• No results found

A self extending lexicon: description of a word learning program

N/A
N/A
Protected

Academic year: 2020

Share "A self extending lexicon: description of a word learning program"

Copied!
14
0
0

Loading.... (view fulltext now)

Full text

(1)

Eva Ejerhad and Hank Bromlay

A SELF-EX1ENDING LEXICON: DESCRIPTION OF A WORD LEARNING

PROGRAM

1. Introduction.

This paper d e s c r i b e s the curr en t i m pl e m en t a ti o n of the lexical an aly zer (MORPH)» of a fi ni t e st ate parser for S wed is h (S U E D I S H - S Y N T A X ). The M O RP H p r o g r a m was d e v e l o p e d by the au th o r s in Augus t 1985» and it is i m pl e men te d in Z LIS P on an LMI CADR Lisp machine. M O RP H is a c o m p u t at i o na l model of the r e c o g n i t i o n and a n a l y s is of w r i t t e n words» with the f o l l o w in g brief c h ar a c te r i s t i c s :

M O RP H is a word an aly zer only» it does not s y n t h e s i z e or g e n e r a t e word forms.

- M O RP H wo rk s left to right through a word» withou t p r e ­ p r o c e s s i n g it in the form of suffix or prefi x stripping.

- M O RP H m o de l s both the p r oc e s s of word r e c o g n i t i o n /ana 1ysis and the p r o c e s s of word a c qu i sit io n.

- M O R P H w o rks by c o m b i n in g the a d v a n t a g e s of the two general m e th o d s for word r e c o g n i t i o n that are availa ble : m a tc h i ng input w ord s with words, and p a rs i n g input m o r p h e m e s into words. By c o mb i n i n g these me thods» the d i s a d v a n t a g e s of using either m e tho d al on e are avoided.

2. Considarationm in daaigning MORPH.

(2)

The v e rs i o n of SUIEDISH-SYNT AX that I d e s c r i b e d in 1983 had a full form lexicon which listed each inflected form' of a word along with its part of speech and m o r p h o - s y n t a c t i c features? e.g .

(defslex flicka noun -*-utr -pi -def) (defslex fl ic k a n noun +utr -pi + d e f ) (defslex flicker noun -*-utr -*-p 1 -def) (defslex f l ic k o rn a noun -♦-utr -*-p 1 + d e f )

The parser uses the part of speech in for m at i o n of a word in d e ci d i ng c o n s t i t u t e n t s t r u c t u r e for an input string, and it uses fe at u r e i n f o r m at i o n w hen c h ec k i ng that ag re e m en t c o n ­ s t ra i n ts are satisf ied . A g re e m en t in gender, number and d e f i n i t e n e s s b e t w e e n a d ja c e nt w o rd s in a noun p h r a s e was and still is ha nd l e d in S UED ISH-SY NTA X by a m e c h a n i s m that filt er s out any c o m b i n a t i o n s of n o n - a g r e e i n g w o r d s , l i k e

* ett /acker - u t r - « - u t r -pi -pi -def -def

f 1 i c kan + u tr -pi -fr-def

This e n su r e s that all noun p h r a s e s that are well form ed with respect to c o n s t i t u e n t s t r u c t u r e c o n s t r a i n t s (e.g. NP — DET A D JE C T I V E NOUN), are also well fo rme d with resp ec t to a g re em ent co ns t r ai n t s. The way this wo rk s in S U E D I S H - S Y N T A X is not te ch n i ca l l y by u n if i c at i o n, but by imposing the r e qu i rem en t that for im me d i at e c o n s t i t u e n t s of a noun p h ra s e in Swed is h no c o n f l i c t s of va lu e s are al lo w e d for the f e at u r es of gender, number and d e f i n i te n e ss . In the e x am p l e above, there is both a gender co nf l i ct and a d e f i n i t e n e s s co nf l i ct For a atore d e ta i l ed te chnical d e s c r i p t i o n of the general p r op e r t i e s of the a g re e m en t me cha n is m , see C h urc h 1903, and for its a p p l i c a t i o n to Sw ed i s h syntax, see Ej er h e d 1903.

We wa nte d to i n cre ase by a large amoun t the number of wo rds that could be s u c c e s s f u l l y dealt with by S W E D I S H - S Y N T A X , b e ca u s e a very large, and p r ef e r a b l y s e l f - e xt e n di n g , lexicon

(3)

In d e si g n i n g MORPH, there we re c o n s i d e r a t i o n s of a theor eti ca l as well as a p r ac ti cal nature. They w ere the f o 1 l o w i n g .

First, we wa nte d to test the h y po t h es i s , that the proper role, and the only role, of m o rp h o l o g y in word p e r c e p t i o n is in the p r oc e s si n g and learning of u n k n o w n words. U n kn o w n word p r o c e s s i n g p r oc e e ds by " d e c o m p o s i t i o n first, retrieval second", to use the theor eti ca l t e rm i n o l o g y of Job Sartori 190^ in c h a r a c t e r i2ing a p o s s i b l e model of word p roc es sin g. The p r o c e s s i n g of known words, by co nt r a st and on our hypothesis' , p r o c e e d s by retrieval of the e n t i r e word and its a s so c i at e d p r o p e r t i e s (part of speech, features, and m o r p h o lo g i ca l d e c o m p os i t io n ) from a list ("retrieval first, d e c o m p o s i t i o n second"). The so ur c e of this hy pot h es i s , and the hybri d model of word p r oc es sin g, was re se a i ch on speech s y nt h e si s (Byrd &- C h o d o r o w 1985. Ch ur c h 1985). A s s o c i a t i n g p r o n u n c i a t i o n s with w r it t e n words is g e n e r a l l y c o n s i d e r e d to be m e di a t ed by two pr oce sse s, one that r e t r i e v e s the stored p r o n u n c i a 11 on of a word, another that de> ives it fi om general

letter to sound rules. Chijrch 1985 g i s/es a good argu me nt wh> the task of a s so c i a t i n g p r o n u n c i a t i o n s with wo rd s cannot be r edu ce d e n t i r e l y to either listing them or d e ri v i ng them:

"Both a p p r o a c h e s h a ve their a d v a n t a g e s and d i s ­ ad van tag es ; the d i c t i o n a r y ap pr o a ch fails for u n kn o w n wo rds

(e.g. proper names) and the letter to sound ap pr o a ch fails when the word does not fo ll o w the rules, which h a pp e n s all too of te n in English. Most speech s y n t h e s i z e r s adopt a hybrid strategy, using the d i ct i o n a r y when a p p r o p r i a t e and letter to sound for the rest."

(4)

c o m p u ta t i o n a 1 l i ng u ist ic s (e.g. W e h r 1 i 1985). For an o v e r v i e w of psycl\o1 ingui st i c work on m o rph olo gy , see H e n d e r s o n 1905, and fot a set of recent st ud i e s of m o r p h o lo g i ca l c o n s t r a i n t s on word i-ecogn i t i o n , see Jarvella, Job, S a n d s t r ö m 8. Sc hreuder

(fortticomi ng ) .

Second, we wa nte d a ps ych o 1 i n g u i s t i c a 11y and c o m p u t a t i o n a l l y p l a u s i bl e model of how w ord s are e nte re d into the lexicon of ^nown words of a la nguage user, as a f u nc t i on of langua ge use. Most e.;isting c o mp u t a t i o n a l m o de l s of natural la nguage deal with the p r o c e s s i n g of s e n t e n ce s r e l a t i v e to a fixed body of language knowledge, rather than with the a c q u i s i t i o n of that knowledge. P e o p l e a c qu i r e new k n ow l e dg e about a language as part of the p r oc e s s of using it, and we h ave to try to si mu l a te at least some as pe c t s of language learning b e hav ior on comput ers , it we want to shed fu rther light on learning by people, and if we want to make c o m p u t e r s more useful,

1a n g u a g ew i s e .

Third, we wanted a u n i d i r ec t i on a l model of word r e co g n i t i o n only, and not a b i d i r e c t i o n a 1 model of both word r e c o g n i t i o n and genera tio n. There was a p e r f o r m a n c e theoretic r e as o n for this, as well as a p rac ti cal one. In s tud ie s of language p r oc e s si n g (e.g. De uts ch &< J a rv e l la 198^), tasks involving p e r c e p t i o n and tasks in volving p r o d u c t i o n of the same linguistic s t ru c t ur e s have s how n in ter est in g d i ff e ren ce s, which u n d e r m i n e s the idea that it is ex ac t l y the same k n ow l e dg e of langu age (linguistic c o mp et enc e) that is at work in both p e r f o r m a n c e pr oce sse s. The p r act ica l c o n s i d e r a t i o n in mo de l i ng only word a n al y s is was that it m i ni m i ze d the problem. Fourth, and another p rac ti cal c o n s i d e r a t i o n , was the limited time a v a i l a bl e for desig nin g, i m pl e men ti ng and testing the new lexical comp on ent , a total of four weeks.

(5)

and mo rph o s y nt ac tic f e atu res w ere al re a d y e s ta b l is h e d, and p r ev e n t e d us from using d i r e c t l y any p r e - e x i s i t i n g mo rp h o lo g i ca l analy zer , such as B l a b e r g ’s two-level m o r p h o l o g y for Sw ed i s h (Blaberg 190^), or H e l l b e r g ’s S w edi sh m o r p h o l o g y in the i m pl e m e n t a t i o n of Doherty, R a nk i n 8». W i r é n (this v o 1u m e ).

3. Description of the morphological analyzer.

3.1. Overview.

The two lexicon model of word p r o c e s s i n g that we a r riv ed at has the fo ll o w in g technical c h a r a c t e r i s t i c s . There are two lexica, one of full words, c a lle d

word-lex icon,

and one of m orp he mes , ca ll e d

morpheme-lex icon

. We will d e s c r i b e the (norpheme lexicon first.

3.2. The morpheme-lexicon.

The m o r p h e m e - 1 e x i c o n is a le xicon of known morph eme s, c o r r e s p o n d i n g to the m o r p h o 1ogica 1 d e c o m p o s i t i o n s of words g ive n in the word lexicon. The iiioi p h em e lex i-oi. is a h ash

t a b l e (a ss o c ia t i on list) with m o r p h e m e s as keys -itid features

as

values.

The a d v a n t a g e of this r e p r e s e n t a t i o n over other r e p r e s e n t a t i o n s ( d is c r im i n at i o n ne tworks, or tr eating words as Lisp atoms) is that the hash table e n ab l e s a group of values to be a s so c i at e d with a si ng l e key and r e tr i e ve d all at once. Si nce m o r p h e m e and word a m b i g u i t y is common, that is useful. Further, the m o r p h e m e - 1e x i c o n is so rte d by values, me an i n g that m o r p h e m e s with the same f e at u r es are gr oup ed together. The f o l l o w i n g is an e x a m p l e of the s t r u c t u r e of en tr i e s in the m o r p h e m e - l e x i c o n (note that “flick" h ere is a morphe me, not what H e l l b e r g 1970 and o t he r s ( Käl lg ren 1906) have c a lle d a technical stem):

KEY

"flick"

VALUE

( #

1 N

2

-WF3

COMBINATORIAL FEATURES

-•-UTR)

BINARY FEATURES

We will now describe the c o m b i n a t o r i a l f e at u r es and the b i na r y

features of a morpheme. The forme r e n co d e the c o m b i n at o r ia l

possibilities of the morpheme, and the latter are s t ri c t ly

(6)

P o s i t i o n 1. The first c o m b i n at o r ia l f e at u r e e n co d e s what part of speec h the m o r p h e m e c o m b i n e s with on its left, what it

takes. The o p ti o n s are:

NIL

m o r p h e m e takes any part of speech on its left m o r p h e m e takes n o th i n g on its left

NOUN VERB ADJ PREP ADV

m o r p h e m e takes s p e c i f i e d part of speech on its left

P o s i t i o n S. The s e con d c o m b i n a t o r i a 1 f e at u r e en co d e s what part of speech a m o r p h e m e makes. The o p ti o n s are:

NOUN VERB ADJ PREP ADV

P o s i t i o n 3. The third c o mb i n a t o r i a l f e at u r e e n co d e s whet he r or not the m o r p h e me can occur in word final positi on. There are

just two options:

-UP m o r p h e me c o m b i n e s with s o m e t h in g on the right, i.e. it is non word final

-♦-UIF m o r p h e m e c o m b i n e s with n o th i n g on the right, i.e. it is word final

The w o rd- f i ria 1 i ty f e at u r e e n ab l e d us to do morphological

a n al y s is wi th o u t zero mo rph e me s , which seemed desirable. An e x am p l e of how it is used is the follow ing :

■ k o / __ #(w b) has f e at u r es ( * N O UN +WF -»-UTR -PL -DEF)

■ k o / _ _ m b) has f e a t u r e s < ♦ NOUN -WF -»-UTR) and no more.

It will get the f e a t u r e s for numbe r and definiteness from what

(7)

P o s i t i o n Bi nar y featu res . For the noun, a d j e c t i v e and verb system, they are;

+UTR +AUX

-UTR --AUX

4-PL +REAL

PL -REAL

+DEF +FIN

-DEF -FIN

fPRS -PRS

The bi nar y f e at u r es are st ric tly binary, i.e. the co mpl e me n t of -«-UTR I S -UTR, etc. Bi nar y f e a t u r es are al wa y s fully specified, thus there are no r e d u n d a n c y ru les that m e di a t e b e t w e e n p a r t i a l l y and fully s p ec i f ie d fe at u r e values (the a bse n c e of a s p e c i f i c a t i o n for a feature m ean s that the item can p ass the r e q u i r em e n t of both the va lu e + and the va lu e - for that feature) . There w ere two re as o n s for this, one being that it is do ubt ful that r e d u n d an c y rules are of any practi cal use in a p r o c e s s i n g model, the other being that the ab se n c e o f r e du n d an c y rules fac ilitates d e b u g g i n g and exten d i ng the lexicon. At any point of w ork in g with it, the fe at u r e s p e c i f i c a t i o n s you see is what the p r oc es sor gets. t o o . Be low are sa mp l e s of gr ou p s o f en tr i e s of lexical m o r p h e me s and partial e n tr i e s of some gramm ati ca l morphe mes .

"flick" ( * N -UF -«-UTR) " g r y t " ( * N -UF -«-UTR)

" k lock " ( M- N -UF + UTR) "po jk" ( » N -UF ^UTR )

"dru 1 1 " ( N -UF + UTR ) "män" ( N -UF -«-UTR) "h i m 1 " ( N -UF -«-UTR )

"a" ( (A A ■^WF -«-UTR -«-PL -«-DEF) a /' de söt-«-a tä-«-r-«-na (A A -«-UF -«-UTR ^-PL -DEF ) a / söt-«-a ta-«-r

(8)

(N N -»■WF ■»-UTR -PL -DEF ) a / f 1 ick-»-a (N N -»-WF -UTR -»-PL -»-DEF ) a / bi-»-n-»-a

( V V ^WF -AUX -»-REAL -FIN -»-PRS) a / verk-»-a (inf) ( V V -»-WF -AUX -REAL) ) a / ver k -»-a ( i mpv )

( N Ivl -»-WF -»-UTR -PL -»-DEF) n / f 1 i ck-»-a-»-n ( N N -»-WF -UTR -»-PL -DEF ) / b i -fn

( M Tvl -WF -»-UTR -»-PL ) or / f 1 i c k-»-or -»-na ( N N 4-WF -»-UTR »-PL -DEF ) ) or / flick-»-or

( V A -»-WF -»-UTR -»-PL -»-DEF) na / de druck-»-na ko-»-r-»-na ( V A -»-WF -»-UTR ^-PL -DEF ) na / druck-»-na ko-»-r

( V A -»-WF -»-UTR -PL -»■DEF ) na / den dr uc k » na ko-»n ( V A -»-WF -UTR -»-PL ^DEF ) na / de druck^-na bi-»-n-»-a ( V A -»-WF -UTR -»-PL -DEF ) na / druck-»-na bi+n ( V A -»-WF -UTR -PL -»-DEF) na / det druck-»-na bi-»-et ( N N -»-WF -»-UTR -»-PL -»-DEF ) ) na / f 1 ick-»-or-»-na

The general ap pr o a ch to s p ec i f yi n g word s t ru c t u r e in M O RP H is e x ce e d i n g l y simple. Word = m o r p h e m e “ m inu s those s e qu e n c e s of m o r p h e me s ruled out by the m o r p h o t a c t i c c o n s t r a i n t s e nco de d in the c o m b i n a t o r i a 1 features. That those fe at u r es go a long way towards s p ec i f yi n g a d m i s s i b l e s e q u e n c e s in real c a ses is illust rat ed by the f o ll o w in g e x a m p l e of three m o r p h e m e s and their sp eci fic a t io n s ;

" a v ” ( (NIL PR EP -W F ) (NIL PR EP -»-WF) )

av '' av-»-led-»-a av / av f 1 ick-»-a-»-n ” led" ( ( * V -WF ) 1 ed / 1 ed-»-er

a V-»-1 ed-»-er

bransch-»-led-»-ande V -»-WF -REAL) ) 1 ed / led 1

"n i ng " ( ('-J N -WF ) n i ng / 1 ed-»-n i ng-»-en N -»-WF -»-UTR -PL -DEF ) ) ning / led-»-ning

G i v e n this set of three m o r p h e m e s

avf ledv ning»

thi number of theoT-etically poss i b 1 e wo rds (up to trii w o r d s ) made out of them are; 3 + 3»’ -»- 3'» = 39.

(9)
(10)

(in a c c o r d a n c e with a h y po t h e s i s of word p e r c e p t i o n of C a r a m a z z a et al 1905).

Below are e x am p l es of the s t r u c t u r e of e n tr i e s in the word- l e x i c o n .

“flicka" (((NOUN +WF -PL -DEF +LITR)) ("flick" "a")) "flickan" (((NOUN -»-UIF +DEF -PL -*-UTR) > ("flick" "a" "n")) "flickor" (((NOUN «-WF -*-PL -DEF + U T R ) ) ("flick" "o."))

"f lickorna" (((NOUN +WF +DEF -t-PL + U T R ) ) ("flick" "or" "na")) These are copi ed with one minor ch an g e from the curren t en tr i e s in the w o rd - l ex i c o n , and they reveal an u n n e c e ss a r y feat ur e of each entry, + W F . S i nc e ev ery word, by d e f i n i t i o n is word final, this f e at u r e shou ld be removed. The e n tr i e s ab ove are very s i mp l e and do not i l lu st rat e what the e n tr i e s of wo rds with a m b i g u o u s a n a l y s es look like. B e low are two ex amp 1e s

-"b iIdrulie'

( ( (VERB +WF --AUX -PRS + FIIM1 +REAL) ("be" "t a 1 ’' " a d e " ))

( (ADJ EC TIV E +UIF + UTR + PL -•-DEF) (ADJEC TIV E +WF +UTR +PL -DEF ) (A DJECTIVE -•-UF +UTR -PL + DEF ) (ADJ EC TIV E -•-WF -UTR + PL ■•-DEF) (ADJEC TIV E -»-WF -UTR + PL -DEF) (A D J E C T I V E + WF -UTR -PL ■»-DEF ) ("be" "tal" " a '■ "d" "e" ) ) )

( ( (NOUN -^UIF -PL -DEF 4UTR) ("bil ( (NOUN +-WF -PL -DEF +UTR) ("bil

Using

MORPH-The central fu nc t i on in the M O R P H p r o g r a m is ca ll e d lookup- word, and it has the f o l l o w in g easy to u n d e r s t a n d d e fi n iti on ;

(defun lookup-w or d (word)

(cond ((1o o k u p - w o r d - i n - l e x icon word))

( (a n a l y z e - w o r d - a n d - a d d - t o - l e x icon w o r d ) ) ((a s k - a b o u t - w o r d - a n d - a d d - t o - l e x i c o n word)) ))

(11)

internall) c o n s i s t e n t m o rp h o lo g i ca l d e c o m p o s i t i o n s of it, r e la t i ve to the curr en t m o r p h e m e - 1 e x i c o n . These a n al y s es are retur ned as the va lu e of the function, and the result i n co rp ora ted in the w o r d - 1 ex i c o n . If the f u n c t i on a n aly ze- Nord". .. is un su c c es s f ul , the p r o g r a m asks tlie use. to supply i n fo r mat io n about the word, as a last lesort. Several u t i l i t i e s for the manual e x t e n s i o n of the lexica are ir.cluded in MORPH, in order to m i n i m i z e the amount of typing done. In that sense it is a good e x a m p l e of a lexicon w r i t e r ’s workbe nch . For example, if a new m o r p h e m e is e x ac t l y like a pr ev i o us l y known one, with respect to part of sp eec h and f e at u r es (and d e gr e e and n a tur e of lexical a m big uit y) , then this in fo r m at i o n is a u t o m a ti c a ll y co pie d when typing the m o r p h e m e it is like, at the a p p r o p r i a t e point in the

in ter act iv e sequence.

As intended, M O RP H has r e pla ced the old lexicon of S W E D I S H - SYNTAX, and the in te r a ct i o n wo rks out well. For example, in the case of an a m bi g o us word like " b e t a l a d e ” above, the new lexicon ha nds the parser all of the analyses, and the parser picks the a p p r o p r i a t e one accordi.'tg to the co nte xt in which the word occurs. The fo ll o w in g tree d i a g r a m s p r od u c ed by SUED IS H-SYNTAX i l lu st rat e the ii.teraction b e tw e e n the lexicon and the parser.

Fig. 1 Ana 1ys av

Hon betalade boken.

Fig. c’ A n a 1 y's a v'

Hon gav honom den betalade boken.

(12)

r i n a 11y » we would like to pres en t some p r e l i m i n a r y figu re s that shtow how long it takes for the p r o g r a m to do looku p-w or d by 1o o k u p - w o r d -1n - 1 e x ic o n (R = re trieval) on the one hand, and by ana 1y z e - w o r d - a n d - a d d - t o - l e x icon (A = analysis) on the other hiand . The words we tested are p r e s e n te d in d e sc e n d i n g order of m o rp h o lo g i ca l c o mp l exi ty , and the last word is an e x am p l e of an a m b i g u ou s l y d e c o m p o s a b l e word.

#kyrk+klockH

#b i b 1 1

#till-«-tal+a nd e# #f r an-» tag-»-ni ng# #kyr k -r t ag + n 1 ng#

#b i 1 -»-dr u 1 1+e# #b i I d-»-r u 1 1-»-e#

(13)
(14)

Church, K., 1985, St re s s a s si g n me n t in letter to sound rules for speech synth esi s. P r o c e e d i n g s of the 23rd Annual Me et i n g of the A s s o c i a t i o n for C o mp u t at i o na l Li ngu i st i c s, pp 2^6-253. D e u t s c h1

W.

. Jar ve11a

S p r a c h p r o d u k t i o n und

H e r r m a n n (eds) , Karl ___ _____ ... A x io ma tik der S p r a c h w i s s e n s c h a f t e n Kl ost e rm a n n, pp 173-199.

el la, R., 190^, A s y m m e t r i e n zwis ch en S p r a c h v e r s t e h e n , in C.F. G r a u m a n n & T. 1 B u hl e r s Ax io m a ti k - Fu nf z i g Jahre F r an k f ur t an Main,

Doherty, P., Rankin, I. &< Wirén, M., 1986, E r fa r e n h e t e r av en i m pl e m en t e ri n g av H e l l b e r g s s y ste m för svensk morfol ogi , i F. K a r l s s o n (utg). F ö re d r ag från V N o rd i s ka D a ta 1 ingV i s 11k d a g a r n a , 11-12 dece mb er, 1985, He lsi n gf o r s.

Ejerhed, E., 1983, K o n g r u e n s i en svensk finite s tat e parser, fo re d r ag vid :e sv en s k a k o ll o k vi e t i sp rå k l ig d a t a b e h a n d l i n g - d a t o r s i m u l e r i n g av ve rbalt bete en de, 26 maj 1983, Lund.

Ejerhed, E., 1985, En y t s t r u k t u r g r a m m a t i k för svenska, i S. A l lé n et al (utg), S v e n s k a n s b e s k r i v n i n g 15, I n st i t ut i o ne n för no rd i s ka språk, G ö t e b o r g s un iv e r si t e t, s 175-192.

Ejerhed, E. Church, K., 1983, F i ni t e S tat e Pa rsing, i F. K a r l s s o n (ed). Pa pe r s f rom the S e ven th S c n a d i n a v i a n C o nf e r e n c e of L i ng u ist ic s, U n iv e r s i t y of Hels in ki, De pa r t me n t of General L i ng u ist ic s, P u b l i c a t i o n No. 10 (II), pp '^lO-^Sø.

He llberg, S., 1978, The m o rp h o l o g y of p r ese nt day Swedish, Stock hol m, Al mq v i st i Wiksell In ternational.

Hende rso n, L., 1985, To war d a p s y c h o l o g y of m orp he mes , in A.W. Elli s (e d ) , P r o g r e s s in the p s y c h o l o g y of language, Vo 1. 1, London, L a w r e n c e E r l b a u m A s so c iat es .

Jarvella, R., Job, R., Sa nds trö m, G. &< Sc hre u de r , R. (f or t h ­ coming), M o r p h o 1o g i c a 1 c o n s t r a i n t s on word r e co gn iti on, De pa r t me n t of General L i n g u i st i c s, U n i v e r s i t y of Umeå.

Job, R. &C Sartori, G., 198^, Mor pho 1 og i ca 1 d e c o m p os i t io n : e v i d e n c e from cr os s e d p h on o l og i c al d ysl ex ia, The Q u a r t e r l y Journal of E x pe r i me n t al P s y c h o l o g y 36A, pp ^35-^50.

Figure

Fig. 1 Ana 1ys av

References

Related documents

The above window shows a small Reload system with two post offices and two domain profiles active. If you have not yet created a domain or post office profiles, then this screen

Davis came before the Board seeking licensure as a Certified Respiratory Therapist.. Davis was convicted of a DUI in

Program faculty are generally satisfied with most of the achievements of our course and programs as reflected in student success while at EOU and in the years that follow.

basement windows, one in the back and one on the side, with glass block windows. They would like to replace the shutters in the design submitted in the color Wedgewood

Data collection for the Class of 2013 ended in February 2014, to align with emerging standards to report data as of 9 months after graduation. Data collection for the Class of

True third-generation information-security management tools, such as Symantec Corporations Expert 4.1 system are able to provide rational decision-support systems by building

a drop in public commitment to research was compensated by business research expenditure.. Businesses sought to conjugate the

This study has raised a number of concerns and questions about architectural photographs of India and they fall into two broad groups: the first group focuses on the documentation