3-416msg1.txt

运用贝叶斯网编的一个识别垃圾邮件的程序

Subject: re : 3 . 395 rule

g . vanden wyngaerd write : > 3 . 387 marttus arnold nyman < manyman @ finuha . bitnet > write : > ( 1 ) a is [ the man who be tall ] _ _ in the room ? > b is [ the man who _ _ tall ] be in the room ? > > why would a more or less random distribution over ( 1a ) and ( 1b ) be > expectable in the acquisition stage , if speaker unfailingly - - and so , > as the only pattern for child to base inductive generalization on - - > produce ( 1a ) ? > the assumption implicit in nyman 's question be that child only > produce what they hear . this be plainly incorrect . children do not hear > form like " buy " , " eat " , or " goe " , yet they all go through a stage > where they produce these form . this can only be because they make > generalisation ( rule , if you like ) , which go beyond what they hear . now > give that the main source of evidence on yes-no question at the child 's > disposal will overwhelmingly consist of simple sentence of the form " is > the man _ _ in the room " , the child could make the generalisation either > in way : in term of linear precedence ( " front the first finite verb " , > yield ( 1b ) ) or in term of hierarchical structure ( " front the finite verb > which follow the subject " , yield ( 1a ) ) . the fact that child do not > make mistake in this respect ( ie do not form ( 1b ) ) clearly show that > the rule be not one learn by experience , the relevant experience not > be rich enough to determine the nature of the rule and not be > able to explain the absence of mistake . there be a number of interest thing about wyngaerd 's statement here . 1 ) form like " buy " , " goe " , and " eat " may occur in child speech , but they be in a minority . the vast majority of attempt at irregular verb be produce correctly . ( gary marcus and his colleague at mit have a monograph on this that be not yet out . ) children do tend to produce what they here , at least statistically . 2 ) the assumption be make , in the absence of any data , that child rarely hear adult produce sentence like ( 1a ) above . this be an amaze claim , and i doubt that it be true . 3 ) even if it be true , wyngaerd be make generalization about learn in the absence of a theory of learning . these last two thing be unfortunately very common . why do we think that it 's ok to say , " there 's no datum on this , but if there be , i ' m sure it would be x " , and expect people to take it seriously ? or " i have no theory of learn , but i ' m sure that it would n't predict x " ? i ' m a phonologist , and i have n't keep up with change in syntactic theory , and i ' m sure that i would n't be allow to get away with statement like " i have no reasonable formal theory of syntax , but if i do , i ' m sure that it could n't accommodate subj-aux inversion , so all theory of formal syntax must be wrong " . but this statement be no different in kind from the other one . one last statement implicit in much work in linguistics : " i have no theory of genetics , ontogeny , or evolutionary biology , but i ' m sure that if i do , modern linguistic assumption about innateness would fit in real well . " maybe we should ask a bite more of ourselve that we often do . - - - joe stemberger