5-1450msg1.txt

贝叶斯学习算法源码

i re-read newmeyer 's original post ( date 30 nov 94 ) and notice that it have an implicit attack on the notion of " basic word order " for comparative purpose . this flow from his earlier question on how basic word order be variously define by various analyst , imply that it lack precision . to the extent that i have understand the response so far , no one have respond to this particular point , so here go - - basically in agreement with newmeyer , but with some criticism of the assume larger implication for comparative reconstruction of syntax . i ' ll stick to what i know about linguistic event in europe since i expect the event to be more familiar to most reader than event in other language family , so that my example may speak to reader ' previous idea about particular linguistic event refer to below , and , under the best circumstance , provoke further discussion . to anticipate , the notion of basic word order be too simple-mind ( simplistic ? ) for diachronic purpose . fn : " could anything resemble latin syntax be reconstruct through comparison of the syntax of the modern romance language ? " bw : should it ? the immediate ancestor of the romance language be not latin but proto - romance . svo ? what about the exception , e . g . , french j ' aus rien fait etc . are n't they hint to certain historical complication . also aux inversion in standard french , e . g . , sont-il venus ? standard french aux inversion be probably of germanic origin , not find elsewhere in romance . as for reconstruction of case ( cf . latin / english ) , with possible implication for earlier word order possibility , case inflection remain on ( third person ) object clitic in all romance language . when we consider deviation from svo word order in romance language , we may suspect that the notion of " basic " word order be a villain to the extent that it invite the analyst to disregard less frequent or syntactically restrict word order . this be not safe in internal reconstruction of syntax , while the most reveal procedure may be to do internal reconstruction of syntax within a language before use the comparative method to reconstruct within group of language genetically relate by other criterion . in fact , the last point above be what be usually do , and be probably the only proper way to proceed , give the intent of the comparative method . always reconstruct a basic vocabulary first , on the basis of sound correspondence - - to justify genetic relationship among the language . then consider syntactic comparison . problem with this necessary procedure will emerge in follow discussion , but it remains necessary unless you want to reconstruct the syntactic evolution of a geographical area rather than of a genetic family . the latter may be a useful complement to assume genetic reconstruction , but i think it be too digressive an idea for me to pursue below . now , what about preverbal object clitic in romance ? another hint of complication to the svo concept of romance . preverbal object clitic means ov - - but i suppose " basic " means when o have a noun not a pronoun as the head . and what aout the fix of multiple object clitic order in romance ? e . g . , case order versus person ( inherent topicality ) order . that can't be reconstruct for latin because the clitic do not arise as distinct entity until proto - romance at the earliest . can a single or prefer clitic order be reconstruct for proto - romance , e . g . , dat-acc ( invariant in spanish and , i think , rumanian , but acc-dat seem to be older in french , now remain only for third person , i . e . , no inherent topicality difference therefore earlier case order remains - - french acc-dat may reflect germanic influence as well ? ) ? or do fix clitic order originally arise independently in various area of romance ? in any case , how can we avoid the comparative method in address the problem of the origin of object clitic order in romance ? in sum , i think that the origin / s and evolution of fix multiple object clitic order in romance be a legitimate issue , and that it cannot be solve without recourse to the comparative method ( among other , of course ) . therefore , the comparative method cannot be dismiss in syntactic reconstruction . fn : " should we therefore reconstruct proto - germanic ( almost surely incorrectly ) as svo ? " bw : newmeyer 's reference to proto - germanic reveal even more than romance that " basic " word order be a villainous concept for reconstruction in this case , as if " basic " word order can be compare across relate language to reconstruct " proto-basic " word order , such that " minor " word order can be ignore or reconstruct as " proto-minor " word order . one principle which emerge from our knowledge of the historical record be that word order change do n't work that way , i . e . , word order cannot be compartmentalize as " basic " and " minor " for diachronic purpose . principle : if we want to develop tool and principle for syntactic reconstruction , we will really have to consider the function serve by the various word order at different time , and detect change in their function diachronically . in other word , word order change in a language must be study in the context of the totality of function of all its word order at any give time . the difference between such a study and the study of change in " basic " word order , if the latter means anything at all , be analogous to the study of phonetic change and phonemic change . if analyst argue about the notion of " basic " word order , this be analogous to argue about different concept of the phoneme , and not about agree upon linguistic fact . the fallacy of reconstruct " basic " word order for proto - germanic on the basis of the " basic " word order in current germanic become obvious accord to the above principle because of the sov word order in subordinate clause in continental germanic . and even this be not as invariant a property of the oldest text as it be of the later standardise language , e . g . , before middle german and dutch . english also clue us in to inversion by its many remnant , even without historical text which reveal more similarity to continental germanic , e . g . , v - first follow a subordinate clause or adverb , still common in the av bible . include english but exclude gothic which mirror to the extent possible the new testament greek syntax from which it be translate , the earliest text in germanic show , as far as i know , a tendency relatively favorable to verb-final in subordinate clause , but much variation , later eliminate in the continental standard . this lead to argument about whether or not the sov tendency be the break-up of an earlier more general indo - european sov tendency ( or earlier fixation ? ) which die in english and scandinavian ( include the still highly inflected icelandic - - create problem even for simplistic functional argument for the evolution of svo in germanic , so that the argument would have to be : it started as a reaction to the increase unreliability of case distinction but spread for social reason to language that do n't need it . such an argument do not strike me as at all unreasonable ! ) . [ although i exclude gothic above , because of its malleable accommodation to greek syntax , its deviation from nt greek be quite reveal . most salient be the absence of a definite article , despite its occurrence in nt greek as well as in all other germanic language . if gothic translation be totally serious about imitate nt greek syntax in all case it could have adopt an unstress demonstrative to imitate the greek article , as later germanic do ( in a sense ) . the most important question about gothic syntax be : be the malleability of its syntax innovative from proto - germanic ? - - if so , we will probably never know in what way the gothic translation of nt greek stretch the limit of that malleability , and distort the colloquial function of word order in speak gothic or more generally in the germanic of the third century . if sov be invariant in some ancestor , then why do it deteriorate in germanic ? at least as puzzle , how / why / when do germanic get aux inversion , even in yes / no question . all indo - european language show wh front for wh question , so be there any reason to believe the proto - language do not . would i be buy what newmeyer be question with the precede argument , cf . fn : " i have the impression that with increase frequency , one come across statement such as the follow in the literature : " most of the attest language in language family x have some syntactic property . therefore we can assume that proto - x have this property . " " bw : [ why with " increase frequency " ? is the implication that copy-cat historical linguist be become less responsible than they use to be ( cf . the discussion of the spread of rumor about the number of eskimo word for snow ) , or that syntax be drive historical linguistics to pot ? ] to be safe , i guess ancillary argument come in , like : consider how areally widespread the ie language be , be it likely that the unanimity of wh first question across ie be not reflective of the proto-language ? ok , so maybe wh front be proto - ie ( get a better explanation for the fact ? or reason to ignore them ? ) , but why aux inversion in germanic question . and do the application of inversion to yes / no question mean that there be a question marker in initial position ( wh question position ) with yes / no question , as in " what / hullo , be you kid me ? " ( such a marker " ibaus " be find in gothic whenever the new testament greek original have " me : " ) . do n't get me wrong . i ' m not suggest that an initial yes / no question-marker be a necessary condition for the inversion innovation in germanic . i ' m only observe that there be the option of such a marker , and anticipate ( if indeed it have not already be suggest ) that some analyst who be overly formal in their approach to syntax may suppose the necessity of such a marker to motivate a mechanical generalisation of inversion from wh to yes / no question . again , because such in - version occur in all germanic should we not reconstruct it for proto - germanic . . qualification on all : it 's variable in gothic where it 's variable in nt greek : pronoun subject usually do not invert with the verb , noun subject usually do . ] a practical point be that , as far as i know , there be an issue about the differentiation of germanic into anything like the modern descendent before the 4 - 5th century . thus , if aux inversion go back that far , do it matter if it spread from one germanic area to another or if it be part of proto - ( west ? ) germanic ? all innovation must spread before we recognise them as innovation in the language ( or dialect ) . historical linguistics do not study the evolution of the idiolect ( if such a concept as " evolution of the idiolect " be even coherent ) . finally , about typological argument . when morphology-a - fossilize syntax and universal-typology of word order congeal in the early 1970 , some suggest that even before ie be sov it be vso . why ? because subject mark inflection follows the verb : v - s , get it ? inevitably , then , some suggest that celtic maintain the most archaic " basic " word order among the ie language . this be another excess in diachronic application of ( whatever ) the notion of basic word order . the principal objection raise be that it be not obvious ( and even unlikely ) that only basic word order morphologise . ( in fact , they may be least likely to morphologise . ) consider that s represent an unstress anaphor in v - s , s may be in a minor word order position . [ and note that if basic word order means when the argument be nominal , not pronominal , then discourse frequency be not criterial of basic , since at least most subject argument be pronominal , if not merely inflectional . furthermore , if " basic " only count the relative frequency of subject nominals , it be base on something which be quite rare in discourse : subject nominal ] . in any case , subject inflection follow the verb stem be a widespread eurasian areal feature , extend into africa in semitic and various other branch of afro - asiatic . to some this may be take as a clue to the correctness of nostratic and other super-family notion , to other an indication of an extremely old innovation have spread ( over millenium ? ) across family regardless of genetic relationship ( cf . the famous balkan area ) . still other may opt for coincidence . various other language also have this feature , e . g . , kanurus in west africa and various new world language . ( welcome to nostratic ? ) question about the function of position after the verb in so-cal sov language be raise by v - s inflectional order . the typological implication of this go beyond my current knowledge , although i dimly remember that various eurasian sov language be different accord to whether or what function post-verb - al position have in such language . this be obvious , for example , in the contrast between eurasian sov language ( with difference amongst themselve ) and the so-cal sovx west african language , where x be an adpositional phrase ( with complication about the origin of adpositional phrase in such language which i will refrain from discuss here ) . since so many eurasian sov language be also v - s inflectional language , i do not know if there be sov language which have a " minor " postverbal position for an anaphoric subject ( such that it turn up in other position in certain construction , so that we can demonstrate that it be not an inflection ) . maybe typology of relevant amerind language will help solve this problem . but maybe we may also anticipate the reappearance of the party-spoil question about the representativeness of curent language type to possible ( and former ) language type . no doubt an unwelcome ( and unhelpful ? ) think to many . in this context it seem worth mention that mathematical probability argument for various word order , as i have see them practice , be not impressive , because of fault in their initial assumption . for example , as far as i know , they invariably assume the validity of count number of language , regardless of the genetic relationship or areal contiguity of the language count . this seem unsound to me , particularly ignore areal contiguity . for example , we find ( i think ) that the eurasian area , consist of umpteen and umpr language , have s . . . o word order ( ignore the position of the verb as an independent variable ) . then in the new world we find large area of contiguous s . . . o order , and separate area of o . . . s order . maybe we should be calculate mathematical probability on the basis of contiguous area rather than number of language . the surprise result may be that o . . . s be more probable than s . . . o ( where area be an independent variable , not individual language , of course . ) i ' m not sure what the implication of this may be , but one may be that o . . . s may have once be a more likely order strategy than it be now , and that the predominance of s . . . o across language count individually be largely the result of areal spread . [ if the spread of s . . . o be old enough , its predominance may even be the result of an absolute increase in the number of " language " in the world all together . however , this be n't really relevant to my criticim of the way probability argument have be apply to syntactic typology , only to the historical implication of adequate assumption about language typology . in fact , i ' ll leave this discussion with the already widely accept suggestion that strategy for information distribution in the clause , esp . in term of old / new , be more relevant to typology than such problematic notion for cross-linguistic comparison as subject and object . similarly , for example , i think that in eurasium , agent . . . patient / theme order largely cut across ergative and accusative language , even though ergative language would have patient / theme , or whatever term you want , as subject . to this extent , s . . . o cannot be universal even in eurasium , without confuse " accusative " definition of subject and object , where subject status be consistent with case-mark , and " ergative " definition , where case-mark be more closely tie to transitivity role than to the mushy cross-linguistic concept of subject . ] fn : " and furthermore , syntactic change can be fairly catyclysmic , restructure grammar wholesale in one generation - - unlikely or impossible with phonological system . " bw : i think the thrust of fn 's suggestion here be largely right , though probably vastly overstate ( on the basis of older idea about creole ? ) . however , question remain in my mind about condition under which these thing happen . within monolingual area ( or among closely relate , mutually intelligible language , ( certain type of ? ) syntactic strategy seem to have the potential to spead quite quickly , perhap almost as quickly as new word and expression , while comparable speed be not generally observe for phonological change ( and be probably " unlikely or impossible " , as newmeyer put it , for both internal linguistic and social reason - - in fact , i think " unlikely " be more accurate than " impossible " depend on the nature of the particular type of phonological change , and i have an example of possible single-generation phonological change in the current english of the african american in new york city area , but i do not have sufficient datum to make this possibility or its implication worth discuss here - - until i get to the section in labov 's new book where he discuss recent innovation in the chicago vowel system i will not be sure if he have other example ) . however , to the extent that language-contact propel " cataclysmic " syntactic change ( in the historical record ) fn 's suggestion be not so clear . for example , english in east la ( a mexican american community ) be more immediately strikingingly different from adjacent english dialect on the phonological than on the syntactic level . and that apply to monolingual speaker of east la english as well as spanish - english bilingual , and to various phonological segment as well as intonational contour ] . in understand historical syntactic evolution , it be most often difficult to distinguish internal evolution from language contact ( hence the sour regard for substratal theory during the late 19th - to-mid 20th c celebration of neogrammarian achievement ) - - and then there be also jakobson 's caution that language contact may not permit evoluton that be not internally possible anyway , rather that it can only promote one possible direction of change as oppose to other possible direction which may flourish under other external circumstance . this too i consider problematic as a blanket statement , but a major consideration to keep in mind as a possible constraint on change in most reconstructive expedition into the unknown past . enough . these be my thought about the problem of use the comparative method alone to reconstruct syntax . but nobody would suggest something so foolish as not enlist all possible tool of reconstruction , include internal and typological in addition to comparative . conversely , it would be equally foolish to shun the comparative method as a reconstructive tool . in any case , the notion of basic word order be not helpful to syntactic reconstruction , if not of doubtful value as an autonomous observation about the synchronic state of any language as well . i mean it 's ok to observe that english and french be both synchronically svo ( speak french even more than english - - not least because of the former 's prefer question formation strategy ) , but without further discussion that do not mean that their syntax be " basically " the same , in any insightful or interest ( to coin an adjective ) sense . i would like reader to react to any of the point i have make above , not least of all the accuracy of the fact i have suggest for various language , since i be not an expert in the area of most of the fact i have present , e . g . , current typological theory , indo - european linguistics , eurasian and new world areal syntactic characteristic . benjus