This language's endonym is _sabasasaj_. I began its development on 25 June 2005. == Phonology == -- Inventory and romanisation -- Phonologically, there are seventeen consonants and three vowels: p_h t_h k_h p t k b d g s h m n w l 4 j i a_" u We write _r_ for /4/, _a_ (and /a/ [a]) for the central low vowel (/a_"/ above), and digraphs in _h_ for the aspirated stops (optionally superscript h where it's available). Everything else is written as above. -- Phonotactics -- Syllables are (C)(G)V(V)(R), where G is /j w/ and R is a continuant, though in older forms of the language the description (C)(G)V(R) was more suitable. In any case, phonemic sequences of vowels, even exceeding two in length, are common. The glide /w/ is only allowed homosyllabically after velars /k_h k g h/. -- Allophony -- The present description draws the division between allophony and phonological rules in the same place as the romanisation; there is one rule whose status is perhaps arguable. /j w/ are realised as pure fricatives [j\ B] unconditionally, except sometimes in complex onsets when they're a little less fricated. Pure approximant realisations [j w] do not exist. The choice of phoneme symbols is apt for aesthetic and historical reasons and because they same phones are realizations of /i u/. There is a barely perceptible tendency to lax (necessarily single) /i a u/ in the direction of [I 6 U] in unstressed (primarily or secondarily) syllables. Two consecutive similar vowels are realised as a long vowel; _aa_ is [A:] instead of [a:]. Sequences of unlike vowels are diphthongized. Within these diphthongs, /a/ is realised [E] after /i/ and [O] after /u/; a different coloring occurs when /a/ is before /i u/, yielding [& 3_+] respectively (that <&> is CXS, i.e. ash). This coloring is not triggered by /j/ or /w/. The duration of long vowels and two-vowel groups is slightly more than 3/2 the duration of short vowels. As for sequences of more vowels, /i u/ become [j\ B] intervocalically, while /a/ becomes [?\] (which slightly lowers adjacent vowels and backs /a/ to [A]). This applies left-to-right in sequences of more than 3 vowels: /auai/ > [aB&i], which example also illustrates that this gliding precedes the /a/-coloring rule. /k k_h g h/ are palatalized to [c c_h J\ C] before /i j/. /h/ is also palatalized in a coda after /i/. -- Phonological processes -- First, a non-process: nasals do not assimilate in place to following consonants, not even before palatals or velars. In fluent speech, when two vowels abut across a word boundary in a constituent and the one before the word boundary is single, it's common to drop it. The remaining processes are exhibited with the topmost applying last. The lower ones also start to include some morphophonology. In CV(V)C where both Cs are among _ph th kh h_, the former deaspirates to _p t k 0_ respectively. This happens less regularly across greater distances than V, though never across an obstruent. _p ph b t th d k kh g_ become _s s w s s r h h j_ syllable-finally. (We can infer that voiceless stops used to be distinctly released in this position.) Fricatives _s h_ cannot stand before voiced stops _b d g_, and are deleted. Voiceless stops are voiced in the same environment, and by the rule above they then become approximants instead of fricatives. _sh_ becomes _hh_. (_hs_ stays.) In verbal agreement markers, _n_ assimilates to _l_ before _l_, and _m_ to _w_ before _w_. I'm not sure whether these apply elsewhere. Most plain stops voice between vowels, or after a nasal and before a vowel (but not next to glides or liquids), including those on the inner periphery of either component of the verb, and a few inflectional morphemes. _s_ or _h_ before a voiceless stop in such conditions can aspirate it and vanish. (NB that this does not happen in stop+/h/!) There is one narrow, stress-driven, exception to this situation: voiceless stops will not become voiced in the onset of a stressed syllable if the previous syllable onset is a voiced stop as well. Thus _wudi'tinas_ 'I'm being followed'. (I suspect this is on its way out.) Certain morphemes with a final _m_ lose it before some consonants. A frequent example is _gam-_ 'four'. -- Stress -- Stress is predictable, and occurs on one of the last three syllables. To find its position for a given word, segment the word into trochaic feet, starting from the left edge, leaving isolated extrasyllabic syllables if necessary; the main stress then falls on the head of the rightmost foot, with secondary stresses on the heads of the preceding feet. Legitimate trochaic feet are H, Ll, and Hl, with the head capitalized, where H and L represent heavy and light syllables, respectively those with >1 and 1 mora, each vowel and coda consonant counting for one mora. Feet are formed greedily: an H foot only ever shows up if there's not a following L syllable. There is one fairly specific exception: if this rule would place the stress on the final syllable in a rule with a final consonant, but it would fall on the penult if that final consonant were absent, the stress is actually penultimate (and the last syllable is not treated as a foot-head). [Rewrite this for counting from the right; it's simpler that way.] Words with no feet, of course, can't take the stress. This imposes a minimal word constraint on content words, which must be able to take stress -- (C)V(C) words are ruled out. It shouldn't be too surprising that closed short-vowelled syllables are insufficient, in the light of the exception about final consonants in syllabification. For verbs this constraint applies only to the first component, which bears the inflection when the second component is separated, and indeed nearly all (unprefixed) second components violate it. The peculiar-looking word breaking conventions in the verbal system are justified by the stress placement. [But what does this say about clitichood?] [I don't know what the realization of the stress is. I also don't know anything about phrase-level intonation, but questions must be different to declaratives. But see THC's prosody thread on the ZBB.] Phrase-level intonation will break at [or otherwise indicate] the start of a heavy sentence-internal constituent (most such are strongly left-branching). -- Other things to note -- Many exclamations contain extraphonological sounds, and accordingly are well served by other letters in romanization: _oo_ [o:] (a generic call) _ah'_ [AX\] (annoyance) _sa'_ [sa?] 'stop it!' Other exclamations are phonologically unremarkable: _ii_ [i:] 'ouch!' == Morphosyntax of the verb == -- The form of verbal stems -- Verbs are discontiguous. Each verb stem has two components, which I will give separated by a hyphen, thus: _pa-ul_ 'lick' _gi-an_ 'push' For a very few verbs the second component may be empty. This happens for instance with some onomatopoeic verbs. Verbs which leave the second component empty are defective, in that they can't form a perfect (for reasons as below). Speakers cope with this by circumlocutions: so although _kukaridiu-_ 'crow (of a rooster)' is a verb, one has to live with _kukarii kukaridiu suuba sirgas_ 'the cock has crowed', lit. 'has gone _kukaridiu_'. I think of this as characteristic of the older language, though; verbs without second components feel like remnants, even if the reason is that they were only recently verbalised. In the "contemporaneous" standard they are granted second components by extension from other verbs. Some semantically coherent series of verbs, like verbs of taste or verbs of emitting light or most numbers, share a second component. _khu-ta_ 'take', _kara-ta_ 'claim' _kja-ram_ 'taste sweet', _iakhi-ram_ 'taste sour/bitter', _haga-ram_ 'taste bland' When two verbs share a first component, one is often a metaphorical or ~synaesthetic elaboration of the other, and sometimes one exhibits a fused directional that the other has lost. The first component of a verb stem is the more important of the two: it tends to have more phonemic substance, isn't treated as badly by derivation, and has more history as part of the verb word than the second component. When verbs are borrowed the first component gets the borrowed material, and some second component is selected on the basis of semantic similarity. [I rejected verbal borrowings instead being treated as nouns with a light verb as that utterly has no precedent.] There are a few verbs which can be said to have no or very little root material in the first component. These always appear with derivational material (like directionals) to flesh this component out. -- Morphosyntactic alignment -- Verbs cross-reference two arguments. I call the slots for these arguments 'nominative' and 'absolutive' (these aren't cases). The language is split-S in terms of verb agreement, and with a somewhat unusual split between the S_a (nom) and S_o (abs) types. This arises from a peculiar alignment of the two roles of transitives with the two cross-reference slots of the verb. As a guideline, this alignment of the nominative and absolutive arguments is determined by the theta-role of the more prototypically animate argument: agents tend to be nom, experiencers abs. So we'd get '(nom) hits (abs)', but '(abs) sees (nom)'. Many verbs are inherently transitive, and when they're used intransitively the same alignment for the remaining argument is used as would be in the transitive case. Verbs used so will simply drop the noun phrase corresponding to the omitted argument. This dropping is frequent in other circumstances as well, whether or not an argument is cross-referenced; there are no obligatory nominal components of a clause. [Does this dropping really accomplish detransitivisation, or does it just decline to specify a value for the argument? What's the theory here? I really need to read on it. Relatedly: if that's the case, what basis is there for selecting the nom and abs nouns as potential arguments but asserting that PPs are never arguments? Something like referentiality when they're dropped?] When there is no (frequent) transitive counterpart of an intransitive verb, the argument's case is selected somehow. [by agency, or often arbitrarily?] In a number of cases, the change in frequency of a transitive verb, or general considerations of naturality, has meant that both argument marking schemata are valid for a particular intransitive sense. Among verbs whose most frequent use is as stative intransitives with an abs argument, there are two common types of transitive counterpart. One type is '(nom) causes (abs) to be in state X'; another type is exemplified by 'find hot'/'be hot'. Note in particular that 'find X'-type verbs have experiencer as abs and thus the argument of the intransitive is promoted to nom. [does it ever happen that the tr is lost and two dissimilar-seeming intrs remain?] Outside of the verbal agreement system, a more usual alignment of agent~experiencer against patient~stimulus is reflected as well. One manifestation is in the word order of clauses with two explicit nominal arguments; another is in imperatives. [though is it really? 'see this' really wins over 'be seen'? Sure it does; for that there's 'show yourself'.] -- Verbal categories -- [overview sorely needed! subsectioning too.] Verbs show a distinction which combines TAM and active/stative-ness, in the sense of 'denoting an action' vs. 'denoting a state'; this shouldn't be confused with the use of 'active' and 'stative' in describing active languages. The basic verb is imperfect, and its most frequent use is present continuous; this is the active. In this form, person and gender agreement markers occur between the two parts of the verb. [variation, eg. 'I have 120 heads'] -- Finite agreement markers -- Here are the standard agreement morphemes. Bracketed stuff appears when phonotactically necessary, but we always prefer leniting a syllable-final stop to vowel insertion. When two insertions are equally good _i_ takes precedence over _a_; when this doesn't resolve the matter an _a_ after both markers is better than one between them. Nominative: _(i)n_ 1sg _(i)m_ 1pl (there's no incl/excl distinction) _li_ 2 (all numbers; once formal?) _th(a)_ 3sg human _s(a)_ 3pl human _k(a)_ 3sg animate _ja_ 3pl animate _u_ 3 inanimate (all numbers) Absolutive: 1 and 2 are the same as the nominative. _is_ 3sg human _(a)wis_ 3pl human _i_ 3sg animate _(a)wi_ 3pl animate _d(a)_ 3sg inanimate _(a)w_ 3pl inanimate The 3sg animate still alternates with _g(a)_ in leniting positions. In the dentals there was once a _t ~ d_ alternation but in the current language these have split into different functions. If only a single 1 or 2 argument is marked (this is never used with a 3rd person argument), the above scheme doesn't tell us whether it's nom or abs. If it's abs, by default we insert _i_ before the person marker; the verb may specify its own insertion here, which we write in parentheses. I call this insertion the absolutive augment. (Note that the intransitive nominative forms are probably odd things to say, without pragma.) _gi-an_: _ginan_ 'I push', _giinan_ 'I am pushed' _gwa(hi)-i_: _gwani_ 'I frighten', _gwahini_ 'I fear' _ba(l|da)-uh_: _balnuh_ 'I harm', _badanuh_ 'I am harmed' One verb is completely suppletive: _(ja|thulu)-pi_: _janbi_ 'I kill', _thulunbi_ 'I die' When both nom and abs arguments need to be marked, the nom precedes the abs. Some combinations are not as given above. Starred ones only occur where phonotactics permits, but retain the _t_ in place of _d_ in otherwise regular forms. _t(a)_ 3sg human nom, 3sg inanimate abs *_st_ 3pl ... *_ht_ 3sg animate nom, 3sg inanimate abs *_jt_ 3pl ... Masses take plural agreement. -- Subordinate verbs -- *Subordinate verbs* are an important class of verbal words, essentially participles, although they are closely integrated with the rest of the verbal system. They formally differ in employing an special class of agreement markers, which can be characterised as relative. These nominalisers signify that the phrase this verb are part of is referential to the role the nominaliser play in it; they do *not* occur in any other cases of coreference. Subordinate verbs are fairly strongly dispreferred _modifying_ a noun, though. This means one can't usually directly render HERE; normally one does an internally-headed relative clause type of thing. The relevant agreement marker is _a_ sing _i_ plur except when the replaced marker is abs and there is no nom (the same situation where the absolutive augment shows up), when instead we get _ta_ sing _ti_ plur. The _t_ is non-voicing. This is identical to the choice of markers used in a large class of nominalizations. The resulting subordinate verb often looks very much like a nominalization (discussed below), differing mainly in the unmodified second component, and the potential to be in the perfect. Note however that verbs in syntactically subordinate positions are often not subordinate in this morphological sense, including in the frequent case where the head is to be taken as a topic. When the superordinate clause with an argument coreferring to an argument of a subordinate verb is higher than one clause away, the lower verb invariably takes a particle indicating this. This particle is _a_ when it's a nom in the higher clause, _u_ when it's an abs. The argument of the verb which these refer to is marked with the verbal agreement system, not with nominalizers. [examples!!] When there are many clauses above this one, _a_ and _u_ will preferentially refer to arguments of a closer clause. [Maybe this system can be used in the short-distance situation too, where we'd ordinarily just use nominalisers. see my Conlang post of 22 Dec 2008.] -- Other argument structures -- [what, if anything, does indefiniteness/nonspecificity do? Nothing, I think; there's not any signalling of definiteness or specificity. Check how this should work wrt incorporation, at least.] The nominative and absolutive agreement positions never both agree with the same first or second person participant. For semantically reflexive clauses, a medio-reflexive sense can be gotten with directionals (see below). There are no verbs with more than two core arguments. What would be an indirect object, if a SAP, can be rendered using the directionals 1 _s(a)_ and 2 _i_, q.v.. [handling of more-than-two-argument verbs? otherwise quirky argument structures?] There are no zero-valent verbs. For weather and other such ambient conditions we have _irua barpi_ 'it's sunny' _dimisna barpi_ 'fishing is going on' (note _dimisna_ is still intransitive) -- Agreement selection -- The humanness and animacy distinctions in the agreement markers are made entirely semantically; gender is not a feature of the noun. Humans of any culture or age or sex merit human agreement, as do other humanoid sentient beings of legend. Other things which are seen as capable of acting with volition, or that are prototypically instruments of volitional actors, get animate agreement: this includes animals and human institutions and body parts and certain tools, but not plants, say, or fire or water. Everything else is inanimate. Deceased humans retain their human agreement, out of respect; dead animals vary between animate and inanimate, depending on their role (food for instance is inanimate). Nonprototypical assignment of humanness and animacy is somewhat narrow, but e.g. inanimate objects which in stories exhibit properties beyond their noun class will often be assigned to a more alive class than their ordinary kin, usually the animate (though there's variation here). This rule has the functional value that there are fewer things the overburdened animate agreement can point to running around in stories where it applies [I did make the rock inan in relay 15, though]. [It remains to be seen what the conculture has to say about the existence and character of spirits, gods, etc., in this respect. I should think there won't be a belief that natural forces are executing the will of anything in their day-to-day behaviour, so that they get inanimate agreement.] Finite verbal clauses as arguments select inanimate agreement. -- More verbal stuff -- A serial verb-like construction is obtained by entirely suppressing the second component of each of the non-final verbs in the chain. All verbs in the chain take agreement markers. This can be considered a coreference strategy; it makes two verbal words corefer. [I invoked these in the LCC2 relay.] In serial chains in the perfect, the whole perfect complex is omitted from nonfinal verbs. The verb _sit-pa_ 'say' when used to introduce reported speech (see a fuller treatment below) often comes in non-final serial position, and more generally appears thus for ideophonic sorts of functions. In fact in the latter function it can either appear or be omitted; it's unclear if anything conditions the variation. [See also the slurping text.] An iterative or perhaps pluractional form exists. It expresses various things ranging from iterative or gnomic through frequentative through even action repeated a number of times (which can amount to a reflection of distributivity of a participant; these latter senses do not however extend to simple plurality marking of said participant). It is formed by reduplicating the first CV of the first part of the verb root; this ignores (and hence goes inside) synchronic verb prefixes. In the perfect this reduplicant still belongs to the first component. _hamphunii_ 'my back is itchy (now)' _hampuphunii_ 'my back is always itching' and _huis siwthuaaawniw kunridul_ (man.PL chest hold-upward-3shn-3pia) 'the men lift the chest', assumedly all together _huis siwthuaaawniw kukunridul_ (ditto w/ redup) 'the men each lift the chest' Glides may be lost in the reduplicant in fast speech, if it's unstressed (so _gagwasni_ 'he always scares me'); this is not standard. -- Perfects -- A perfect (which is stative) is made by detaching the second part of this imperfect construction and placing it before the verb, generally prefixed by a semantically-determined perfect marker. _am_ 'until the end', cessative-like _pak_ 'with (often destructive) force' _i_ 'ahead, out (in a single direction)', generic _um_ 'on (contacting the surface)', 'with only minor effect' _bi_ emphasis on the imperfect action; particularly useful to avoid an extended sense of the usual perfect (more in lexicon) Many perfect markers are identical to, or at least cognate with, directionals, though their senses are broader. For truly spatial meanings, verbal directionals tend to be expressed as the perfect marker in perfect clauses; the more idiomatic the combination of directional and verb, the less likely they'll separate. _sanniaj ian ginda_ 'I pushed (have pushed) the door open' 'have', invariably when alienable and sometimes when inalienable, is expressed as the perfect of _khu-ta_ 'take, acquire'. A perfect marker is never used in this meaning. For specifically expressing 'have taken' rather than 'have', the perfect marker _bi_ is useful: _kaakhasali bida khunda_ 'I have taken your torch' -- Preverbs -- [I should provide an introduction to preverbs. I also may want to refine this, especially the classifierly ones, in view of the incorporation hierarchy.] Preverbs never cause voicing alternations. A number of uses of preverbs are worth distinguishing, with respect to their productivity and frequency of use. (1) Lexical compounding, supplying a generic argument to the verb to describe a conventional activity or state. Sometimes this causes detransitivisation or rearrangement of argument structure. This is productive, though not as common as it could be, given the limited selection of prefixes; there is far from one for every noun. _suhuum-un_ 'make calm', _phinsuhuum-an_ 'quiet an infant' Some common cases are (1a) Indication of the body part affected or used. _tha-nu_ 'tie', _kaatha-nu_ 'tie the hands' (or 'by hand'), _numtha-nu_ 'tie the feet', _hamtha-nu_ 'tie the tail' _iakhisu-is_ 'stand on a point', _numiakhisu-is_ 'stand on one foot' _khiri-ta_ 'pinch', _kaakhiri-ta_ 'pinch with the hand' _phian-tu_ 'be 120', _kharaphian-tu_ 'have 120 eyes' (1b) Indication of location. (2) Classification of an argument, without affecting the argument structure. This can happen (2a) referentially, i.e. to indicate that the argument is identical to a fairly topical entity referred to somewhere else. [This isn't a very plausible use, per Mithun. I should revise.] But topic pronouns are preferred to this; the norm is non-referential use of classifiers. The frequent classificatory uses are (2b1) Use with the empty root, often to fulfil the constraint that the empty root must carry at least one of a preverb and a directional. (2b2) Use with a small collection of verbs of physical manipulation. (2b3) Selection of an often metaphorical sense of the basic root appropriate to humans, with the prefix _ta-_. (3) Classification of the referent of this verb itself. Not actually found on verbs, as they generally don't refer to things, but prevalent on nouns. When this is found on verbs, generally there isn't agreement marking for the relevant slot, and the referentiality of the object is decreased as is utual for incorporation. _wudi(ti)-as_ 'follow', _ljawudias_ 'tail' There are a few preverbs particularly suited to classifying abstract nouns. -- Directionals -- [directional elements after the first part of verb are long decided and just need writing up. Check the relay writeup.] Among potential spatial contrasts, containedness is basically ignored in directionals. Most directionals have both stative and dynamic senses; some have only one or the other. If it is desired to have the senses of two directionals, a serial construction with the empty verb to stack them is what's usual: _haumaa siarua gi phu-*sug*-i-0 tai-*nsi*-h-lu_ 'a bird flew *passing in front of* a tree' Directionals are not applicative in their semantics, i.e. their basic function is not to add an argument syntactially. They can add semantics of motion to otherwise quite motionless verbs without adding an argument: _diman ias daninnii_ 'one fish got through' It's quite common to have an adpositional phrase in a clause providing the reference point of the spatial sense, generally the most bleached adposition _gi_. This phrase is not a core argument (as is generally true of PPs). _sanniaj gi ian ginniin_ 'I have been pushed (out) through the door' _sanniaj gi sugu khulugin_ 'I'm sitting far from the door' In their productive uses, directionals have less semantic extension away from motion or position than perfect markers. Many perfect markers are semantically broad and overlap motion or position somewhat, and some of which are identical with directionals of narrower sense. Directionals can be required with certain verbs. In some cases the verb-directional compound has fused, and the directional semantics they contribute can combine idiosyncratically with particular verbs. Directionals can give a causer interpretation to the nom which the base lacks, there being no dedicated causative morphology in the language, and in this case they often do add an argument. Verbs that are formed this way are annotated (cd) in the lexicon. _dju-ihi_ 'be oily (nom), find oily (tr)' _djuhi(ri)-ihi_ 'smear with oil, anoint' Likewise, a sort of middle or reflexive can be derived for a transitive verb by using it intransitively with directional _s(a)_. Either the nom or the abs can be the retained argument, according to degree of agentivity; there are a few verbs for which this approaches fluid-S. Note that _s(a)_ has a special absolutive augment, with which it appears as _phi_. _tahallida_ 'I am hugging you' _tahasinda_ 'I am hugging myself' _kharanjunjulinbaa_ 'you drive me mad' _kharanjunjuphinbaa_ 'I drive myself mad' When used with an abs intransitive with an inanimate argument, it can suggest the non-involvement of a nom, in contrast to the clause with _s(a)_ omitted. _mjiru namandaan pagih sigas_ 'the child cracked the pot' _namandaan dauih sihsar_ 'the pot cracked (with no evident cause)' If there's already a causative directional, _s(a)_ replaces it: _suhuumilliun_ 'I am calming you' _suhusnun_ 'I am calming myself' (actively) cf. with a different directional _suhulinun_ 'you find me calming', _suuphinun_ 'I find myself calming' As another example, the base _djuhi(ri)-ihi_ 'smear with oil' derives _djus-ihi_ 'smear oneself with oil' [Approximate first and second person senses rendered by _s(a)_ and _i_.] In common speech these will occasionally appear with another directional; this is the first crack appearing in the unity of the multifunctional directional system. -- More verbal bits -- The simple imperative is just both parts of the verb with no other marking necessary, though patient marking is permissible, in which case the imperative will generally be identical to a one-argument finite form. _nup-lu_ 'run', _nuslu_ 'run!' Regular absolutive augments do not manifest themselves in imperatives, even where the participant being imperated gets abs agreement. Whether irregular ones do is a matter on which there is variation. One noteworthy case is _thulu(dau)bi_ 'die!' vs. _ja(dau)ispi_ 'kill him!' A few old imperatives survive consisting of solely the first part of the verb, with an extension _-i_ (or maybe not, but certainly if it's necessary to guarantee bimoraicity). These are not productive and limited to only a few verbs, and do not allow usual argument marking. _khu-ta_ 'take', _khui_ 'take it!' This simple imperative, formed either way, is quite direct. More indirect polite constructions are usual [what, though?]. The negative marker is _kha_. It appears just before the second component. In the present continuous it is preceded by a word break. In the perfect its presence tends to discourage perfect markers from appearing; when they do appear they are prefixed as part of the same word before the negative marker. Thus in the usual construction only word order distinguishes the two TAM forms: _ballin khauh_ 'you are not hurting me' _khauh ballin_ 'you haven't hurt me' Negative imperatives are, analogously to positives, formed by omitting person marking from the negative construction. _badan khauh_ 'don't hurt me!' A desiderative 'want to X' is formed by preposing the particle _mia_ as a separate word, inside any perfect particle. _diman mia khunita_ 'I want (= want to get) a fish' Other irrealis particles in the same syntactic position are _thiu_, _khal_, both most common in subordinate clauses. I think they partition semantic space according to whether whoever the clause is empathising with wants the event to happen. A fairly central example for the one is _thuwabi thiu jawis gwauni_ 'I fear they have been killed' -- Special verbs -- There is an empty root _-pi_, taking absolutive arguments, whose basic meaning is something like 'exist, exist at, go/come'. It can never appear with a completely empty first component, and never takes a nonzero perfect marker. It most frequently takes a directional, to express 'be in position X' or 'go to position X'. _haumia ruwibi_ 'some birds are up high' The location is still not a core argument: _haumia siarui gi ruwibi_ 'some birds are high above the trees' If a directional is not used then a verb prefix must be, classifying the sole argument. If there's no better prefix to use _na_ will do. _siarui khiwpi_ 'some trees exist' If both a directional and a verb prefix are used, the verb prefix can classify either the absolutive argument or the location; this has some ambiguity, but it's generally not a problem when an explicit NP is provided. _hwinniwpi_ 'water is going through' or 'it is going through the water' The empty root is defective in not forming an iterative, having nothing to form it with. This gap is filled by the use of the aspectual root (just below), as _bubuk-u_. The aspectual root has the form _b(a)-hu_, but the _h_ is deleted except initially in the perfect, so that it looks like _b(a)-u_ but still selects preconsonantal allomorphs of the agreement morphemes. It is used to express process phases, such as 'get X/start Xing', 'stay X/keep Xing'. The root tends to take one (3rd inanimate) abs argument, which is the clause it's modifying. It also often takes a controller as a nom argument, but this controller is still usually echoed in the subclause. [This leads to matters of raising or control. What can I say about them?] The derivatives of this root are made using directionals, following the primary metaphor 'in/at' is 'not happening', 'away from' is 'happening' and derivatives, and lesserly the metaphor 'up' is 'to a higher intensity', 'down' is 'to a lower intensity' and derivatives. There are a number of directionals commonly used; some less frequent ones are also possible. _bi-u_ inceptive 'get X, become X, start Xing, begin to X' (< 'ahead, out') _buk-u_ 'stay X, keep Xing' (< 'far from') _bas-u_ cessative 'become not X, stop Xing, finish Xing' (< 'in') _bum-u_ 'stay not X, keep not Xing' (< 'on the surface') .. _bari-u_ 'get more X, start Xing more intensely' (< 'up') _baru-u_ 'be really X, X intensely, X out' (< 'high') _basu-u_ 'get less X, start Xing less intensely' (< 'down') For example _nubinlu bi-n-da-u_ 'I start running'. [I took out _nsi_ and _gu_ for "failing aspects".] There are no derivatives of the aspectual root which are basically perfective, or view events as complete wholes. The aspectual root is also not used for a simple 'do'; this is a different verb, _sai-hu_. But there are a few other roots which behave like the aspectual root but add other semantics, like _phuri-u_ which adds suddenness. This has the exceptional form _phuriaa-u_ for a pure punctual sense. -- Reference tracking -- Each block of discourse has a topic. The topic is not explicitly marked on introduction. In the overwhelming majority of cases the topic is made to be the only topicworthy argument of the main verb. This may require some extra syntactical contortion to be arranged, frequently a cleft. But if, say, neither argument of the main verb is topicworthy but one of the main verb's arguments has a topicworthy argument, that will be the one chosen. (Also, if the content of the sentence is explicitly topic-establishing -- "lemme tell you about X" -- there's no need for contortion.) Subsequent sentences may mark which, if either, of their main arguments is the topic, by supplying what is essentially a pronoun. If the topic is not relevant and this sentence is establishing a new one, then it's common to signal this by supplying an expression of setting or time. When one of these new-setting words appears but the topic is to stay the same, it's typically clefted. These topic pronouns manifest especially often in long sentences, or when ambiguity demands their presence (e.g. the topic is human and the verb has two human markers). They are also, it seems, more likely to occur with the main verb of the sentence. Topic pronouns appear just before the verb, in the same slot as the long-distance nominalization references _a u_, and inside mood. The particles themselves are _ha_ topic is the nom argument _hu_ topic is the abs argument These topic particles are the language's solution to the problem of identifying a referent through long sequences of clauses. Besides them there are no other mechanisms for this with scope greater than one clause. [There still may be pivoting mechanisms within conjoined clauses or something, but conjoined clauses are dispreferred.] To repeat myself, here is the order of certain particles in the verb phrase: 1. perfect complex 2. mood (irrealis, desiderative) 3. _a u_, _gia giu_, _ha hu_ 4. verb == Morphology of the noun == -- Deverbal nouns -- The vast majority of nouns are related to verbs by a single scheme of derivation, which is by far the most productive method of derivation in this language. The direction of derivation is usually evidently from verb to noun, so we'll call the relations we'll be describing below deverbal nominalization. It's likely that some derivations historically went the other direction, even though they fit the nominalization schema (perhaps because the nouns have been reshaped a bit too): _butha_ 'Moon', _buth-il_ 'shine moonlight' _irua_ 'Sun', _iru-il_ 'shine sunlight' which can even so do verby things like _kharabubutithil_ 'her eyes cast beams of moonlight' Nouns look a little like finite verbs in the (unmarked) imperfect, with two differences. Firstly, the second component of the verb is generally dropped, or sometimes modified (but sometimes left untouched): if modified then often it's denied its final vowel. Secondly, in place of one of the person agreement markers, we have a nominalizer. The most frequent nominalizer _a/i_ also marks subordinate verbs, and in fact nouns and subordinate verbs should be viewed as two subclasses of the same word class. [Are there semantic correlates of the changes to the second component? I'm speculating in particular whether changing it to _-a_ yields instrument-type semantics.] This nominalizer, singular _a_ / plural _i_, can nominalise on either the abs or nom argument, whichever yields a more useful result (for a given stem, the choice is lexically fixed). The _a ~ i_ alternation marks nominal plural, not plurality of the verbal argument. _ta_ pl. _ti_ is used as a transitive abs nominalizer when the nom is fundamental; alternatively, it just emphasises the transitivity, foregrounding the roles of both arguments. _ta_/_ti_ cannot cooccur with a nom marker from the verbal concord system. _tapuia_ 'married person' (not possessed), _tapuita_ 'spouse (of)' (nearly always possessed) _ta_/_ti_ can also be resultative (because the TAM system isn't available to make the distinction): _baltau_ 'wound (n)'. _i_ is another nominalizer; its function is not yet clear, though it seems to be common for deriving animate but non-human nouns, and nearly universal in animal names. It (always?) selects a totally empty second component. It typically takes _-win_ in the plural. _tha_ pl. _thi_ is used for human actors as nom argument, especially when the unmarked nominalization is taken. [example] _d_ seems to be used in some nominalizations with singulative force. They pluralise normally. _thiuijaga_ (v) 'they come swarming' _thiuida_ pl. _thiuidi_ (n) 'locust' but, per the next derivation, _thiuiu_ 'swarm of locusts' _u_ forms some mass nominalizations. Masses take plural agreement. _ukhaisuurli_ 'it burns (down) to leave it (ash) behind' _ukhaisuu_ 'ash' Alternation between the singular and plural nominalizers is the regular way of marking nominal number. So _phinkaramdata_ (v) 'we claim it as ~an inheritance' _phinkarama_ pl. _phinkarami_ (n) 'hereditary possession; ~right (loosely)' (_phin-_ 'parents, ancestors' ,_kara-ta_ 'claim (as one's own)') _sanniuju_ (v) 'it lets (things) go through' _sanniaj_ pl. _sanniij_ (n) 'door' (_sa-ju_ 'let go', _nni_ 'through') Many personal names are nominalizations of verbs. These typically replace the second part of the stem with _(a)m_. So _pulam_ from _pul-as_ 'unite', _siwtam_ from _siw-il_ 'be dear'. A common hypocoristic pattern replaces this _m_ with _ssi_, and applies the other sympathetic modifications of the diminutive (of course, there are a number of variants). Other nominalizers are _du_ with no special plural, and _dua_ sg. ~ _dui_ pl. These nominalizes on the spatial setting of the event, usually the place but also potentially the path or the motion, or indeed the temporal setting where this makes more sense than the spatial. _phinisam_ (v) 'he is born' _phinduih_ 'birthplace' It is atypical to make a true noun from an iterative. Nouns already have the time-stable sense that the iterative would generally contribute, plurality has other modes of expression, and distributive nuances are generally taken up by directionals. [What of derivation starting from nouns? It's probably a marginal phenomenon, but tries to use the same material as deverbal nouns. I don't entirely understand this question, though.] -- Other nouns -- Where the nominalizer has no plural or is absent or not identifiable, there are other pluralization strategies. A generally common one is -_(w)in_. _mjaai_ pl. _mjaaiwin_ 'cat' Some nouns have peculiar plurals; these are among the clearest specimens of nondeverbal nouns. For instance, there's _sulu_ pl. _suwis_ 'hand', which looks manifestly related to _suu-ta_ 'grab', but which can't be a nominalization therefrom. -- Possessives -- There are two species of possessive marking. The least remarkable is formed by a series of suffixes reminiscent of verbal agreement markers. _na_ 1sg _ma_ 1pl _lia_ 2 _ta_ 3sg _ua_ 3pl (all classes) So _mjaaina_ 'my cat', _suluta_ 'his/her/its hand'. Pluralization of the base occurs where natural: _suwisua_ 'their hands', cf. _suluua_ 'their (collective) hand'. Many nominalizations instead show *internal* possession, by using appropriate verbal agreement markers alongside the nominalization marker. This is especially true of nominalizations with _t_, which in their basic form can also be taken as 3s possessed, and have no explicitly unpossessed form. _dauitah_ 'his/her/one's/someone's/a face', _dauinah_ 'my face' _thipulta_ 'his/one's/someone's wife', _thipullia_ 'your wife' Note especially the cases in which the internal possessives merely replace a final -_a_ and thus look very suffixal. This second strategy is declining overall, but remains strong for canonical possessables and most _t_ nominalizations. Indeed there is some bleeding between the two strategies, and they probably have the same origin (observe that the marking on _thipulta_ is formally identical to what the first strategy would give). The 3pl marker is especially uncomfortable internally, and can be found externally (_dauitahua_) even where other markings are internal. [Maybe _ua_ gets encroached on by _ta_, relying on plurality of the base?] Both of these possessives, when they can't be interpreted as members of the verbal system, are pretty much restricted to concrete nouns; what might be rendered with abstract nouns generally have more verbal argument-taking patterns. Also note that many nouns don't select possessive marking on their arguments at all. This isn't really a comfortable situation for nouns, though, and so given a deverbal nouns whose semantics don't differ much from the word's verbal original, an expression with the verb will be preferred to attempting to form a possessive-marked noun. _daguthih_ 'merchants' _haumia dadaguiwiam_ 'bird merchants' -- Other schemes of derivation: verbal -- There are verbs with second component _-innu_ imperf., particleless _nu_ perf., nominal truncation _-in_, rendering the sense 'be something/one which habitually X'. Compared with the iterative, this derivation tends to name a formalized role or title. The argument defaults to abs, but many verbs override this, especially if they introduce a second core argument or contrast with another _-innu_ derivation: _dimthana_ 'he is fishing' _didimthana_ 'he fishes regularly' _nu dimtha_ 'he is a fisherman (by vocation)' This last one with the abs marker (_nu dimis_) would render 'he is a fish'. Cf. also the noun _dimthan_ 'fisherman'. A formal noun for the likes of 'someone who happens to be fishing now' can't be gotten, but given the insubstantiality of the noun/verb distinction this is not a problem. For some verbs, the formally unmarked nominalizations exhibit narrowing in sense. Many such verbs have _-innu_ derivations with the same narrowing, and the ones that do often permit a related sense even for the underived verb: _hwasi_ 'pig' _hwasiniuh_ 'I am grunting', also used as 'I'm acting like a pig' _nu hwasin_ 'I am a pig' _hwasininnu_ 'I am becoming a pig' [This can't be the only derivation whose form is a replacement of the second component. What, then, are others? Perhaps 'be fit to', which might also serve for 'tendentially do' with iterative?] There are about no forms of compounding looser than the use of a verb prefix. There are certainly fixed expressions for particular referents, but their syntax is no different than would be expected in a less fixed description. _aagariwia insiasuui_ 'fallow deer', lit. 'pronking (= four going up from the ground) deer' [rework this, in view of _mu_ -- why?] In particular there are no forms of words which occur uncomfortably in these fixed expressions (like plurals in English compounds do). -- Other schemes of derivation: nominal -- There are nominal augmentative and diminutive derivations. Each of their main exponents is a suffix: diminutive -_si_ and augmentative -_thun_. However, both allow sympathetic phonetic modifications to a prominent syllable of the root: for the diminutive this is inserting a _j_ onset glide, for the augmentative inserting an _n_ coda in a previous syllable, and backing the vowel of the syllable, if it happens to be _i_, to _u_. These modifications give a more intensified (colloquial? involving-the-listener?) feel. _siarua_ 'tree', _siaruathun_ 'big tree', _sianruathun_ 'BIG tree' == Other word classes == -- Pronouns -- The independent singular personal pronouns each have two stems, a shorter one used with an adposition and a longer one used elsewhere. 1s in, ijan 2s phan, phawan 1p muru 2p phuru The (longer) pronoun is not required in clauses with SAP arguments, and has emphatic force when it appears. Third person pronouns don't exist. Among the features which help to fill their place are topic-markers _ha hu_, long-distance nominalization particles _a u_, their clusters with pronouns, and preverbs. -- Determiners -- [Figure out deictics. I've used forms _naa_ 'this' pron., _anaa_ 'this' det., _puanaa_ 'this fact or situation', _hwaa_ 'here', _phjaa_ 'so' without a system containing them.] -- Quantifiers -- These are things like _nabas_ 'other', _talas_ 'every', _khas_ 'no'. [A list with pretensions of completeness would be nice.] Quantifiers can modify finite verbs, giving the semantics of e.g. 'every time X happens'. This is generally not done on the main verb. When used with negative quantifiers this does not induce negative concord on the verb. -- Sentential adverbs -- These form a tightly closed class. They do *not* include the sort of derivation from the adjective that SAE considers the canonical adverb, and hardly could in our (near-)absence of the class 'adjective'; but there are some members of the class with the same semantics, such as _hwusuu_ 'fast, quickly'. The unmarked position for sentential adverbs is clause-initial [refer to the LCC2 text]. Many sentential adverbs are not limited to modifying a VP, but can go elsewhere. [exs] Probably representing a larger cohort, _danu_ 'only' is derived from a verbal stem with the suffix _-u_, which was historically the 3rd inanimate nom marker. -- Exceptional simple adjectives -- The only words in Sabasasaj which have a claim to being adjectives are four colour terms. They are not the whole slate of colour terms in the language, but the others are newer and are regular stative verbs. They are _kwui_ 'black', _nari_ 'white', _thaw_ 'red', and _thjassi_ 'yellow, green' which is a diminutive of 'red'. In attributive position these words only modify nouns, which as expected they precede. For predicative use, as expected, they've been verbalised in the experiencer pattern, with second component _-il_ [compare this to the 'shine foolight on' sense it has elsewhere]. -- Adpositions -- There are a very few postpositions which are tightly bound to their head (not always a noun). I'm not sure whether they can be called clitics, as they have no phonological effects on the head and the syntactical tests I know don't apply. They don't bear stress themselves. There are no spatial adpositions with precise spatial senses among them, since the verbal directional system already handles these distinctions quite adequately, and finer-grained spatial distinctions are made with words that are still analysable as verbal. Indeed, the numerical majority of adpositions are most frequently, or only, adverbial clause markers. _gi_ is the semantically empty spatial preposition, indicating the point of reference of a directional. It can also render generic spatial-locative senses when there is no directional, but this reading is essentially unavailable when the _gi_ phrase is directly the argument of a verb. _ba_ is the analogue of _gi_ for temporal location, and more broadly 'when X', 'in the condition that X is true'. This is used for hypothetical and counterfactual constructions 'if X', with irrealis verbs. _tusjudu ba igwis luadih_ 'in case of fire, break glass' _din_ renders reason clauses 'because X' with the realis and purpose clauses 'in order to X' with the desiderative. With nouns it can be a benefactive 'for'. _haumia mia gwasuuwwii din siarua gi thjitiriwwus_ 'we're throwing (rocks) up into the tree to scare the birds away' _siarua gi paus thjitirimaw din haumia pi suuh_ 'the birds have dispersed because we have thrown (rocks) up into the tree' _pi mia iili in din?_ 'would you go for /my/ sake?' [recipient?] Postpositions form clusters with the long-distance subordination reference particles _a_ and _u_, in which the postposition appears in second position, e.g. _gia_ 'to it' (nom in a higher clause). In this use the reference need not be long-distance. _giu liaiaur thipulta judatikha_ 'he heard his wife (as she was) calling out to him' There are no analogous forms for the topic particles _ha hu_. -- Conjunctions -- The principal conjunctions are _im_ 'and', emphasizing parallelism _us_ 'but', emphasizing contrast These apply to both NPs and VPs, not that there's much difference between the two. However, they're very dispreferred in an embedded clause if not joining NPs. Also frequent is _naadin_ '(and) so' _im_ in particular often causes stripping of parallel material in the first of its two conjuncts. == Syntax == -- Basic clause order -- Basic clause order is verb-final. Of the verbal arguments, when both are given explicitly (which is not the most common situation), the more agent- or experiencer-like one goes first. Thus the effective word order agrees with SOV in a language with more typical case framing. _insiasuui innandaui thiuii_ 'a deer sees mushrooms' _insiasuui innandaui sjagawar_ 'a deer eats mushrooms' Placing an argument after the verb is a focussing technique. Subordinate clauses do not permit focussing by right dislocation, generally speaking [though perhaps I want to have some standard syntactical dodge to wrap important subordinate clauses to make it permissible]. Non-core arguments and adjuncts can never focus by postposition. [cf. my focus Conlang post of 6 Aug 2008, http://archives.conlang.info/thu/danzo/whoncontal.html .] There are a variety of types of short clauses, especially those providing setting, which can appear before the clause making the semantic thrust, or after, in the position of a main clause. When these appear initially the clause following them itself permits focussing, so perhaps they're not main clauses. -- Non-canonical clauses -- All interrogative clauses have a distinctive intonation. Yes-no questions are formed by adding _puukhau_, lit. 'it is not true', after the declarative clause, which doesn't change formally. This originated as a tag question but is now invariable (in particular it keeps its _kha_ even when the main verb is negative -- but are negative questions even used?). In short clauses the tag can be omitted where pragmatics makes the question status clear; the intonation remains interrogative. [I'm not sure if there's wh-extraction, or any of that jazz.] A typical cleft uses as main verb the empty root with a preverb, and precedes it by the original sentence nominalized on that argument. [example!] -- Basic order in other phrases -- Head-lastness is the general rule. This applies at least to subordinate verbs and complements of adpositions (which are thus postpositions). However, determiners follow their nominal head. Left branching to a fairly significant depth is commonplace, and an especially favored phrase structure. For words with two left dependents, there is usually a notable break in intonation between them, which I have been known to write as a comma. Even so, for such words, it's severely awkward for the right dependent to be substantially large, so that the left dependent is separated from the head by lots of material; a verb with two such arguments is better off focusing the right dependent. Given a bunch of verbs which one wants all to have coreferential arguments, the preferred tree structure is a (relativisey) chain, not a bush: [[[[V1] V2] V3] thing they refer to], not [[V1] [V2] [V3] thing they refer to]. (Admittedly these look mostly the same on the surface.) -- Subordination, and the noun/verb distinction -- There is no bright line between nouns and verbs in this language; subordinate verbs blur the line quite nicely, but probably have more in common with nouns than with finite verbs. Both subordinate and finite verbs appear in many syntactic dependent functions, generally predicating something of the head. Some examples: _thalau irua_ 'hot sun' _daunu hamthaaata hwasi_ 'pig with a knotted tail' The dependents of a verb (or noun) in a clause are typically noun phrases, most typically themselves clauses headed by a single noun or subordinate verb, (co)referring to some subset of the arguments of the head which are not nominalizers. It's also fairly common to see finite verbs in the modifier position. This contributes one of two senses: it can be interpreted as referring to the verbal event itself, or as a noun which is topical to the clause. If one wants to be clear that a subordinate verb or noun is modified by another word referring to an argument which is a nominalizer or a speech act participant or a topic marker (i.e. something other than a 3rd person agreement marker such as normally expects an explicit argument), this modifier can be followed by _mu_ (diachronically related to _im_ 'and'). I'm not sure whether it's an enclitic; in this respect it's parallel to the postpositions. Using _mu_ is in relative terms dispreferred as a coreference strategy to nesting clauses. [example where the two alternatives can be clearly contrasted] Subordinate verbs are usually arguments of the clause they modify. When the head of a subordinate verb is a noun, this yields an appositive sense (since the noun can often be thought of as derived from a verb ~'be N'). This example no longer has the relevance it did: _dimthan njuiphaa siaidua khai thiaas_ 'the fisherman doesn't realise how sick he is' -- Reported speech -- There's no morphosyntactical differentiation between direct and indirect speech [I think]. The first person can be used logophorically in reported speech, in situations characterised by enough empathy. Reported speech can be introduced by a form of _sit-pa_ 'say'. It is preferred to serially chain this to another verb expressing manner of speech (e.g. _lia-ur_ 'shout') than to give that other verb the speech as argument. == Miscellany == -- Numbers -- Underived numbers are verbs 'be (a group of) n'; they take absolutive arguments. The number system uses as its primary units 1, 10, and 120. The small numbers are _dan-as_ '1' _ni-ui_ '2' _sin-a_ '3' _gam-a_ '4' _thau-a_ '5' _labi-a_ '6' _lida-a_ '7' _iri-a_ '8' _iih-a_ '9' _khjup-a_ '10' Fractions are formed with the directional _suu_ 'apart': _gasuu-a_ '1/4'. Half has its own verb root _khwat-as_ with the transitive sense 'take half of'. Numbers are nominalised with an _a_ as nominaliser. Those with _-a_ as second component retain it, e.g. _sin-a-a_; 120 is _phian-a-s_. In some fixed expressions, e.g. numerators of complex fractions, but rarely otherwise one sees the plural nominaliser _i_ instead. Round numbers greater than 20 are formed as fractions of 120 (_phian-tu_). The fraction construction has a highly formal variant with the word _phianas_ '120' present, but standard speech omits it. It's rarely ever unclear when large integers are meant and when fractions because of number-agreement on the head (at least in verbal constructions), but to make things even clearer real fractions often use _khisuua_ 'part': _gasuuaa dimin_ '30 fish' _gasuuaa diman khisuua_ 'a quarter of a fish' Formed this way are _thausuu-a_ '24' (or _phianas thausuu-a_, and similarly for the rest) _gasuu-a_ '30' _sinsuu-a_ '40' _khwat-a_ '60' There's also _lau-a_ '20' which could very well be a reduction of _labisuu-a_ '1/6'. The numbers 60, 40, 30, 20, 10 are used additively (and greedily) to form numbers between 10 and 120. There is no subtractivity, so 119 is formed as 60 + 40 + 10 + 9. _thausuu-a_ '24' is generally not used in additive combination. For numbers used verbally, as they usually are, additions are formed simply as serial verbs, larger first: _khjuswis niwisui_ 'there are 12 of them' For numbers used nominally, the summands are strung together with _im_. Role nominalizations of numbers with _innu_ carry the sense 'be number N' (as an identification number), of which an especially frequent specialization is 'be Nth'. This extends even to _dan-innu_ 'be first', _ni-innu_ 'be second'. (Compare _dan-innu_ with _pu(l|wi)-innu_ 'be the one and only'.) 'N times' constructions use a number verb with a verbal argument. -- Systematic matters of word choice -- The language is path-prominent, as opposed to manner-prominent. Path is specified very naturally with directionals, commonly with the empty root. == Notes on diachrony == The two parts of the verb were clearly separate at some point. The first component looks like the original main verb, taking mostly suffixes. The second component in the earliest and most prototypical cases was a former object which became genericised and mandatory (such as _dim-na_ < distantly _n@dim- nad_ 'fish for a fish'). (Some explanation for the prevalence of vowel-initial ones seems necessary, but enough layers of analogy and grammaticalisation mixed with sound change should do the trick.) Later, after all verbs started requiring one many were given less than prototypical complements by this standard. Indeed, other parts of the grammar got sucked into the slot of the second component, as the former copula of class membership which yielded _-innu_ (and which, whatever it originally was, was not itself a verb.) Surely following some precedent, complements began spreading from words to semantically similar ones; this says something about the classificatory function (_iakhi-ram_ 'be sharp' + 'taste'?). The fact that the second component vanishes in serial constructions, and appears elsewhere in perfects, suggests that its value before it joined the verbal complex was something like an indicative. [How then did the clippings in nominalisation come about? This must have been a significant case of analogy.] Some TAM distinctions of the original verb were likely lost. The perfect marking with initial particles is likely old, and it may be the original marking of the perfect category (if there were other TAM marks, they may have occurred with both perfect and non-perfect, creating a two-dimensional system). Were perfect particles separate words? The truncation of second components in nominalization is peculiar and probably ravaged by analogy and other such developments; I highly doubt it's supportable as a regular sound change outcome. Some nominal/verbal pairs are from different originals (~suppletive); an example is _da-am_ 'show', noml. _da--h_. The phonation of initial stops among the more contentful verbal affixes is interesting. As a general rule, directionals, which are suffixal, exhibit voiced stops; verb prefixes, which of course are prefixal, exhibit voiceless (plain or aspirated) stops. Perfect markers, which are prefixal, exhibit all three series, but those with voiced stop initials overwhelmingly are identical with directionals, and probably were directionals originally. This is understandable in light of the sound changes [though I have to be _very_ careful about the ordering wrt the extension of the domain of application over the second component of present continuous verbs.] We postulate two series of stops in an earlier stage of the language, voiced and voiceless. The general scheme of development, using the exemplars _t d_, was t > {th d d t th} / {#_ V_ N_ C_ s_} d > {t 0 d 0} / {#_ V_ N_ C_} with loss of the _s_ in *_st_ and analogously for _h_ < /x/, except that sometimes quirks of syllabification could keep _d_ around in *_Cd_ and lose the other bit instead (as in _ba(l|da)-uh_). Voicing was also prevented in a stressed syllable following one in which the voicing rule applied. I hope this whole pattern isn't too scattered, and wonder what else *_nd_ could become that would help. Of course, roots with initial voiced stops aren't at all uncommon. I imagine initial material has been lost (_dim-na_ < _n=dim nad_ or so?), but that's somehow less than satisfying. A change far enough back to perhaps not be visible in many synchronic alternations (but maybe the _m_ disappearance?) is _m n_ > _0 r_ intervocalically, with loss of _m_ sometimes producing some concomitant vowel coloring(?). There was also doubtless some remnant nasalisation at the time, but it's gone without a trace. I imagine this also to have been the first introduction of _r_ to the phonology. I imagines a marker of shape _s_ for [+human] being separable from the agreement markers that preceded it in an earlier stage. We suppose the animate markers were once nom sg +hum _kt_ < _ks_ already in the ancestor; in Sab this had an early instance of the reduction _kt_ > _ht_ > _th_, which then could reduce again in the main developments nom sg -hum _k_ nom pl +hum _Gs_ [if final G developed like this, current Sab should show losses of invariant coda _-j_ before certain Cs] nom pl -hum _G_ acc sg +hum _is_ acc sg -hum _i_ acc pl +hum _wis_: of course _w_ is a segmentable element here too, and suggests a restructuring. acc pl -hum _wi_ It would then be sensible for this _-s_ to be the latest accretion. Perhaps human participants demanded a pronoun after the verb at some point. If it were, there shd have been a phase during which, even when menant for the nom, it showed up after the acc; analogy from nom intransitives should have fixed this.