are twice as many more third person subjects (100) than first and second person subjects (47 each) , the latter's proportions among high accessibility markers is much higher (2 . 35 times more) than that of third person subjects . 1st person 2nd person 3rd person Total 47=41 . 2% 47=41 . 2% 20=17 . 5% 114=100% Table 4 . Zero/Pronoun subjects in Lotan (1990) Concentrating on zeroes versus pronouns only (by ignoring full NPs) , we still see a preference for first/second persons to be coded by zeroes more often than third persons . Table 5 presents the actual numbers . Note that whereas for first/second persons these percentages are also the percentages of zeroes out of the total number of references , this is not so for third persons , which are also referred to by full NPs . The percentage of references to third persons by zero out of the total number of references is 4% . 1st person second person 3rd person 19=40 . 4% 20=42 . 5% 4=25% Table 5 . Zero subjects in Lotan (1990) (percentages calculated out of numbers of high accessibility occurrences per person) These different zero-subject ratios for first/second versus third persons repeat themselves . In interviews and short stories rendered into written Hebrew (Noga pieces analyzed in Ariel 1990) , third person referents are 3 . 5 times more likely to be coded by pronouns than by zeroes (counts here do not include lexical NPs) . First person referents , on the other hand , are 1 . 4 more times likely to appear as zeroes than as pronouns . Second person referents are intermediate (1 . 9 times more pronouns than zeroes) . Table 6 presents the relevant data: 1st person 2nd person 3rd person Zero 168=57 . 7% 13=35% 38=22% Pronoun 123=42 . 3% 24=65% 135=78% Total 291=100% 37=100% 173=100% Table 6 . Zero versus pronoun choices in interviews and short stories in Hebrew (analyzed in Ariel 1990) Colloquial Hebrew contains many more overt pronouns than written Hebrew . Although the recording of a conversation between intimates (see Note 12) showed that pronouns are now more prevalent for all persons , the person differences are still consistent: first person subjects were 2 . 45 as likely to occur as pronouns as in zero form . For second person referents the ratio was 1 . 7 in favor of pronouns , and for third person referents the ratio was 12 times in favor of pronouns (and we should also note that in addition , third persons were often referred to by full NPs , which the first two persons were not) . Third person referents are by default of rather low accessibility . They are not automatically extremely accessible as the speaker and the addressee are . Hence , they require special circumstances to merit pronominal or zero reference . One such special case is being the discourse topic . Indeed , in Ariel (1990) I showed that of the meager 4 . 5% of third person pronouns (relatively high accessibility markers) referring to antecedents mentioned in a previous , rather than a current paragraph (low accessibility in terms of distance) , 92% referred to the global discourse topic of the text . Topicality , then , renders an entity highly accessible . However , even if we concentrate on the one continuing discourse topic in Levy (1995) , which constituted 36 . 5% of all third person references and over two thirds of the zero/pronoun choices for third persons , the percentage of high accessibility marking is not as high as might be expected: Almost half of the references to the discourse topic employed full NPs . Table 7 lists third person expressions referring to the one human discourse topic in Levy (1995): Full NP Pronoun Zero Total 31=45 . 6% 17=25% 20=29 . 4% 68=100% Table 7 . Third person references to the global discourse topic (Levy 1995) These discourse data are echoed by similar observations re third versus first/second person references in other languages . Jelinek and Demers (1994) , who discuss languages of the Northwest coast of North America , observe -- just like Mithun (1991) -- that although third persons are usually zero-marked on the verbs , third persons are not really zero-marked in discourse: "The 3sg . deictic element has important discourse uses . It appears frequently in narratives to mark continuity of reference and sequential action across clauses" (715) . Furthermore , a set of demonstrative particles that are neither arguments nor predicates serve to mark a contrastive reference . These are limited to third persons , and appear as adjuncts , thus "compensating" for the zero marked verb . Malotki (1982) argues that while Hopi has no person inflection , it has only first and second person pronouns . The third person counterpart is a demonstrative pronoun . Similarly , Allen (1995) found an asymmetry between first/second person zero subjects and third person zero subjects in Inuktitut . Even in agent role , where referents are most often Given (and hence , an overt NP occurred only in 1% of the cases overall) , third person arguments were overt rather than zero in 41% of the cases (though third person subjects were relatively marginal) . What is constant in all these findings is that third persons are consistently coded by relatively lower accessibility markers . The actual differences in coding do vary (deictic/demonstrative particles vs . pronouns/agreement ; demonstrative vs . personal pronouns ; overt vs . zero subjects) , but the overall principle remains , linking the less accessible third person references with relatively lower accessibility marking . The difference in degree of accessibility between first/second and third person referents explains why in many languages there is a difference in zero-subject options according to persons , allowing first and second person zeroes much more freely . Such a distinction is predicted by Accessibility Theory , since the speaker and the addressee , who are prototypically more accessible than third person referents , should be referred to by using higher , less informative forms of accessibility markers . Zeroes are obviously the least informative forms . Such differences among the persons have been noted for Finnish , for example . Andrews (1993) finds that in South-Western Otom? , whereas first and second person prefixes and suffixes (more attenuated forms) on predicative nominals distinguish dual and plural , pronouns (less attenuated forms) must be used for third person referents . Cohen (1995) notes this for spoken Latin (as reflected in Plautus' Psudolus) . Grammarians have assumed that English does not allow zero subjects , but many have argued against this accepted wisdom (as witnessed by a vibrant e-mail testimony to that effect on The Linguist list in February 1993) . In any case , it is quite clear that although English zero subjects are limited to main clauses , and although initial auxiliaries tend to be omitted as well , zeroes are quite severely restricted to first and second persons (e . g . don't know/picked him up , did you?) . Philips and Reynolds (1987) , who examined prospective jurors' responses , found that subject deletion "occurs far more often with first-person singular pronouns" (77) . The prospective jurors simply did not use second person subjects , but although they did refer to third persons , they did not use zero subjects there . However , you as subject was frequently dropped in judges' courtroom language use (Philips 1984) . Ixil (a Mayan language) also tends to use zeroes for first and second persons , but not for third persons . In fact , the common zero references for second persons in imperatives (see Sadock and Zwicky 1985 , Gilligan 1987) can also be cited in favor of the thesis connecting zeroes and salience . According to Accessibility Theory , the referent in this case is predictable , given the nature of the action depicted by the verb (a plausible action to be requested or demanded from an addressee) and the salience of the addressee (Thrashner 1974) . In other words , the identity of first and second person referents (but not third person referents) can often be deduced much more easily . They should then be coded by higher accessibility markers -- zero versus overt pronouns in the cases mentioned here . I suggest that a similar though less extreme coding difference between first/second and third person referents is reference by a full pronoun (for third persons) versus reference by an agreement marker (for first/second persons) . The more reduced forms are invariably reserved for the more highly accessible referents . The general finding is then that first and second person referents are consistently highly accessible , but third person referents are only extremely accessible when they happen to be the continuing discourse topic(s) . Even so , many other third person entities are mentioned in the same discourses , causing great variability in degree of accessibility for third person referents . Thus , there is nearly a 100% correlation between first/second person and extremely high accessibility . Not only do first person markers (pronominal , agreement or zero) refer to the speaker (at least partially for plural references) , these references are also made to the discourse topic in many cases (in Levy 1995 , 31 -- 79 . 5% --out of the 39 first person uses refer to the discourse topic , although the writer-narrator is not the discourse topic) . On the average , only 25 . 7%-50 . 9% (depending on the text) of third person subject referents are perceived to be highly salient (as judged by 0/pronoun choices) . This means that whereas the degree of accessibility associated with third persons varies greatly within and across texts , this is not so with respect to first/second person references . Extremely high accessibility , the relevant factor for the accessibility account , remains constant for first and second persons across all texts . 4 . Alternative accounts for the person agreement pattern Even though I have presented evidence that accessibility theory can indeed account for the prevalent verbal person agreement favoring overt marking for first/second persons and zero marking for third persons , we should examine other theories which can potentially account for these data: a typological markedness account (4 . 1) , and a frequency-driven morphologization account (4 . 3) . Based on textual counts , I will argue against each of them (4 . 2 , 4 . 4 respectively) . I should stress , however , that I am only arguing against these theories as accounting for the prevalent marking diffrential . Not only are these theories functional in other linguistic changes , they can actually motivate other , marginal person verbal agreement patterns (see Ariel 1998 , and see also section 8 below) . 4 . 1 . Typological markedness: Third person verbs are unmarked , first/second person verbs are marked A possible explanation for the universal finding that third person verbs are commonly marked by zero , as opposed to the overt person markers signalling first and second persons , can be offered on the basis of markedness distinctions between the persons . An explanation relying on markedness was the explanation adopted by most researchers (see Greenberg 1966 , Kury?owicz 1968 , Givón 1976 , Moravcsik 1978 , 1987 , Dixon 1979 , Bybee 1985 , 1988 , Lapointe 1987) , and is the one presented in typological textbooks (e . g . , Croft 1990 , Mathews 1991 , Hopper and Traugott 1993) . As Bybee (1985: 135) put it: "Zero marking . . . tends to occur on the most frequent and less conceptually marked items" . In fact , the non-markedness of third persons is mostly seen to derive from their high frequency in texts . Such theories assume that "grammars code best what speakers do most" (Du Bois 1985: 363) . Since frequently used forms tend to be short , third person references and/or third person verbal forms , assumed to be the most frequent , should be shortest , and often zeroes . This explanation is particularly suitable for those who argue for an NP detachment source for agreement development , since they cannot otherwise motivate why it is first/second persons which are marked whereas third person is zero marked , when it is precisely third person which commonly occurs in NP detachment constructions . Note that typological markedness cannot itself motivate the recurrent pattern of zero marking for third person verbal agreement (as "virgin zeroes") . It needs to rely on another hypothesis which would explain why a language should develop any inflectional paradigm where persons are marked on verbs in the first place . A possible reason is that verbs need to be related to subjects (minimally) , and agreement signals the grammatical relation of predicate-argument . Now , there is , of course , no reason why predicate-argument relations need to be overtly specified for first/second persons , but not for third persons . If such a marking is functional , it is equally functional for all persons . The predication marking function of agreement seems to better motivate inflectional paradigms where all three persons are marked on the verb (at least originally -- see Ariel 1998) . The typological markedness account , then , has to posit not only predicate-argument marking of all persons , but also a later stage in which a reduction of only the third person marker occurs (so-called non virgin zeroes , see Croft 1990) , as a result of the relative high frequency of third person verbal forms . Note , however , that inflections do not always show a marking that is only sensitive to persons , but rather to the specific gender and number within the person , distinguishing between 'you-masculine-singular' , 'you feminine singular' , 'you masculine plural' , 'you feminine-plural' . If so , then typological markedness would have to assume that each variant of third person verbs (including 'they-feminine' where available , for example) is more frequent than each first/second person verbal forms . The relevant data for this is presented in 4 . 4 (see (15) and Table 14) , which makes it clear that at least some third person verbal forms are not at all frequent (e . g . , third person feminine-singular) . In 4 . 2 I will ignore this problem , and assume that the pre-agreement stage includes only three basic pronouns , each specialized for one person , perhaps with additional morphemes for gender and number (similar to the pronominal Chinese system) . I will then argue that a serious problem for typological markedness is that not all frequency counts show third person verbal forms to be the most frequent verbal forms in discourse (4 . 2) . I will therefore conclude that typological markedness is not necessarily responsible for the prototypical agreement paradigm . 4 . 2 . Empirical evidence against typological markedness: Inconsistent frequency counts "Zero marking . . . tends to occur on the most frequent and less conceptually marked items" (Bybee 1985: 135) . Greenberg (1966) had originally argued for a frequency-markedness correlation , with Zipf's (1935) law motivating this correlation "economically" (and see also Haiman 1983) . Indeed , previous frequency studies found a difference in the frequency of occurrence of first , second and third person verbal forms . Bybee (1985) quotes Juilland and Chang-Rodr?guez' (1964) findings for written Spanish and Rodr?guez Bou's (1952) findings for spoken Spanish . Between 41% and 51% of all verbs (depending on the specific tense and genre) are third person singular , between 22%-31% are first person singular , and only between 4% and 16% are second person singular . Philips and Reynolds (1987) also state that third and first person verbs are the most common in their data (prospective jurors answering lawyers' questions) . Such data corroborate the typological marking claim , even though they cannot really explain the close affinity between first and second person marking , when in terms of frequency , first and third persons are much more similar to each other . Let us now check a few more texts . What I found was that different genres show different person ratios (although one would need to examine larger quantities of text in order to establish a secure correlation between genre and relative person frequencies) . In Levy (1995) , the narrative about a single human topic , the findings are even more extreme than those for Spanish . There are 4 . 75 times as many third person verbs as first and second person taken together , as shown in Table 8: No . of occurrences % of occurrence 1st person: 39 14 . 8% 2nd person: 11 4 . 2% 3rd person: 214 81% Total: 264 100% Table 8 . Frequency of verbs according to person in Levy (1995) Third person verbs seem to be the most common ones , then , motivating the shortness of the verbal forms for third persons . The markedness scale predicted from such findings (if obtained on a larger scale) is presented in (11) , and it perfectly fits the prevalent person agreement marking pattern (scales here and hereafter are proportional to the percentages of occurrence in the cited texts): third 1st 2nd (11) Least marked <---------------------------------------------------------------------> Most marked The next text I examined , Morris (1994) , is a collection of slightly edited American English personal stories . Unlike the third person narrative , here first and third person verbs are equally frequent: No . of occurrences % of occurrence 1st person: 224 45 . 6% 2nd person: 45 9 . 2% 3rd person: 222 45 . 2 Total: 491 100% Table 9 . Frequency of verbs according to person in Morris (1994) Note that such findings (if obtained on a larger scale) would predict the following marking scale , which is different from the scale resulting from the previous text: 1st/3rd 2nd (12) Least marked <---------------------------------------------------------------------> Most marked It predicts that inflection in second person should be most frequent , while there should be no person inflections for both first and third person . This is obviously not true . Alternatively , if zero marking is assigned to only one category , which is at least as frequent as the next one , zero marking should have been assigned to either first or third person verbal forms . This would imply that on the basis of such texts , first and third person zero marking would have been equally prevalent in the languages of the world . This is definitely not the case . Of course , the data as such do not contradict typological markedness , because the thesis is formulated in its weak version , namely that the least formally marked form is of greater or equal frequency as a more marked form . But they certainly do not lend it substantial support . A third discourse , the face-to-face Hebrew conversation (Lotan 1990) , shows no difference in the frequency of first , second and third person verbal forms (the differences here are not statistically significant): No . Percentage 1st pers 47 32 . 4% 2nd pers 45 31 . 0% 3rd pers 53 36 . 5% Total 145 99 . 9% Table 10 . Frequency of verbs according to persons in Lotan (1990) According to these conversational data (if obtained on a larger scale) , either all or none of the persons should have been zero marked on the verb . Again , due to the weak formulation of the typological markedness thesis , these findings do not constitute a counter-example to typological markedness , but they predict that person verbal markings/zeroes would be equally distributed among the three persons in the languages of the world . This is obviously false . A fourth source I checked for the distribution of the three persons reveals a fourth pattern . Table 11 presents the number of verbs for each of the persons used in children's free conversations among themselves (on three separate occasions , February 2 , 7 , 9 1996 , children's ages 7:5 , 7:3 , 5:8 and 5:6) . The data is limited to all and only future tense verbs (exclusive of those used as imperatives) , since it was conducted for a specific purpose (see section 5 below): 1st person 2nd person 3rd person Total 51=56 . 7% 11=12 . 2% 28=31 . 1% 90=100% Table 11 . Frequency of future tense verbs according to person in children's Hebrew conversations Such findings (if obtained on a larger scale) would correlate with the following markedness scale , which does constitute a counter-example to the predictions made by typological markedness (see scale 11 above) . It predicts that if only one person is zero marked on the verb , it would be first person: 1st 3rd 2nd (13) Least marked <------------------------------------------------------------------> Most marked A fifth text I studied was Harel (1992) . The findings are presented in Table 12: No . of occurrences % of occurrences 1st person: 370 45 . 7 2nd person: 255 31 . 5 3rd person: 185 22 . 8 Total: 810 100 . 0 Table 12 . Frequency of verbs according to person in Harel (1992) The formal markedness predicted from such relative frequencies (if obtained on a larger scale) is the symmetric scale in (14) (note that second person is 1 . 4 times more frequent than third person , and first person is 1 . 45 times more frequent than second person): 1st 2nd 3rd (14) Least marked <------------------------------------------------------------------------> Most marked This too is an unequivocal counter-example for typological markedness re verbal agreement . It predicts zero first person marking as opposed to overt marking for second and third persons , a pattern which is not attested , as far as I know . The findings presented in Tables 8-12 have quite inconsistent results for the typological markedness explanations . Levy (1995) seems to support the non-markedness of third person verbal forms , since these are the most frequent verbal forms there . Morris (1994) and Lotan (1990) are consistent with it only because typological markedness is formulated in its weak version . The children's data and Harel (1992) clearly refute the assumption that third person verbal forms are unmarked in terms of verb form frequency . Du Bois (1987b) , who notes that genres differ significantly in the functional pressures they impose on the speaker (he specifically mentions narratives versus conversations and first versus third person narratives) , concludes that for argument structure , it is narratives which are the better source for teasing out discourse strategies used for smooth information flow . Greenberg (1966: 45) notes that narratives would have many more third person references , and suggests conversations for the study of person . The following quote from Croft (1990: 87) emphasizes the need to consider oral data for markedness research: Generally , one attempts to use the "unmarked" text style , that is conversation or oral narrative , rather than written genres . One reason for doing this is that studies . . . have indicated that the textual frequencies for certain "marked" categories increase in formal written styles , and hence , they are not such reliable indicators of correlations between text frequency and other markedness criteria (Actually , what this really indicates is that there is a correlation between informal , oral style and some , if not all , unmarked categories) . Milroy (1992: 5) in effect agrees with Croft , especially regarding the appropriate source for analyzing linguistic change: it is in spoken , rather than in written , language that we are able to detect structural and phonetic changes in their early stages . Milroy (1992) emphasizes that conversational data specifically is superior to other oral genres , if linguistic change is the object of inquiry: it is typically in the day-to-day situational context of speaker interaction that structural changes take place , and it is in these contexts that they have to be investigated (Milroy 1992: 32) . In conversation , Milroy argues , speakers do not mind so much if a linguistic change they are making causes some vagueness or ambiguity . They know that their messages contain many redundancies anyway , their addressees can rely on freely available contextual cues , and they can always stop them and ask for some clarification . This is particularly relevant for the change at hand , which reduces forms (pronouns) and makes them more opaque . Hence , it seems that Harel (1992) , Lotan (1990) and the children's data are the appropriate texts for examining linguistic change . These conversational data do not support the typological markedness proposal . But then , the Spanish findings for spoken discourse were quite different , and supportive of the typological markedness proposal . I therefore suggest we draw the conclusion that texts drastically vary in the frequency of their verbal forms according to person (Note that the number of patterns of distribution equals the number of sources I checked) . Thus , while I do not wish to argue that first/second person are always more frequent than third persons , I think that the findings in 4 . 2 argue against assuming that all in all third person is by far the most frequent person in verbs . If third person cannot be shown to be consistently the most frequent verbal form , the typological marking account fails , since it crucially relies on third person verbal forms being the most frequent ones in change-initiating discourses . 4 . 3 Frequency-driven morphologization For typological markedness we have concentrated on third persons verbal forms . However , we can try and account for the differential verbal markings by focussing on the overt first/second person markings . Instead of explaining the zero marking on third person verbal forms , one can motivate the overt markings on first and second persons (see Bybee et al . 1990) , i . e . , the fusion of first and second person pronouns , but not third person pronouns , with verbal forms . It is only reasonable that " . . . in order for a gram to fuse with a stem it must occur contiguous to the stem with sufficient frequency and be in the same phonological phrase or intonation unit as the stem" (Bybee et al . 1990: 29 , and see Givón 1971) . This resembles the accessibility account . But note that the accessibility account is oblivious to whether first/second person pronoun-verb pairs are less , equally or more frequent than third person-verb pairs . All it relies upon is that the proportion of referents of extremely high accessibility is higher for first/second person than for third person referents . For the frequency-driven morphologization account , pronoun-verb pairs are directly compared across the three persons , and the most frequent one(s) are expected to encourage inflection formation . For the frequency-driven morphologization account to work , all we need assume is a high frequency of first/second person pronoun-verb adjacent cooccurrences , as opposed to a low frequency of third person pronoun-verb cooccurrences . In other words , one can claim that regardless of whether third person verbal forms in general are or are not the most frequent , when it comes to pronominal references , perhaps first/second persons predominate , and hence the tendency to reduce first/second person pronouns to agreement markers (see Croft 1990: 80) . Indeed , I think such an assumption can perhaps explain why certain Hebrew cognition verbs have an agreement marker only for first person in present tense , although Hebrew present tense does not usually have any person agreement (see Ariel 1998) . For the predominant agreement pattern , the explanation could argue roughly as follows (William Croft p . c . ): Accessibility considerations explain why the prototypically less accessible third person referents are mostly referred to by overt NPs , whereas the consistently highly accessible first/second person referents are referred to by pronouns . The rest takes place automatically , due to the high frequency of the cooccurrence of first/second person pronouns with verbs , as opposed to a relatively lower counterpart cooccurrence for third person pronouns . 4 . 4 . Empirical evidence against frequency-driven morphologization: Inconsistent frequencies of first/second versus third person pronoun-verb cooccurrences The frequency-driven morphologization explanation is based on direct and absolute inter-person frequency comparisons . It is the high frequency of the cooccurrence of subject pronouns next to verbs which should govern the formation of overt agreement markers . Thus , absolute numbers of references to highly accessible antecedents should be compared (because they are the ones which would be coded by pronouns in the pre-agreement stage) . Since my data come from Hebrew and English , subject pronouns will be the high accessibility markers counted for English , and both pronouns and zeroes will be counted for Hebrew . Once we compare these ratios , however , it is again not clear why it is third person verbs which are consistently unmarked for person . In other words , it is not invariably the case that first/second subject pronouns (and zeroes) outnumber third person subject pronouns (and zeroes) . For instance , in Harel (1992) (the Hebrew phone conversations between acquaintances) , there are 3 . 95 times as many first person and 2 . 75 second person zero/pronoun subject references as there are third person zero/pronoun references (for actual data see Tables 1,2) . Such findings seem to support the frequency-driven morphologization hypothesis regarding the creation of agreement markers for first/second persons . However , in Levy (1995) (the Hebrew narrative) , there are more third person high accessibility references adjacent to verbs (1 . 4 times more than first person , 5 times more than second person) . This would predict third/first overt marking versus zero marking for second person . Yet another pattern emerges from the children's data of future tense . There are over 4 times as many first person zero/pronouns as there are second and third person zero/pronouns . While these findings may indeed motivate first person agreement marking , there is an equal number of second (10) and third (9) person zero/pronouns . These two should then be predicted to be zero marked . Needless to say , none of these predictions are borne out . Similar numbers characterize the ratio of subject pronominal references in the Hebrew interviews and short stories presented in Table 6 above and in the English personal narrative (Morris 1994) . In the interviews and short stories , there is hardly a difference between third and first person pronoun-verb combinations (1 . 09 times more third person) , but there is a large gap between these two and second person pronoun-verb combinations (5 . 1 times more first persons , 5 . 6 times more third persons) . This means that according to the Hebrew interviews and short stories , as well as Levy (1995) , it should have been first and third person pronouns which should have cliticized onto the verbs in a zero-subject language . But a language without zero subjects (where I count both zeroes and pronouns) is the more plausible origin for agreement creation . Counting this way , although there are more first than third person zero/pronoun references in the short stories and interviews (1 . 7 times more -- see the data in Table 6) , the gap between third and second person zero/pronoun references is significantly larger (4 . 7 times more third person zero/pronouns) . Hence , again one would expect first and third person pronouns to have developed into agreement markers rather than first and second person references . In Morris (1994) there are almost 2 first person subject pronouns per one third person subject pronoun , but about 2 . 7 third person subject pronouns per one second person subject pronoun . Similarly , in Huebner's (1983) data on the acquisition of the English pronominal system by an adult (as a second language) , while there are 1 . 66 more first than third person pronouns (as expected) , there are twice as many third person pronouns as there are second person pronouns (an unexpected finding) . Agreement should have then developed for both first and third persons , not for first and second persons (the gap between which is 5 . 3 in Morris , 3 . 33 in Huebner) . The data for Huebner (1983)(mostly narratives , some dialogues) are presented in Table 13: 1st person 2nd person 3rd person Total 1777=52 . 6% 533=15 . 8% 1070=31 . 6% 3380=100% Table 13 . Frequency of subject pronouns according to person in Huebner (1983) I have not analyzed the data on persons according to number and gender in the data presented so far . But this is relevant for languages where each agreement marker corresponds to a specific free pronoun , as is the case in Hebrew . I believe it sheds light on other languages as well . Note that the second person feminine verbal forms are not just the masculine forms + some feminine gender marker , although this is the common Hebrew pattern elsewhere . Rather , they are directly derived from the free feminine pronoun (at -> t , aten -> ten 'you-fem-sg/pl' , just like the masculine forms are: ata -> ta , atem -> tem 'you-masc-sg/pl') . There is no way to derive the feminine forms here from the masculine ones , for the regular Hebrew feminine markers added are a , at , or et (and never a deletion , as in the singular form here , or m -> n , as in the plural forms) . In other words , I believe that collapsing verbal or pronominal forms into three groups (the three persons) distorts the picture , where plural and feminine references are of quite low frequencies . If fusion results from adjacency , then we should consider actual adjacent forms , rather than the abstract category "person" . And actual adjacent forms are different within the same person category (cf . he , she , it) . Looking at Morris (1994) , it would seem that an English-style language developing inflection would need to skip a few positions if it were to develop the common first/second person agreement pattern: (15) I: 197 > They: 50 > He: 45 > You: 42 > We: 24 > It: 14 > She: 3 . It is the first , fourth and fifth positions on the frequency scale above which routinely develop person agreement markers on verbs . They are not , however , the most common pronouns . Third person they and he are more frequent than you and we . Table 14 presents the frequency of subject pronouns and zeroes according to person , gender and number in Levy (1995) and Harel (1992): Levy (1995): Harel (1992) Least frequent: Y-fm-sg/Y-fm-pl/Y-ms-pl 0 Y-fm-sg/Y-fm-pl/They-fm 0 She/They-fem 2 She 9 Y-msc-sg/They-ms 11 Y-pl-ms 21 I 15 We 26 We 24 They-ms 27 He 45 He 67 Y-ms-sg 234 I 341 Most frequent Table 14 . Frequency of zero/pronouns in Levy (1995) and Harel (1992) (Y = 'you' , persons in bold = out of their "proper" place in the frequency hierarchy) . If we assume that agreement markers arise simply because of high cooccurrence frequencies of specific subject pronouns , but not others , with verbs , then it is not at all clear why feminine agreement ever arises . Note that all feminine subjects are invariably in the least frequent categories , often with no occurrences at all . Ignoring Levy (1995) , where the most common subject is 'he' with 'we' following , consider the conversational data from Harel (1992) , which initially seemed to constitute a more reasonable source for the unmarked inflectional paradigms according to the frequency-driven morphologization proposal . Note that agreement indeed develops for 'I' and 'you-msc-sg' , which top the list , but it then ignores the more frequent 'he' and 'they-msc' , and develops for 'we' and 'you-pl-msc' . It then hops over 'she' , but does develop for 'you-sg-fem' and 'you-pl-fem' , skipping 'they-fem' , which shares their very low frequency . In Huebner's (1983) data , whereas 'I' is the most frequent pronoun , 'he' is equally frequent as 'you' . 'They' is twice as frequent as 'we' , which in turn is only slightly more frequent than 'she' (1 . 16 times more) . A frequency count for Latin pronouns is quoted in Greenberg (1966: 35) . Whereas 'I' and 'you-sg' are indeed by far the most common pronouns , third person singular is is about 5 times more frequent than 'we' , and 7 . 65 times more frequent than 'you-pl' . Nonetheless , as we know , agreement tends to develop for 'we' and 'you-pl' much more often . In the data from Spanish quoted by Bybee (1985) (and mentioned above) , second person plural forms do not occur at all in the spoken data , nor in the preterite of the written data . They constitute 1% of present indicatives in the written data . In the Lorge magazine English pronoun count quoted in Greenberg (1966: 35-6) , he and it each considerably outnumber we (2 . 75 and 2 . 9 times more respectively) . Finally , note that although it is only common-sensical that typological marking and frequency-driven morphologization should complement each other , the findings in 4 . 2 and 4 . 4 have quite the opposite results for the two kinds of explanations (regarding virgin and non-virgin zeroes) . Levy (1995) seems to support the non-markedness of third person verbal forms (only if genders and numbers are calculated together) , since they are the most frequent verbal forms . It nonetheless manifests a frequent third person pronoun-verb cooccurrence , which would support the grammaticization of third rather than first/second person agreement marking through frequent adjacency of pronouns and verbs . Thus , based on Levy (1995) , one theory predicts zero third person marking , while the other predicts overt third person marking . The data from Harel (1992) , on the other hand , argue against the assumption that third person verbal forms are unmarked in terms of verb form frequency (they should therefore be marked morphologically) , but it is nonetheless somewhat compatible with assuming that agreement arises simply due to the frequent cooccurrence of certain but not other pronouns (first/second versus third person) with verbal forms (third person verbs should then be zero marked) . Thus , it seems that accounting for the differential verbal markings by contrasting first/second with third person pronouns (the frequency-driven morphologization account) fares no better than accounting for it by counting first/second versus third person verbs (the typological markedness account) . Each of the accounts seems to gain support from some textual counts , but is refuted by others . The Accessibility Theory account is the only one which is supported by all the textual counts . This is so because Accessibility Theory argues that verbal person inflections (deriving from free pronouns) develop in response to form-function coding considerations encouraging an appropriate fit between highly accessible referents and highly uninformative , non-rigid , and short forms (cliticized pronouns , agreement markers and zeroes -- see Ariel 1985 and onwards for evidence supporting this claim) . Reduction (possibly followed by cliticization and fusion) of pronouns should occur only when referents are extremely accessible . We have seen typical examples of this general phenomenon in 2 . 2 above (and see Ariel 1990 , 1991 , 1996 for many more examples) , and specifically for first/second versus third person referents in 3 . 1 above . Accessibility Theory can then provide an adequate explanation for the common differences in person verbal markings: first and second person referents are more accessible than third person referents (other things being equal) . Hence , in some languages , first/second person references are codable by pronouns , whereas third persons are coded by demonstrative pronouns only (lower accessibility markers) ; in many languages zero subjects are freer for first/second persons ; and in yet others , pronouns referring to first/second persons are more prone to undergo a process of de-stressing and reduction , which may later lead to cliticization and finally fusion with the verb , a state we identify as full-fledged agreement . I suggest that the grammaticization process initiated for the speech act participants is prevalent because they are consistently of high salience , rather than consistently high frequency . Third person referents are extremely salient only when they are the continuing discourse topic (but see Table 7 above) . On other occasions they are quite inaccessible , certainly in comparison to the speaker and the addressee . There is thus no consistently high degree of accessibility associated with third person references . 5 . Hebrew future tense person inflections: A test case Note that Accessibility Theory , NP detachment and typological markedness all present plausible explanations for the person verbal inflectional paradigms found in many of the world's languages (especially if we ignore frequency problems) . They actually complement each other in principle , since whereas typological markedness motivates zero markings for third person verbs , Accessibility Theory motivates overt marking for first/second person verbs . Moreover , both theories are independently well-motivated , accounting for a wide range of data in addition to verbal person inflection patterns . Perhaps we can assume that both of them are sometimes at work in tandem , converging on encouraging overt versus reduced/zero markings (Givón 1991 explicitly makes this assumption) . The reason for this is that both seem to call for the use of reduced/zero forms for the same linguistic codings: often enough the most frequent forms are also the highest accessibility markers (e . g . , the current discourse topic is both frequently referred to and is routinely coded by high accessibility markers) . However , we cannot say that the two theories always work in tandem . I have already shown one reason for this , namely that high frequency (of high accessibility) -- the typological markedness criterion -- is distinct from consistency (of high accessibility) -- the accessibility criterion . These two criteria do not always overlap , and it is the latter which determines which pronouns develop into person markers . Here I will argue that accessibility theory can motivate the creation of Hebrew future verbal agreement when typological markedness cannot be said to operate for another reason: Neither NP detachment nor predicate-argument relation marking , which must be presupposed prior to the application of typological markedness , can be said to be involved in this grammaticization process . I have proposed (Ariel 1990) that new cliticized pronouns are now emerging in Colloquial Hebrew future tense . First note that as predicted by Accessibility theory , prior to the inflection formation stage , the zero subject option should diminish . This is crucial , since it assures that overt pronouns , rather than zeroes , be used as subjects . This is clearly true for Hebrew future tense . In my intimate conversation data , whereas most of the past tense first/second person references were coded by zero (82 . 5%) there were no future tense zero subjects . Data from children's conversations (see Table 11 above) show that zero subjects occurred with future tense verbs mainly in frozen expressions and impersonals . Out of the remaining 62 optional zero/pronoun choices , zero was chosen only in 7 cases (11 . 3%) . Such numbers are the opposite of the zero/pronoun pattern in past tense mentioned above , and attest to the vanishing referentiality of the future agreement markers . Indeed , native speakers , asked to quickly translate from English to Hebrew clauses such as: I/you/he went/go/will go (out of context) provided overt pronouns for all third persons , and crucially , they did the same for first and second persons in future (and present) tenses . But for the first/second person past tense , they offered zero subjects . In other words , future tense first/second person agreement markers are not taken as referential forms , unlike the counterpart past tense forms (hence the independent pronouns supplied by the translators) . Accessibility theory then predicts that first and second person pronouns will be reduced in pronunciation , due to the very high degree of accessibility consistently associated with their referents . Table 15 shows the frequency of occurrence for reduced pronouns (e . g . , an or ni for ani 'I') in the different persons and tenses . Clearly , the great majority of first and second person future verbs occur with cliticized pronouns , although this is not the case for present and past , nor for third person future . Past present future 1st person 1=3 . 2% 15=26 . 8% 10=76 . 9% 2nd person 0 3=20% 1=100% third person 1=5% 3=10 . 7% 0 Table 15 . Percentages of reduced pronouns out of the total use of pronouns in the 3 tenses according to person (full NPs not included) (intimate conversation -- see Note 12) Past tense inflections are still perceived as transparent (the person agreement marker is clearly identified as a separate unit , which is , moreover , formally similar to the independent pronoun -- see Table (16)) . It is conceived as referential for first and second persons , so zero subjects can be freely used with it . Future tense person agreement inflections are not transparent (the verbal forms are "portmanteau" forms) and hardly felt as referential anymore . A quick look at Table 16 shows how past first/second person agreement markers still resemble free pronouns , whereas third person verbs do not contain an element resembling the third person pronominal form (singular first person was no doubt analogized to second person agreement) . This is the prototypical agreement pattern: Person Pst:Root+ Prs infl Indep . Pron . Future 1st safar + ti ani ?espor/yispor 2nd f safar + t at t+isper+i 2nd m safar + ta ata t+ispor 3rd f safra + ? hi t+ispor 3rd m safar + ? hu (y)+ispor 1st pl safar + nu anu/anaxnu n+ispor 2nd pl f s(a)far + ten/m aten/m t+ispor+na 2nd pl m s(a)far + tem atem t+isper+u 3rd pl f safru + ? hen/m t+ispor+na 3rd pl m safru + ? hem (y)isper+u Table 16 . Past and future inflections in comparison to free pronouns (shown with verb root 'count') Now , compare past and future agreement markers . The latter are not as transparently related to free pronouns , although the pronoun consonant is there for all 2nd person forms and for plural first person . Again , no such traces are discernible for inflected verbs used for third persons . According to accessibility theory , since first/second person future forms are no longer transparent , they are not perceived as containing referring pronominals (they have undergone a desemanticization of their referential force) . Overt pronouns must then be used . I propose that a new verbal paradigm is now emerging for future tense (see the numbers for reduced pronouns in Table 15) , and again , since third persons are not consistently highly accessible , but first and second person referents are , it is only first/second persons which trigger consistent pronoun reduction , followed by cliticization . This cliticization may very well eventually lead to the creation of obligatory person markers . The missing steps are the freezing of one form of reduced pronoun (i . e . , either an or ni for ani , 'I') , and the obligatory adjacency of the subject pronoun to the future verb . What is interesting about this new paradigm is that it is evolving even though the marking of the predicate-argument relation is still successfully performed by the original verbal agreement . Thus , whereas the pronominal traces of the verbal agreement markers are quite opaque in future inflections , since the verbal forms themselves are "portmanteau" forms , they quite unequivocally mark the different persons (similar to third person singular s in English) . Hence , whereas the predicate-argument function is served by the existing system , reference tracking has become difficult due to the diminishing referential power of the agreement . It is this problem , then , that encourages the new verbal agreement formation . The new paradigm cannot also be motivated by the wish to code third person verbal forms by the least marked forms (the creation of non-virgin zeroes according to typological markedness) , since they already are the basic and least marked verbal forms . Last , this development is unrelated to NP detachments , which were rare in general , and did not at all occur for future tense in my data . The development of a new inflectional paradigm from independent pronouns then attests to speakers' wish to reduce the forms referring to first/second persons regardless of the potential usefulness of marking predicate-argument grammatical relations and/or abiding by the typological markedness principle . 6 . Nominative and ergative verbal person agreement This paper is concerned with nominative verbal person agreement patterns . However , if verbal agreement is sensitive to the degree of accessibility associated with the nominative argument in accusative languages , it should equally be sensitive to the degree of accessibility associated with ergatives in ergative languages . Indeed , the accessibility account proposed here is partly paralleled in an account of ergative agreement marking proposed by Du Bois (1987a,b) . Briefly , Du Bois (1987a) argues that (third person) absolutives are not marked on verbs , because these are commonly introduced by lexical forms . Overt pronominal affixes are then superfluous . This is comparable to the accessibility approach , in that both approaches examine verbal agreement in terms of its role in reference to entities . Note that Du Bois analyses a phenomenon which was mostly seen as functional for disambiguating argument roles -- ergative marking -- and shows its verbal marking to be primarily a reference tracking system . I also argue that nominative verbal agreement functions for the most part as a reference tracking system . In this view , nominative , as well as ergative verbal agreement markers , are primarily referential expressions , rather than pointers as to predicate-argument relations (see also Dixon 1979 , Croft 1990: 105) . Both analyses argue that Givenness/accessibility is responsible for verbal agreement , rather than markedness considerations . This follows from the referential function attributed to verbal agreement . Note that Dixon (1979: 79 , 1994) , who does emphasize the differences between verbal and nominal case marking , is puzzled by the fact that in nominal case marking the (conceptually) unmarked case (nominative for accusative languages , absolutive for ergative languages) is often 0 marked on nouns , as it should be , but quite the same (conceptually) unmarked cases are consistently overtly marked on verbs (agreement is often overt for nominatives , as well as for first/second person absolutives) . If we assume different paths of development for nominal and verbal markers (which is supported by the different forms assumed as the historical sources for these markers -- see Dixon (1994: 185-206)) , it is not surprising that verbal agreement patterns differently from nominal case marking in terms of markedness . They serve different functions . In verbal agreement , the indexing of case (whether accusative or ergative) is merely a by-product of the agreement . It is on a par with person , gender and number , namely a feature guiding the addressee as to the intended referent . But in nominal case marking , the indexing of case is itself the primary function . Du Bois' (1985) notion of competing motivations is very relevant for agreement marking (again , whether ergative or accusative) . If verbal agreement is the result of grammaticization of very high accessibility marking , then we expect overt marking (highly attenuated referring expressions) for highly accessible discourse entities . Now , just as I have concentrated on consistent accessibility differences between the persons , Du Bois discusses consistent Givenness differences between agents (ergatives) on the one hand , and intransitive subjects and accusatives (absolutives) on the other hand . Since ergatives consistently refer to Given (actually , highly accessible) discourse entities , we expect them to be overtly marked on the verb , since such markers code referents of consistently high accessibility . The same applies to nominatives in accusative languages (though to a lesser extent , because nominatives include intransitive subjects as well) . Since absolutives do NOT consistently refer to highly accessible entities , no verbal marking is expected . Similarly , since third persons (in general) do not consistently refer to highly accessible entities , third person coding is not expected to (often) grammaticize as verbal inflection . First/second coding is . There are , then , generalizations about the correlations between Givenness/accessibility and case on the one hand , and between accessibility and person on the other . However , ergatives , nominatives and absolutives must take all persons . Hence , converging , as well as conflicting motivations , are bound to present themselves . Indeed they do . Third person absolutives are expected to be non-overtly marked by both the case and the person criteria (both third person and absolutive case do not correlate with Givenness/high accessibility) . Indeed they are not (in Mayan) . First/second person ergatives are predicted to encourage overt verbal marking by both criteria (since both agents and speech act participants are expected to be highly accessible) . Indeed they are (in Mayan) . Theoretically , third person ergatives show a clash between the person and the case predictions (low accessibility for person , high accessibility for case) , but statistical findings (Du Bois 1987b) show that in effect they are mostly Given . They are accordingly marked in Mayan . Two other person/case combinations potentially motivate competing marking patterns: first/second person absolutives and third person nominatives . The former belong in a case not characteristic of Given entities , but they themselves are clearly and consistently highly accessible , despite that role . They are indeed marked in Mayan . The latter belong to a case role that more often than not is Given (nominative) , but they themselves are not consistently highly accessible , as we have seen in the counts above . Hence , the lack of uniformity in verbal marking according to persons . In Mayan languages (see Du Bois 1987a) and in Yawa (Papuan , see Dixon 1994: 76) , at least , whereas third person absolutive is 0 marked on the verb , first/second persons are overtly marked . In Choctaw-Chickasaw (2 close dialects , Muskogean -- see Dixon 1994: 37-8 , based on others) , agents , patients and datives are marked on the verb (intransitives being assigned to any one of these roles) , and again , first/second persons are overtly marked in all roles , although third person is not . Thus , the patient role , which is expected to represent non-Given entities for the most part , is nevertheless overtly marked on the verb for first/second persons , because first/second persons are consistently highly accessible . The same applies to third person nominatives in many languages , which are zero-marked rather than overtly (see section 3 above) . Thus , the verbal marking for third person ergatives (and nominatives) and for first/second absolutives is potentially influenced by two competing motivations: Case role calls for one type of accessibility marking (low for absolutives , high for nominatives) , and person calls for the opposite type of accessibility marking (high for first/second persons , low for third persons) . This is why first/second person absolutives may "diverge" from the nonmarking of absolutives , and why third person nominatives may "diverge" from the overt marking characteristic of nominatives . Looked at this way , the inconsistent marking for persons in both nominative and ergative verbal agreement systems is only apparent , and can be motivated in exactly the same way , namely by reference to form-function correlations between referring expressions and consistent high degrees of accessibility of certain discourse entities . 7 . Counter-examples to typological marking and to Accessibility Theory Table 17 sums up the differences in what the relevant data is according to Accessibility Theory , frequency-driven morphologization and typological markedness . It will help us see which data forms a counter-example to each theory . Typological Frequency-driven Accessibility Theory Markedness Morphologization What person is reduced/zero? 3rd person 1st/2nd person 1st/2nd person . What form is compared for frequency? verbal forms nominative pronouns Reduced nominative in pre-agreement stage . pronouns in pre-agreement stage . How is frequency calculated? Absolute numbers Absolute numbers, Proportional numbers compared across the compared across the 3 per person , compared 3 persons . persons . across the 3 persons . Table 17 . Relevant data for Typological markedness , Frequency-driven morphologization and Accessibility Theory regarding verbal person agreement In the end , then , it seems that the controversy hinges on what linguistic elements are considered paradigmatic alternatives (or "what to count" -- see Croft 1990: 87-8) . I am claiming that the relevant paradigmatic alternatives in this case are often (though not always) referring expressions , rather than verbal forms . Moreover , the relevant counts are mostly intra-person counts , rather than inter-person ones (at least not directly so) . I believe that this is why Accessibility Theory can explain linguistic marking patterns which constitute counter-examples to other theories . English third person singular simple present tense is the only person to be marked on the verb . Greenberg (1966: 44) further notes that first person is zero-marked in Dutch , while third person has an overt person marker . These are counter-examples to a theory which argues that third person verbal forms are unmarked (typological markedness -- see 4 . 1 above) . However , they do not necessarily form counter-examples to Accessibility Theory , because what is compared under Accessibility Theory is the linguistic code or codes (in cases of double-marking) used to specify the entity involved , and not the verb (alone) . Hence , in English we would compare a , b and c below , and indeed find that precisely because of the overt marking on the verb , third person references are signalled by more lexical information than first/second persons: (16) a . I (verb)+zero b . You (verb)+zero c . S/he/it (verb)+s In other words , unlike typological markedness , Accessibility Theory does not determine that third person verbs absolutely be zero-marked/least marked . It only has predictions for nominal markers (wherever they may be) , and it predicts that if and when grammaticized , third person referents (verbal person agreement included) will be consistently coded by relatively lower accessibility markers than first/second person referents (i . e . , longer , more informative forms) . More patterns are then compatible with Accessibility Theory , among them (1) in Table 18 below , where no person agreement occurs on any verb , (2) , the prevalent agreement pattern , where only first and second persons show verbal agreement , and (3) , where only third person verbal agreement occurs . Of course , there can be many other sub-patterns , but we will make do with these . What patterns 1-3 in Table 18 have in common is that third person referents are invariably coded by lower accessibility marking than first/second person , be it overt subject vs . zero subject (in 1) , pronoun/lexical NP vs . agreement (in 2) , or overt agreement vs . zero agreement (in 3) . Note that whereas (1) and (2) are consistent with both Accessibility Theory and typological markedness , (3) , where the third person verb is more marked than the first/second person verb , is only consistent with Accessibility Theory . In addition to the English and Dutch counter-examples mentioned above , in Modenese Italian , third person present tense shows more marking than first and second persons (according to Gilligan 1987) . In Nez Perce too only third person verbs are marked (see Mithun 1996) . According to Mithun , third person referents are first introduced overtly , and then followed by zero subjects but nonetheless referred to pronominally on the verb . First and second persons are routinely referred to by zero subjects , and their verbs are zero-marked . Hence , first/second persons are coded by a higher accessibility marker (zero vs . bound pronominals in this case) . Genuine counter-examples for Accessibility Theory would be patterns obligatorily requiring lower accessibility marking for first/second person referents than for third person referents . Thus , if there were languages which allowed zero subjects for third person referents , but not for first/second person referents (4 below) or alternatively , if in a language with free zero subjects , only first/second person verbal agreements occurred (5) , then typological markedness would constitute a better account , for these are not counter-examples for it , although they are for Accessibility Theory , since in each of these cases first/second persons are coded by lower accessibility markers than third persons (overt vs . zero subject , overt vs . zero verbal agreement) . As far as I know , whereas there are languages (albeit very few) which manifest pattern (3) , there are none which pattern as in (4) or (5) . This is surprising under a typological markedness account . 1st pers 2nd person 3rd person 1 . Ref Exp . zero zero Overt V Agr zero zero zero 2 . Ref Exp zero zero Overt V Agr Overt Overt zero 3 . Ref Exp zero zero zero V Agr zero zero Overt 4 . Ref Exp Overt Overt zero V Agr zero zero zero 5 . Ref Exp zero zero zero V Agr Overt Overt zero Table 18 . Five types of agreement types (Ref Exp = Referring expression , V Agr = Verbal agreement) Moreover , recall that (a) typological markedness requires an additional mechanism to explain the initial creation of agreement (be it NP detachment , predicate-argument marking or Accessibility Theory) , while Accessibility Theory does not ; (b) typological markedness cannot explain why zero subjects are more prevalent for first/second persons than for third persons , while Accessibility Theory simultaneously motivates the prevalent verbal agreement marking as well this and as many other distributional and grammatical restrictions . Crucially , (c) typological markedness relies on an assumption that third person verbal forms are the most frequent in discourse , but we have seen that they are not necessarily so . Accessibility Theory relies on the assumption that first/second person referents are consistently highly accessible , and we have seen evidence that this is a consistent finding across text types . 8 . Conclusions: Distinguishing between zero agreement and absence of agreement Is zero always a marker of nonmarkedness? It often is , as in singular (most often zero-marked) versus plural (often overtly marked) forms . Thus , when a formal distinction is required between two forms , it is indeed the less frequent form which will normally be formally marked . But zero is not always the unmarked member of the paradigm (Dressler 1987: 14) . For example , whereas Chinese zero/pronoun choices indeed show zero to be the unmarked member (in terms of frequency -- there are more zeroes than pronouns according to Li and Thompson 1979) , English zeroes (unmarked formally) are extremely marked nonetheless , in that they are heavily restricted in their distribution , as compared with overt pronouns . This appears to be a paradox . However , according to Accessibility Theory , there is no zero third person agreement in fact (and see also Mithun 1986b) . No inflection is not the same as zero inflection . Inflection simply never arose for third person (and if it did , it would tend to disappear or be reanalyzed) . Hence , the (formally) unmarked , basic verbal form associated with third persons is the one used for third persons , just because it has no person marking . So-called third person verbs are unmarked because no functional pressure (that is , an accessibility-based form-function correlation) motivated such a marking . As Croft (1990: 158-9) argues: The final criteria for markedness are typological frequency and typological distribution (dominance) . The typological-frequency criterion is relatively simple to account for: if a grammatical semantic category is very infrequent , it simply will not be expressed as a distinct grammatical category in many languages (emphasis added) . In other words , I am suggesting with Mithun (1991) that third person verbal markers are not useful enough to merit grammaticization , because third person referents are not consistently highly accessible , whereas first and second person marking is often grammaticized , because of the consistent high accessibility associated with their referents (and see B?tori 1982 for a similar analysis for Hungarian) . Although theorists of markedness have debated which markedness criteria should be used (see Cairns 1986 and Odlin 1986) , how consistent the results of various markedness criteria should be , what grammatical components are amenable to a markedness analysis (see the various articles in Eckman , Moravcsik and Wirth 1986) , the terms considered members of a contrast set seem to be trivially taken as a given (but see Croft 1990: 87/8) . While this may often be justified , I suggest that the opposition taken for granted here (verbs used for third persons versus verbs used for first/second persons) is not always the relevant one for natural language coding . It is only relevant when predicate-argument relations are to be marked , but often enough this is not what speakers choose to mark , at least not initially . Instead , we should measure the form coding the referent separately from the form coding the verb , even if they are in specific instances merged . Looked at this way , verbal forms are now equally long and complex for all the persons , of course . Common ways of referring to the various persons , on the other hand , are not: (17) 1st person: inflection 2nd person: inflection 3rd person: pronoun/demonstrative/full NP . In other words , the verbal form used for third person is indeed the least marked form , but at least originally , it is the unmarked basic verbal form , used for all persons , and not only for third persons . It is not specific for third persons , even though it often seems to be used solely for them . And the reason it seems to be dedicated to third person is that bound pronouns tend to merge with this basic verbal form only or mostly for first/second persons . I therefore endorse Mithun (1986b) and Croft's (1990: 271) distinction between zero-marked verb agreement and absence of agreement . I believe that Benveniste (1971: 197-8) in effect held this very position too . The third person "only presents the invariable inherent in every form of the conjugation . . . the 'third person' is not a 'person' ; it is really the verbal form whose function is to express the non-person" . Once we assume that so-called third person verbal forms are not actually zero-marked , because they do not really form part of a contrast pair with first/second person verbal forms , the data presented above as counter-examples for typological markedness (English , Dutch , Nez Perce) are no longer relevant , since in the absence of an opposition between forms (verbs , in this case) , typological markedness (between verbal forms) is simply inapplicable . Summing up , I have argued that for the most part , it is not verbal forms alone (typological markedness) and not absolute numbers for pronoun-verb cooccurrences (frequency-driven morphologization) for the three persons which should be compared . We should compare verbal forms alone only if overt subjects are equally present or absent (zero) for the three persons . Since third persons are normally coded by overt subjects , but first/second persons are often coded by zero subjects accompanied by overt verbal agreement , we should actually compare the coding for the subject AND the verb taken together . We then see that in fact , it is first/second persons that are less marked formally , because what for third person is coded syntactically (a subject-verb combination) is coded morphologically for first/second persons (verb+person agreement , see Croft (1990: 80) for syntactic versus morphological marking) . But combinations of first/second person subjects with verbs are not necessarily the most frequent ones! Hence , it must be the ratio of very high accessibility references per person which is crucial for this marking . Thus , even if third person verbs are more frequent , and even if third person pronoun-verb cooccurrences are sometimes more frequent , what determines the finding that first/second , rather than third person pronouns become person agreement markers is not their frequency , but rather , the proportion of very high accessibility references that each performs . These textual frequencies are consistently close to 100% for first/second references , whereas they are between 26%-51% for third person references (see 3 . 2 above) . Accessibility Theory , then , offers a specific application of the frequency-driven morphologization hypothesis . Indeed , once we apply frequency counts to categories and comparisons defined by Accessibility Theory , the usual assumptions about grammaticization processes involving adjacent forms apply: Highly frequent adjacent forms tend to fuse together . But in this case , it is intra-person very high accessibility frequency comparisons , rather than inter-person straight frequency comparisons which determine which pronouns will grammaticize into agreement markers . References Allen , Shanley 1995 . Null subjects and null objects in early Inuktitut . Paper presented at the ellipsis workshop in the Functional Approaches to Grammar conference . Albuquerque , new-Mexico , 7 . 25 . 1995 . Anderson , Stephen R . 1985 . Inflectional morphology . In Shopen (ed . ) vol . III 150-201 . Anderson , Stephen R . 1988 . Inflection . In Hammond and Noonan (eds . ) 23-43 . Anderson , Stephen R . and Edward L . Keenan 1985 . Deixis . In Shopen (ed . ) Vol III . 259-308 . Andrews , Henrietta 1993 . The function of verb prefixes in South-Western Otom? . Arlington: Summer Institute of linguistics and university of Texas at Arlington . Ariel , Mira 1985 . Givenness marking . Tel-Aviv university doctoral thesis . Ariel , Mira 1988 . Referring and accessibility . Journal of Linguistics 24(5) , 65 87 . Ariel , Mira 1990 . Accessing NP Antecedents . London: Routledge , Croom Helm Linguistics Series . Ariel , Mira 1991 . The function of accessibility in a theory of grammar . Journal of Pragmatics 16(4) , 141 161 . Ariel , Mira 1996 . Referring expressions and the +/- coreference distinction . In Thorstein Fretheim and Jeanette Gundel (eds . ) Reference and referent accessibility . 13-35 . Amsterdam: John Benjamins . Ariel , Mira 1998 . Three grammaticalization paths for the development of person verbal agreement in Hebrew . In Jean-Pierre Koenig (ed . ) Discourse and cognition: Bridging the gap . Stanford: CSLI/Cambridge University Press . Auger , Julie 1993 . More evidence for verbal agreement marking in Colloquial French . In William J . Ashby , Marianne Mithun , Giorgio Perissinotto , and Eduardo Raposo (eds . ) Linguistic perspectives on the Romance languages . 177-98 . Amsterdam: John Benjamins . Auger , Julie 1994 . Pronominal clitics in Quebec Colloquial French: A morphological analysis . PhD dissertation , University of Pennsylvania (IRCS report 94-29) . Auger , Julie 1995 . Les clitiques pronominaux en Fran?ais parle informel: Une approche morphologique . Revue Quebecoise de linguistique 24(1) , 21-60 . Bally , Charles 1932/1965 . Linguistique g?n?rale et linguistique Fran?aise . Bern: A . Francke Verlag . Barlow , Michael 1992 . A situated theory of agreement . Outstanding dissertations in linguistics . New York: Garland . Barlow , Michael and Charles A . Ferguson 1987 . eds . Agreement in natural language . Stanford: CSLI . B?tori , Istv?n 1982 . On verb deixis in Hungarian . In Weissenborn and Klein (eds . ) , 155-65 . Amsterdam: John Benjamins . Benveniste , Emile 1971 . Problems in general linguistics . Translated by Mary Elizabeth Meek . Coral Gables , Fa: University of Miami Press . Bergstr?sser , Gotthelf 1928/1983 . Introduction to the Semitic languages . Translated with notes and bibliography and an appendix on the scripts by Peter T . Daniels . Winona Lake , Indiana: Eisenbrauns . Bopp , Franz 1816 ?ber das conjugationssystem der sanskritsprache . Frankfurt an Main: Andre?ischen . Borer , Hagit 1983 . Parametric syntax . Dordrecht: Foris . Borer , Hagit 1986 . I-subjects . Linguistic inquiry 17(3) , 375-416 . Bosch , Peter 1983 . Agreement and anaphora . London: Academic Press . Brugmann , Karl and Berthold Delbr?ck 1911 . Grundriss der vergleichenden grammatik der Indogermanischen sprachen (Vol . 2 , revised edition 1911 , 1916 ; vol 3 , 1893) . Strassburg: Tr?bner . Bybee , Joan L . 1985 . Morphology . Amsterdam: John Benjamins . Bybee , Joan L . 1988 . Morphology as lexical organization . In Hammond and Noonan (eds . ) 119-41 . Bybee , Joan L . 1994 . The grammaticalization of zero: Asymmetries in tense and aspect systems . In William Pagliuca (ed . ) Perspectives on grammaticalization , 235-54 . Amsterdam: John Benjamins . Bybee , Joan L . , William Pagliuca and Revere Perkins 1990 . On the asymmetries in the affixation of grammatical material . In William Croft , Keith Denning and Suzanne Kemmer (eds . ) Studies in typology and diachrony , 1-42 . Amsterdam: John Benjamins . Bybee , Joan L . , Revere Perkins and William Pagliuca 1994 . The evolution of grammar . Chicago: The university of Chicago Press . Cairns , Charles E . 1986 . Word structure , markedness , and applied linguistics . In Eckman et al . (eds . ) 13-38 . Chafe , Wallace L . 1970 . Meaning and the structure of language . Chicago: Chicago University Press . Chafe , Wallace L . 1976 . Givenness , contrastiveness , definiteness , subjects , topics , and point of view . In Li (ed . ) 25-57 . Chafe , Wallace L . 1987 . Cognitive constraints on information flow . In Russel Tomlin (ed . ) Coherence and grounding in discourse , 21-51 . Amsterdam: John Benjamins . Chafe , Wallace L . and Chao , Wynn 1986 . Indefinite NPs and the interpretation of discourse-based null elements . In Eckman et al . (eds . ) 65-84 . Chomsky , Noam 1982 . Some concepts and consequences of the theory of government and binding . Cambridge: MIT Press . Cohen , Even-Grey 1995 . Accessibility markers in Latin and an assessment of the criteria determining them . A seminar paper , Tel-Aviv university . Comrie , Bernard 1978 . LSA Summer Institute class notes . Comrie , Bernard 1981 . Language universals and linguistic typology . Oxford: Blackwell . Comrie , Bernard 1986 . Markedness , grammar , people , and the world . In Eckman et al . (eds . ) 85-106 . Comrie , Bernard 1988 . Linguistic typology . In Newmeyer , Frederick J . (ed . ) 1988 . Linguistics: The Cambridge survey . Vol . I , 447-61 . Cambridge: Cambridge University Press . Corbett , Greville 1991 . Gender . Cambridge: Cambridge University Press . Croft , William 1987 . Agreement vs . case marking and direct objects . In Barlow and Ferguson (eds . ) 159-79 . Croft , William 1990 . Typology and universals . Cambridge: Cambridge University Press . DeLancey , Scott 1981 . An interpretation of split ergativity and related patterns . Language 57(3) , 626-57 . de Saussure , Ferdinand 1959 . Course in general linguistics . Translated into English by W . Baskin . New-York: Philosophical Library . Dixon , R . M . W . 1979 . Ergativity . Language 55(1) , 59-138 . Dixon , R . M . W . 1980 . The languages of Australia . Cambridge: Cambridge University Press . Doron , Edit 1987 . On the complementarity of subject and subject-verb agreement . In Barlow and Ferguson (eds . ) 201-18 . Dressler , Wolfgang 1987 . Introduction . In Dressler (ed . ) 3-22 . Dressler , Wolfgang 1987 . ed . Leitmotifs in natural morphology . Amsterdam: John Benjamins . Du Bois , John W . 1980 . Beyond definiteness: The trace of identity in discourse . In Wallace Chafe (ed . ) The pear stories , vol III . Roy Freedle (ed . ) Advances in discourse processes , 203-74 . Norwood: Ablex . Du Bois , John W . 1985 . Competing motivations . In John Haiman (ed . ) Iconicity in syntax . Amsterdam: John Benjamins . 343-65 . Du Bois , John W . 1987a . Absolutive zero: Paradigm adaptivity in Sacapultec Lingua 71(1-4) , 203-22 . Du Bois , John W . 1987b . The discourse basis of ergativity . Language (63)4 , 805-55 . Du Bois , John W . Ms . Reference and identification . Eckman , Fred R . , Edith A . Moravcsik and Jessica R . Wirth 1986 . eds . Markedness . New York: Plenum . Even-Shoshan , Abraham 1982 . Ha-Milon Hechadash . [A Hebrew dictionary] . Jerusalem: Kiryat-Sefer . Fassi Fehri , Abdelkader 1987 . Agreement in Arabic , binding and coherence . In Barlow and Ferguson (eds . ) 107-58 . Foley , William A . and Robert D . Jr Van Valin 1984 . Functional syntax and universal grammar . Cambridge: Cambridge University Press . Foley , William A . and Robert D . Jr Van Valin 1985 . Information packaging in the clause . In Shopen (ed . ) , Vol . 1 , 282-364 . Fowler , Carol A . and Jonathan Housum 1987 . Talkers' signaling of "New" and "Old" words in speech and listeners' perception and use of the distinction . Journal of Memory and Language (26) , 489-504 . Gilligan , Gary M . 1987 . A cross-linguistic approach to the pro-drop parameter . USC dissertation . Givón , Talmy 1971 . Historical syntax and synchronic morphology . CLS 7 , 394-415 . Chicago: CLS . Givón , Talmy 1976 . Topic , pronoun , and grammatical agreement . In Li (ed . ) 149-88 . Givón , Talmy 1983 . ed . Topic continuity in discourse: A quantitative cross-language study . Amsterdam: John Benjamins . Givón , Talmy 1991 . Markedness in grammar: Distributional , communicative and cognitive correlates of syntactic structure . Studies in language (15)2 , 335-70 . Givón , Talmy 1993 . English grammar: A function-based introduction , Vol . I . Amsterdam: John Benjamins . Greenberg Joseph H . 1966 . Language universals , with special reference to feature hierarchies (Janua Linguarum series minor 59) . The Hague: Mouton . Greenberg Joseph H . 1978 . How does a language acquire gender markers? In Greenberg (ed . ) Vol . III: Word structure . 47-82 . Stanford: Stanford University Press . Greenberg Joseph H . ed . 1978 . Universals of human language . Stanford: Stanford University Press . Grimm , Jacob 1812 . Article in Hallesche Allgemeine , Zeitung , 7 February 1812 , Sp . 258f . Gundel , Jeanette K . , Kathleen Houlihan and Gerald Sanders 1986 . Markedness and distribution in phonology and syntax . In Eckman et al . (eds) . 107-38 . Gundel , Jeanette K . , Kathleen Houlihan and Gerald Sanders 1988 . On the functions of marked and unmarked terms . In Michael Hammond , Edith A . Moravcsik and Jessica R . Wirth (eds . ) Studies in syntactic typology , 285-301 . Amsterdam: John Benjamins . Gurman-Bard , Ellen 1995 . The control of intelligibility in dialogue: The messy , the sticky , and the Oyster-catcher's egg . Presented at Haifa university , April 4 . 1995 . Haiman , John 1983 . Iconic and economic motivation . Language 59(4) , 781-819 . Haiman , John 1985 . Natural syntax . Cambridge: Cambridge University Press . Haiman , John 1991 . From V/2 to subject clitics: Evidence from Northern Italian . In Traugott and Heine (eds . ) vol . 2 . 135-57 . Hammond , Michael and Michael Noonan 1988 . (eds . ) Theoretical morphology . San Diego: Academic Press . Harel , Zvi 1992 . Yossi , believe me as you never believed me before . The Haaretz supplement , November 13 , 1992 , 20-25 . Harris , Martin 1978 . The evolution of French syntax: A comparative approach . London: Longman . Hawkinson , A . and Larry Hyman 1975 . Hierarchies of natural topic in Shona . Studies in African linguistics (5) , 147-70 . Heine , Bernd and Mechthild Reh 1984 . Grammaticalization and reanalysis in African languages . Hamburg: Helmut Buske Verlag . Helmbrecht , Johannes 1995a . The syntax of personal agreement in East-Caucasian languages . Ms . Helmbrecht , Johannes 1995b . The typology of 1st person marking and its cognitive background . Ms . Hockett , Charles F . 1966 . What Algonquian is really like . International journal of American linguistics (32) , 59-73 . Hopper , Paul J . 1991 . On the principles of grammaticization . In Traugott and Heine (eds . ) vol . 1 . 17-35 . Hopper , Paul J . and Elizabeth Closs Traugott 1993 . Grammaticalization . Cambridge: Cambridge University Press . Horn , Lawrence 1984 . Toward a new taxonomy for pragmatic inference: Q-based and R-based implicature . In Deborah Schiffrin (ed . ) Meaning , form and use in context: linguistic applications , 11-42 . Washington: Georgetown University Press . Horne Tooke , John 1798/1968 . Diversions of Purley (2 vols) . London: Scholar Press . Huebner , Thom 1983 . A longitudinal analysis of the acquisition of English . Ann Arbor: Karoma . Huehnergard , John 1987 . "Stative" , predicative form , pseudo-verb . Journal of Near Eastern studies (47)3 , 215-32 . Janssen , Theo A . J . M . to appear . Deixis and reference . In Geert Booij , Christian Lehmann and Joachim Mugdan (eds . ) Morphology: A handbook on inflection and word formation . Berlin: Mouton de Gruyter . Jelinek , Eloise 1984 . Empty categories , case and configurationality . Natural language and linguistic theory (2)1 , 39-76 . Jelinek , Eloise and Richard A . Demers 1994 . Predicate and pronominal arguments in Straits Salish . Language 70(4) , 697-736 . Jespersen , Otto 1922 . Language , its nature , development and origin . London: George Allen & Unwin ltd . Jespersen , Otto 1924/1965 . The philosophy of grammar . New York: W . Norton & company . Jespersen , Otto 1937/1984 . Analytic syntax . Chicago: Chicago University Press . Juilland , Alphonse and E . Chang-Rodr?guez 1964 . Frequency dictionary of Spanish words . The Hague: Mouton . Gesenius , W . - E . Kautzsch - A . E . Cowley 1910 . Gesenius' Hebrew grammar . As edited and enlarged by E . Kautzsch . Translated from German by G . W . Collins . Translation revised by A . E . Cowley . Oxford: The Clarendon Press . Kemmer , Suzanne 1993 . Get from Suzanne! Kuen , Heinz 1957 . Die Gew?hnheit der mehrfachen Bezeichnung des Subjekts in der Romania und die Grunde ihres Aufkomments . In G?nter Reichenkron et al . (eds . ) Syntactica und Stilistica: Festschrift fur Ernst Gamillscheg zum 70 . Geburtsag . T?bingen: Niemeyer . Kuno , Susumo 1976 . Subject , theme , and the speaker's empathy -- A reexamination of relativization phenomena . In Li (ed . ) 417-44 . Kury?owicz , Jerzy 1968 . The notion of morpho(pho)neme . In Winfred P . Lehmann and Yakov Malkiel (eds . ) Directions for historical linguistics , 65-81 . Austin: University of Texas Press . Lambrecht , Knud 1980 . Topic , French style: Remarks about a basic sentence type of Modern Non-Standard French . BLS 6 , 337-60 . Berkeley: BLS . Lambrecht , Knud 1981 . Topic , antitopic and verb-agreement in non-standard French . Pragmatics and beyond vol . II: 8 . Amsterdam: John Benjamins . Lambrecht , Knud 1994 . Information structure and sentence form . Cambridge: Cambridge University Press . Lapointe , Steven G . 1987 . Toward a unified theory of agreement . In Barlow and Ferguson (eds . ) 67-87 . Lehmann , Christian 1987 . On the function of agreement . In Barlow and Ferguson (eds . ) 55-65 . Levy , Gideon 1995 . That I should have to defend Arafat? The Haaretz supplement . February 17 , 1995 , 14-16 . Li , Charles N . ed . 1976 . Subject and topic . New-York: Academic Press . Li , Charles N . and Sandra A Thompson 1979 . Third-person pronouns and zero anaphora in Chinese discourse . In Talmy Givón (ed . ) Syntax and semantics 12: Discourse and syntax , 311-35 . New York: Academic Press . Lyons , John 1977 . Semantics . vol 1&2 . Cambridge: Cambridge University Press . Malotki , Eckehart 1982 . Hopi person deixis . In Weissenborn and Klein (eds . ) 223-52 . Mathews , Peter H . 1991 . Morphology . Cambridge: Cambridge University Press . McCloskey , James and Kenneth Hale 1984 . On the syntax of person-number inflection in Modern Irish . Natural language and linguistic theory (1)4 , 487-533 . Meillet , Antoine 1912 . L'?volution des formes grammaticales . Reprinted in Meillet 1958 Linguistique historique et linguistique g?n?rale . Paris: Champion . Miklosich , Franz 1868 . Vergleichende grammatik der slavischen Sprachen . Reprinted , Heidelberg , 1926 , (Wien: Braunm?ller) . Milroy , James 1992 . Linguistic variation and change . Oxford: Blackwell . Mithun , Marianne 1986a . Disagreement: The case of pronominal affixes and nouns . In Deborah Tannen and James E . Alatis (eds . ) Language and linguistics: The interdependence of theory , data , and application . Proceedings of GURT 1985 , 50-66 . Washington , DC: Georgetown University Press . Mithun , Marianne 1986b . When zero isn't there . BLS 12 , 195-211 . Berkeley: BLS . Mithun , Marianne 1988 . Lexical categories and the evolution of number marking . In Hammond and Noonan (eds . ) 211-34 . Mithun , Marianne 1989 . Historical linguistics and linguistic theory: Reducing the arbitrary and constraining explanation . BLS 15 , 391-488 . Berkeley: BLS . Mithun , Marianne 1991 . The development of bound pronominal paradigms . In Winfred P . Lehmann and Helen-Jo Jakusz Hewitt (eds . ) Language typology 1988: Typological models in reconstruction , 85-104 . Amsterdam: John Benjamins . Mithun , Marianne 1996 . New directions in referentiality . In Barbara Fox (ed . ) Studies in anaphora , 413-35 . Amsterdam: John Benjamins . Moravcsik , Edith A . 1978 . Agreement . In Greenberg (ed . ) vol . IV , 331-374 . Moravcsik , Edith A . 1987 . Agreement and markedness . In Barlow and Ferguson (eds . ) 89-106 . Morris , Celia 1994 . Bearing witness: Sexual harassment and beyond -- everywoman's story , 36-38 ; 38-43 . Boston: Little , Brown and company . Moscati , Sabatino , Anton Spitaler , Edward Ullendorff and Wolfram von Soden 1969 . An introduction to the comparative grammar of the Semitic languages . Wiesbaden: Otto Harrassowitz . Munro , Pamela . 1974 . Topics in Mojave syntax . University of California , San Diego doctoral dissertation . Nadasdi , Terry 1995 . Subject NP doubling , matching , and minority French . Language variation and change . (7)1 , 1-14 . Newmeyer , Frederick J . 1991 . Iconicity and generative grammar . Language 68(4) , 756-796 . Newport , Elissa L . and Richard P . Meier 1985 . The acquisition of American Sign Language . In Slobin (ed . ) Vol I . 881-938 . Odlin , Terence 1986 . Markedness and the zero-derived denominal verb in English: Synchronic , diachronic , and acquisition correlates . In Eckman et al . (eds . ) 155-68 . Paul , Hermann 1886/1995 . Prinzipien der Sprachgeschichte . London: Routledge . Perkins , Revere D . 1992 . Deixis , grammar , and culture . Amsterdam: John Benjamins . Philips , Susan U . 1984 . Contextual variation in courtroom language use: Noun phrases referring to crime . International journal of the sociology of language (49) , 29-50 . Philips , Susan U . and Anne Reynolds 1987 . The interaction of variable syntax and discourse structure in women's and men's speech . In Susan U . Philips , Susan Steele and Christine Tanz (eds . ) Language , gender , and sex in comparative perspective , 71-94 . Cambridge: Cambridge University Press . Reinhart , Tanya 1981 . Pragmatics and linguistics: An analysis of sentence topics . Philosophica (27)1: Special issue on pragmatic theory , 53-94 . Rizzi , Luigi 1982 . Negation , wh-movement and the null subject parameter . In Luigi Rizzi (ed . ) Issues in Italian syntax , 117-84 . Dordrecht: Foris . Rodr?guez Bou , Ismael 1952 . Recuento de vocabulario espanol , Vol III . Rio Piedras: University of Puerto Rico . Ross , John Robert 1970 . On declarative sentences . In Roderick A . Jacobs and Peter S . Rosenbaum (eds . ) Readings in English transformational grammar , 222-72 . Waltham , Mass . : Ginn and Company . Sadock , Jerrold M . and Arnold M . Zwicky 1985 . Speech act distinctions in syntax . In Shopen (ed . ) , vol . I , 155-96 . Sanford , Anthony J . and Simon C . Garrod 1981 . Understanding written languages . Chichester: John Wiley and sons . Shopen , Timothy (ed . ) 1985 . Language typology and syntactic description , vol . I: Clause structure and vol III: Grammatical categories and the lexicon . Cambridge: Cambridge University Press . Silverstein , Michael 1976 . Hierarchy of features and ergativity . In R . M . W . Dixon (ed . ) Grammatical categories in Australian languages , 112-71 . Canberra: Australian Institute of Aboriginal Studies . Slobin Dan Isaac 1973 . Cognitive prerequisites for the development of grammar . In Charles Ferguson and Dan Isaac Slobin (eds . ) Studies of child language development , 175-208 . New-York: Rinehart and Winston . Slobin Dan Isaac 1985 . ed . The crosslinguistic study of language acquisition . Vols . 1 and 2 . Hillsdale , N . J . : Erlbaum . Slobin Dan Isaac 1992 . ed . The crosslinguistic study of language acquisition . Vol 3 . Hillsdale , N . J . : Erlbaum . Steele , Susan 1977 . Clisis and diachrony . In Charles N . Li (ed . ) Mechanisms of syntactic change , 539-79 . Austin: University of Texas Press . Tanz , Christine 1980 . Studies in the acquisition of deictic terms . Cambridge: Cambridge University Press . Taraldsen , Knut T . 1980 . On the nominative island condition , vacuous application , and the that-trace filter . Indiana: Indiana university linguistics club . Terken , J and S . G . Nooteboom 1988 . Opposite effects of accentuation and deaccentuation on verification latencies for Given and New information . Language and cognitive processes 2(3/4) , 145-63 . Thrashner , Randolph H . Jr 1974 . Shouldn't ignore these things: A study of conversational deletion . University of Michigan , Ann Arbor doctoral dissertation . Thurneysen , Rudolf 1892 . Zur Stellung des Verbums in Altfranz?sischen . Zeitschrift f?r romanische Philologie 16 . Elizabeth Closs Traugott and Bernd Heine eds . 1991 . Approaches to grammaticalization . Vols . 1-2 . Amsterdam: John Benjamins . Van Valin , Robert D . Jr 1987 . Case marking and the structure of the Lakhota clause . In Johanna Nichols and Anthony Woodbury (eds . ) Grammar inside and outside the clause , 363-413 . Cambridge: Cambridge University Press . Vendryes , Joseph 1925/1952 . Language: A linguistic introduction to history . Translated by Paul Radin . London: Routledge & Kegan Paul ltd . Wartburg , Walther V . 1969 . Problems and methods in linguistics . Translated into English by Joyce M . H . Reid . Oxford: Blackwell . Watkins , Calvert 1962 . Indo-European origins of the Celtic verb . Dublin: Dublin institute for advanced studies . Weissenborn , J?rgen and Wolfgang Klein 1982 . eds . Here and there: Cross-linguistic studies on deixis and demonstration . Amsterdam: John Benjamins . Windisch , E . 1869 . Untersuchungen ?ber Ursprung des Relativpronomens in den Indogermanischen Sprachen . In Georg Curtius (ed . ) Studien zur griechichischen und lateinischen Grammatik . Vol 2 . Leipzig . Wright , Joseph 1905/1968 . The English dialect grammar . Oxford: Oxford at the Clarendon Press . Yule , George 1981 . New , current and displaced entity reference . Lingua (55) , 41-52 . Zipf , George Kingsley 1935 . The psychobiology of language: An introduction to dynamic philology . Cambridge: MIT Press . NOTES * I would like to thank Joan Bybee , Bernard Comrie , Shlomo Izre'el , Suzanne Kemmer and Marianne Mithun for very helpful comments on previous drafts of this paper , and Gila Batori for statistical analyses . I am deeply indebted to William Croft for criticizing me harshly , proposing to me alternative explanations , and patiently responding to all my valid and invalid counter-arguments . I am convinced that our lengthy correspondence and discussions over this topic greatly improved the quality of my argumentation . Last , I am most grateful to Sandra Thompson , who patiently and meticulously read the paper , made very many insightful comments , and forced me to clarify numerous points which had been quite obscure before . I hope this version reflects how much I appreciated her comments .