Synopsis of Hopper & Traugott Chap. 5

Handout for LING 319/519, SARS 319/519

Grammaticalization

Get print version of this file here.

  1. Unidirectionality This chapter looks systematically at the issue of Unidirectionality of grammaticalization, i.e. the hypothesis that once grammaticalization begins, it goes forward inexorably toward final morphologization even though there may not necessarily be an end stage. That is, grammaticalization goes forward and does not go back; but it may not necessarily reach the complete and final state we recognize as (e.g.) morphologization of a lexical item (e.g.) into a morpheme. But retreat is not possible.

    General Principles of Unidirectionality.

    • Diachronic issues such as generalization

    • decategorialization

    • increase in grammatical status

    • renewal

    • as well as synchronic issues such as

      • Variability resulting from grammaticalization

      • "layering" arising from grammaticalization

    • Counterexamples to unidirectionality?

  2. Paths of grammaticalization, and Meillet's suggestions:

    Example of Greek future tha which Meillet says is

    lexical item --> morphology

    In fact (as H&T point out repeatedly) this is not direct: instead, it is more like "lexical item in certain highly constrained local contexts is reanalyzed as having morphological or syntactic functions, or

    lexical item in specific contexts--> syntax --> morphology

    A:
    item becomes syntactically fixed,
    a "construction"
    Example:Lit. Tamil inru meel 'today+top'
    B:
    eventually fuses morphologically
    --> inimee 'from now on'
    Assumption: A precedes B, not vice versa  

    Nothing deterministic about this:

    • Changes do not have to happen
    • Changes don't have to go to completion
    • Changes don't have to move all along the cline
    • Changes are often interrupted, with a messy incomplete "system" not moving in any direction (such as the quasi-modals 'need', 'dare' in English.)

    Some people would like to claim that once something starts, its end is implied or can be expected, and progress is inexorable or inevitable. No evidence for this; there are strong constraints on how change occurs; and on directionality. Compare Sapir's notions of drift which is different, perhaps a more abstract idea about change, or "a metacondition on the way in which the grammar of a language as a whole will change. (R. Lakoff 1972).

    Generalization

    On generalization, note some differences: some say generalization leads to weakening of semantic content, with phonetic erosion. But H&T don't agree; instead see

    loss of older, ... more concrete meanings and development of newer more abstract ones that [...] cancel out the loss.

    Refer to previous discussion of this (more informativeness, pragmatic strengthening.) generalization means

    1. increase of polysemies and/or
    2. increase of range of meanings of a morpheme (Kurylowicz).
    3. Also may mean more syntactic freedom (such as not restricted to verbs of motion, or transitive verbs, etc. e.g. French pas )

    Generalization of Meaning

    What constraints on kinds of meanings that are subject go grammaticalization?

    Lexical fields: there are semantic relations that occur in lexical fields , e.g.

    • color terms, verbs of saying, psychological verbs, verbs of motion...

    • relational items (e.g. kinship)

    • taxonomies (animals, body parts, ships, land forms, folk taxonomies of plants, e.g. medicinal, etc. )

    • complementary things (non-gradable pairs w. excludable middle)

    • antonyms

    • directional oppositions (go-come)

    • synonyms

    • polysemies ...

    Usually lexical items that are chosen are general, superordinate and not specialized, so go, come might be used, but not "ooze, flutter". (But see the Tamil aspectual verbs, which start out with some general ones, but gradually involve more specialized ones.) What are chosen are basic words of a field, a.k.a. hyponyms Or, a form is less basic, but is then generalized, and then grammaticalized:

    • Latin. ambulare --> aller (French)

    • --> AUX Nous allons voir ça; on va voir ça;!

    • "We're going to see that, i.e. we'll see about that!"

    General lexical items then take on more and more generalized functions, as used in more and more contexts: wider distribution, more polysemies. Former inferences are grammaticalized.

    Difference with lexicalization changes, where some meanings become narrowed, e.g. cf. German sterben 'die' vs. Eng. starve 'die of hunger.' Dispreferred items (taboo items etc.) become narrower and narrower.

    Grammaticalization does not give evidence of this narrowing, so this absence predicts constraints on possible developments (see examples of negative phrases and negative incorporation, pg. 98, where extant negative form is negative of the weaker member; can be strengthened by an implicature cancelable upward to stronger meaning.)

    Another constraint: avoidance of homonymic clash, so if two lexical items 'fall together' (because of some phonological development) one or another will be replaced, esp. if meanings are opposite. Most homonymic clash is lexical not grammatical, so multiple grammatical homonymies can exist, and without problems:

    Cf. English grammatical morpheme -th Can mean:

    1. State of being s.t., e.g. dead/death, broad/breadth, wide/width, long/length, true/truth. etc.

    2. 'Ordinal' number: thirteenth, fifty-fourth

    3. Other possibilities?

    The two meanings can be homonymic and not clash, because they occur in very different contexts. Also true of the English -s which functions as:

    • Plural marker for most nouns: cats

    • third-singular of most verbs: knows, reads

    • Possessive marker of most nouns: John's

    No problem with homonomy: John' s cats sleep a lot, while his dog chases cars.

  3. Generalization of grammatical function As grammatical forms develop, they will develop more and larger range of meaningful morphosyntactic purposes:

      Eg. English be V-ing

    • first: agentic constructions in the passive ("is being brought into heaven")

    • then stative contexts: know-ing, lik-ing...

    Cf. Also Finnish example of reanlysis (singular agentive N's) then generalized to new environments via:

    • singular NP's and pronouns

    • plural pronouns and plural agentive NP's

    • plural non-agentive NP's

    Spread along a hierarchy of NP's that are more subject-like to less subject-like NP's? (Timberlake) This hierarchy would probably be unlikely to be reversed (H&T).

  4. Decategorialization is a process whereby something that is clearly marked (either by morphology or by function) as a member of one grammatical category (e.g. N, V, Adj) shifts to be more marked or functioning as a member of another category.

    In lgs. with lots of morphology, N's can be determined to be N's by their morphology and the affixes etc. they can take. Same with V's. In a language without much overt morphology, what is a N or V is known by the function they perform, so when an item in English like prepositional "off" is used as a verb ( Off the pigs! a Black Panther slogan of the 60's and 70's) we know it's a verb from its function and context. Often this is along a cline from major to minor, e.g.

    major category (--> adjective/adverb) --> minor category.

    H&T give examples of shift from major to minor as more important(?) than my example of shift from minor to major; e.g. development of Eng. while as a conjunction instead of as a N. When used as a conjunction, while cannot:

    • take articles or quantifiers

    • be modified by adj or demonstrative.

    • serve as a subject or other argument of Vb

    • only appear clause-initially

    • subsequently be referred to by anaphoric pronoun

    These are all negative; positively, while has gained ability to link clauses and indicate temporal relationships (and also act as a 'concessive' marker) in discourse, which it couldn't do as ordinary N. So this is not decay or deterioration but functional shift from one role to another. The trappings of noun morphology are thus discarded, since not needed.

    Compare development of Tamil/Kannada postpositions from verbs, such as paattu from the verb paar meaning 'see.'

    • paattu (past ppl. of paar meaning 'having seen') loses all morphological complexity; reduced to one form.

    • expands range from those where actual vision takes place to other verbs of cognition, e.g. kuruDan avare paattu siriccaan 'The blindman laughed at him.'

    • Expansion to function of attention-direction e.g. naayi enne paattu koleccadu 'The dog barked at me'

    • Expansion to use as attribution of 'negative-intent' as in

      iNNekki varakkuuDaadu NNu sonneen-ee aanaa iNNekki paattu vandiingaLee!
      today come-prohib unquote said-1png-emph but today deliberately came-2png-emph
      I told you not to come today, but you deliberately came today anyway!
      (Or: 'I told you not to come today, but you went and did so, anyway!)

  5. Development of Nominal Categories and of Verbal Categories: two typical paths. These are clines:

    • A cline of grammaticality, making reference to hierarchical categories relevant to constituent structure.

    • clines of decategorialization i.e. starts with a full category (N, V) with then loss of morphological trappings of the category.

    But do we have unidirectionality along the cline? Are the clines continua, or rather clusters at points with grammatical properties resembling ("family resemblance") certain other kinds of things:

    • Auxiliaries
    • prepositions
    • articles
    • something seen in some other languages.

    Clusters are not rigid, fixed points, but gathering places, like iron filings around magnets.

  6. Period of Overlap between older and newer functions, so not a situation with everything lined up like cars of a train. Some people. use term 'chaining' to show non-linearity; H&T prefer 'layering' allowing for multiple origins of a grammatical form, i.e.

    	____________
    	___________________
    			_________________________
    					_________________
    					_____________________	
    						_________________________
    								_________________
    									Etc. 
    
    

    Remember that once embarked upon, grammaticalization does not have an inevitable result; may be arrested etc. We can't work back from any one form to see a clear path; and we can't illustrate a particular cline with only one form. (Not enough historical record, they say; I would bet we might find examples in Tamil with its long time depth, hs). What is important is the issue of directionality between adjacent forms on the cline not demo. of complete sequence of events for a given form.

  7. What are some typical clines? Lehmann 1985 has given following:

    • Relational Noun -->

    • Secondary Adposition --> (kind of prep/postposition)

    • Primary Adposition --> (kind of prep/postposition)

    • Agglutinative Case affix -->

    • Fusional case affix.

      Examples:

      • Relational Noun: nouns such as 'top, side, way' or Tamil pakkam 'side'

      • Secondary Adposition: short phrases such as 'in/on the way', 'beside', 'ahead of'

      • Primary Adposition: reduced, maybe monosyllabic, indicating purely grammatical relations (instead of locational) such as 'by, of, with, to.' Some of these may be themselves locational and relational, e.g. 'by'. Can be easily cliticized.

      • Agglutinative Case affix: Hungarian examples; Tamil (see below)

      • Fusional case affix. Latin example -ibus , Russian -yx, -ov, -amy etc.

      Again, not strictly discrete, but arbitrary clustering points on a cline. Most forms we find will not fit clearly into one or another of named categories, but will land somewhere on the cline.

    • Example of the relational Noun: development of Tamil case and postpositions.

      • Relational Noun: pakkam 'side' as in maDraas pakkam 'Madras-side' (the Madras area, the Madras region)

      • Secondary Adposition: pakkattule which is pakkam plus loc. ule e.g. maDraasukku pakkattule 'near Madras' (Madras-dative pakkam-loc.)

      • Primary Adposition: eduttaaple 'opposite' from the verb edir 'opposite' + taan 'itself' + poola 'like.' with phonological reduction of edir-taan-poola including. deletion of long vowels in poola which otherwise doesn't happen.

      • Agglutinative Case affix --> 'Ablative' marker -lerundu, -lendu which is the Literary Tamil loc. -il plus irundu past participle of iru 'having been' e.g. maDraas-lerundu 'from Madras.

      • Fusional case affix: Dravidian doesn't get fusion in the way that IE lgs. do, but maybe the accusative -e which can't be analyzed as having any source, i.e. whatever its source, it's now obliterated and shortened (from Literary Tamil ai ) to -e : avan-e paatteen 'him saw-I'.

    • Verb to Affix Cline:

      • Full Verb -->

      • (Vector Verb -->)

      • Auxiliary -->

      • Clitic -->

      • Affix

      That is:

      • Full Verb: verbs with full lexical meaning, all grammatical characteristics, full morphology etc. like other verbs.

      • Vector Verb: may be a special category? See Hindi, Kannada, Tamil (below)

      • Auxiliary Verbs are finite, have tense/mood etc., but syntactically specialized (e.g. English Aux inverted in Q's); maybe can't be marked for tense, or (in Dravidian) tend to not be marked for PNG. Many examples in many lgs. Can have examples of both full and aux status, e.g. English. have as full vb, aux (have gone etc.) and new development of hafta .

      • Cliticization: have becomes 've e.g. we've gone and done it again or the French cliticization of forms of avoir added to future, e.g. j'aurai, ils sauront 'I'll have, they'll know .

      • Affix

    • Peter Hook's examples from Indo-Aryan languages. Hook gives the examples of Indo-Aryan (IA) verbs that he calls 'vector verbs' that start out as main verbs like lenaa, denaa and are used in Hindi and other languages for aspectual and other purposes. Vector verbs are supposedly intermediate between full/main verbs and auxiliaries, since they retain lexical identities as well. (Compare English 'keep' with full lexical identity but also "quasi-partial auxiliary" status as in 'keep (on) Vb-ing'. Also 'use' in constructions like 'used to' Vb.)

      The Hindi data are paralleled by data from other IA lgs. which don't have the same frequency, declining to zero in some Dardic lgs. (Shina, Kashmiri) which some people say aren't really IA lgs. anyway. Hooks' account shows how vector verbs contribute notions of perfective aspect, emphasis on completion/completedness, full affectedness of the verb's object, and involvement of an agent. Hook

      • Hook looks both at earlier texts, with fewer examples of this (lower frequency) and
      • other IA languages, with lower frequencies of use, but
      • in some cases, larger number of these verbs (Marathi has more than Hindi).
      • Hindi verbs that can be used: go, give, take, throw, strike, let go, get up, come, sit, fall and some others.
      • In Marathi etc. these verbs only used with verbs inherently non-specified for completedness , i.e. they add info.
      • In Hindi, can be redundant (spread to more environments, e.g. verbs of communication).
      • HS: in Tamil etc., this also true of completive viDu , but if used with sollu 'say' etc. it makes it more definite , e.g.

        iNNekki varakkuuDaadu NNu solliTTeen-ee aanaa iNNekki paattu vandiingaLee!
        today come-prohib unquote saidCOMPL1PNG+emph but today deliberately came-2png-emph

        I definitely said NOT to come today, but you came today anyway!

      In other words, Tamil COMPL ( (v)iDu ) is probably polysemous, having meanings of both COMPLETIVE, DEFINITE/DEFINITIVE, and maybe some others. Hindi verbs also have this polysemy.

      According. to H&T, Hindi can't be said to have any kind of AUX (or vector) category of Verb that is different from Marathi; or perhaps we can't say Hindi vector verbs are a kind of AUX, but Marathi's are not?; there is a scale of grammaticalization here, obviously, and it's messy. Perhaps the Marathi ones aren't as fully grammaticalized as those of Hindi. But all Indian lgs. (as noted) have these to some extent.

    • Schiffman paper on the Tamil aspectual system .

      1. The Hindi/Marathi contrast shows that simply statistically alone, Hindi has more 'compound verbs' (main vb + vector vb) than does Marathi, or any other Indo-Aryan lg. According. to H&T, statistics here mean something i.e. "prima-facie evidence of degree of grammaticalization." (p. 110).

      2. Differences in kinds of main verbs: in Marathi, can only occur if main verb is unspecified for completeness; in Hindi, can be redundant and occur with 'completed' verbs, so maybe means something else?

      3. Also in Hindi, the vector vb analogous to go is the commonest one, while in Marathi, many more verbs can be used. Parallels with the "secondary adposition" situation, with more verbs, more semantic concreteness, less generality.

    • Single clines vs. multiple clines In this section, they discuss how some verbs (or whatever) are recruited more than once, i.e. they have multiple paths, i.e. we can get:
           		      _________> A	
      	        	     /	
      	 source _____/
      	  	        \
      	             \__________> B
      
      
      or conversely, we can also get merger:
       		
      
      	       A______
      		         \
      		          \_________ new item
                         /
      	      B_______/ 
      
      

      Merger occurs when sources from slightly different domains converge on one grammatical. domain, if there is pragmatic, semantic and synactic appropriateness. Metaphor: convergence in 'semantic space.' Examples given are convergence of REFLEXIVES and MIDDLE VOICE, common cross-linguistically. Also EVIDENTIALS and CONDITIONALS.

  8. Processes participating in unidirectionality

    • Specialization

      Example of Hindi verbs where go vector vb is becoming more common; speculation is that the others will become less. Some verbs are gaining ascendency and becoming more general, and therefore more broadly functional. Specialization doesn't necessarily mean no coexistence, but some things must themselves become specialized, or lose out. May depend on

      • which semantic types involved

      • sociolinguistic contexts

      • discourse genres

      • etc.

      Another good example: modern French specialization of negative particle ne plus nouns, now usu. pas 'pace, step' Other nouns were in used, some have remained and become specialized ( point, personne, que, jamais, plus, etc.) but some have gone out ( mie, gote, amend, areste, beloce, eschalope ). Now point denotes only "emphatic negation contradicting previous assertion" ('No, it's not true, there isn't any at all. ) and pas is now the negative marker period. Remember that the choice of these items and the grammaticalization of them is motivated by discourse constraints, i.e. originally pas would have been restricted to verbs of motion, but now generalized.

    • Divergence

      • Example of English a/an and one . Originally, one form, Old English an [a:n], which normally would have become [o:n] but instead on the one hand we get [w@n] 'one' and on the other, the cliticized forms a/an i.e. unstressed.

      • Malay classifiers. (For background, see similar data from Indonesian. )

        The Malay and Indonesian data show how the grammaticalization of numeral classifiers recruit lexical items with particular shapes (at least with inanimate objects) and then use them as classifiers. Because they retain some resemblance (in most cases) to the original lexical item, persistence is illustrated: if the noun being classified is orang "person" you don't get a classifier with it, i.e. you don't get *seorang orang But in the case of *sabatu --> sawatu --> satu you can get satu batu . If the classifier for a term is used, then you don't get s(u)atu which has now taken on the meaning "one", so you can get either satu rumah 'a house' or sebuah rumah 'one-bunch house'. So we now have divergence of the forms satu and batu which is part of the origin of satu (*suwatu <-- subatu) .

        phonological divergence is not necessary however: we can get in Dravidian forms like Tamil viTT-iDu 'leave-DEF' "get out of here!" and the same for Kannada: biTT-biDu "ibid." where one form is the lexical item, followed by the aspect marker based on/derived from it.

    • Renewal: existing meanings take on new forms.

      Most interesting point they make here is the quote from Schwegler (1988) who says that there is a "psychological proclivity" for the development of negative emphasizers, and that they have their point of origin in contexts of contradiction i.e. emotionally loaded contexts. So the French system is a case in point; the English development reuses NEG in various ways, i.e. not , no way (Jose) .

      Renewal by a non-cognate item to effect semantic expressiveness underlies most examples of development of innovative periphrasis which Langacker calls the major way to "achieve perceptual optimality in syntax."

        What is periphrasis?

      • Meaning is not discernible from constituent elements
      • construction shows syntactic unity at some level
      • new construction competes paradigmatically with older morphology.

      After renewal, new form may undergo reduction, e.g. not --> n't . This gets us into recursivity and arguments about "reduction to zero" and renewal etc. H&T say no; how can there be a stage where you can't say something you need to? Rather, the new form says it better or more expressively, or more communicatively, etc.

  9. Synchronic Result of Unidirectionality Persistence of older forms and constructions paralleling new ones can be called layering (or variability (!) As new forms emerge, older ones remain, half-buried, i.e. layered (think archeology!).

    See examples (p. 124) for different ways in English to express tense and aspect; assume that the most bonded forms are the oldest, the more periphrastic and less reduced forms are newer.

    Syntactic examples from Estonian (could also be German, Russian) with two kinds of relative clauses:

    • One clause embedded before the Noun, i.e. German Ich sehe den [in-Berlin-wohnenden] Mann "I see the in-Berlin-living man" vs.

    • Relative. clauses follows NP, with a relative ppl: Ich sehe den Mann, der in Berlin wohnt "I see the man who lives in Berlin.

    In Estonian the pre-NP embedded type is 'older, more learned' but more modern language uses a Np+rel.ppl type. Both coexist.

    Presumably these 2 types have a pragmatic difference; others may reflect older historical layers of (e.g.) Object Verb syntax instead of newer Verb Object syntax.

    More counterexamples, from lexicalization? H&T reject. But grammaticalization as a unidirectional process, and lexicalization as non-unidirectional, may intersect.

    Another problem: the non-intersection of Output2 and Output1 and the attempts of speakers2 to repair or cover-up their "mistakes" when their output doesn't match. Known as hypercorrection. May be typical of adults rather than children? May obscure or confuse attempts to detect abductive changes (reanalysis).


haroldfs@ccat.sas.upenn.edu
last modified 1/27/05