Incremental Identification of Inflectional Types 
Petra Barg and James Kilbury 
Ileinrich-Heine-UniversitSt Diisseldorf, 
Seminar flu Allgcmeinc Sprachwissenschaft, 
Univcrsitiitsstr. 1, D-40225 Diisseldorf, Germany 
E-maih {barg,kilbury}((~ling.uni-duesseldorf.de 
Abstract 
We present an al)proach to the increnmntal ac- 
crual of lexical int'ornmtion fl)r unknown words t;hat 
is consl;raint-based and conll)atil)le with s(.andard 
unification-liased granlnmrs. All;hough the tech- 
niques are -ind(:l)en(lent and can l)('. al)plied 
to all kinds of informal;ion, ill (;his 1)al)er we concen- 
(;rate on the domain of German noun intl('ction. We 
show how morl)hological intbrnm.tion, est)ecially in- 
flectional class, is successfully acquired using a tyl)e- 
based HPSG-like analysis. 15lrthernlore, we sketch 
an alternative strategy which nmkes use of finite.- 
sl;ate (;rans(lucers. 
1 Introduction 
Systenls for nai;tn'al  l)roc('ssing nms(: deal 
adequately with "unlmown" words, i.e. lex(:mes (hat 
either have been newly coin(:(l or else have not been 
included in a 1)articular lexicon (of. Kilbury et al. 
(1994)). l{ather l;han simply regarding mlknown 
words as noise, our syst;(:m insl;(;ad uses t:h(-ir con- 
text as a source tbr 1;11(: systematic accrual of lcxical 
information (;hat (:all 1;hen lie ul;ilized. 
()m' al)l)roach (litthrs in signitican(; resl)ects frolll 
those of oi;her investigators. It is designed fin' 
unifical;ion-bas(:d grallllnar fbrmalisms wilh t.yl/cd 
featur(: stucturcs as in HPSG and is not restricl;ed to 
simple nlorl)hosyni;acl;ic fea~llre8. \[11 COlltrasl; ~;o sIA/- 
l;istical approaches like that of (Brent, 1!)!11), which 
often do not work increnmntally and are intended 
tbr tilt application to large corl)ora, ours instead 
aims at a detailed grammatical analysis of individual 
sentences with a maximal use of their information. 
While systems like l;hat of (Hahn el, al., 1996) deal 
with the general acquisition of concepts, we are con- 
cerned exclusively with the acquisition of s(A'ucl;ural 
linguistic infornmtion. 
All, hough we deal here with German noun in- 
ttection, in a ti'amework close to that of (Riehe- 
mann, 1998) and (Koenig, 1999), the techniques are 
-indel)endent and al)ply to other ldnds of 
lexical inibrmation as well, as is shown in (Walther 
and Barg, 1998) with respect to valency intbnnation. 
Thus, in contrast to (Ehrlich and Rapaport, 1997), 
who employ tailored algorithms for the acquisition 
of intbnnation al)out nouns and verbs, we introduce 
an apl)roac.h thai; is COml/letely general with )'esl)ect 
to the kind of structural linguistic information ac- 
quire& 
2 German noun-inflection classes 
There is a w/st literature on German noun inflection 
represented in recent studies by (Cahill and Gazdar, 
1999), (Clahsen, 1999), and (Neef, 1998). Here we 
stmnnarize only essential points and ignore highly 
irregular and archaic inilections (cf. figure 4 below). 
Gernmn nouns l)ear gender (masculine, tbininine, 
neutt,r) and are inflected for number (singular, 1)lu- 
rail and case (nontilmt.ive, accusative, dat.ive, geni- 
(ire). \Vith the exeq)(;ion of class NWN (e.g. mast 
Ha.'uvr 'farnwr': wi(;h gen sg f~aucrn,), all nonfl:minine 
nouns build genitive singular with -s. 
The "regular" (of. ()lahsen (1999))but; "atytfi- 
cal" (cf. \\qlnderlich (1999)) nouns of class NA (e.g. 
Auto ~car') build their plural forms in -s. The l)lural 
forms of all o(;her (i.e. %yl)ical') (:lasses nmst end 
ill a so-called schwa syllable -c, -el, - c'r, or -en (i.e. 
l)honetically all unstressed \[o\] ta)llowed by a sonoranl, 
f~'olll \[1 , .\]). 
S(;l'()llg lIOllllS add -e for l)lural ill class NS (e.g. 
A'r'm, 'arm', pl A'rm, c) and class NU (e.g. d~'zt 'l)hysi- 
clan', pl /{rztc) if (;he sl;enl itself does not already 
end in a schwa syllable (e.g. Kabcl 'cable', pl Ka- 
bell. Class NU flwthermore umlauts the stem (i.e. 
replaces a, o, ,it,, au with ii, 5, ii, iiu, respectively), as 
does (:lass NR (e.g. Mann 'man', pl \]VIiinncr), which 
adds - er. 
The remaining (:lasses (NM, NWN, and NWS) 
t'onn their plural ill -n (e.g. Schraube 'screw', pl 
Sch, rauben). The nonnominative singular stenl in 
class NWN (e.g. lIasc 'hare', gen sg Hasen) and 
class NWS (e.g. Glaubc 'belief', gen sg Glaubcns) 
is identical with the phlral fornL while N\VN excel> 
(;iolmlly adds no -s in genitive singular. 
All (:lasses except NA build dative plural by 
adding -n (;o the phn'al Ibrm if it is not already 
present (e.g. Miinncr 'men', dat pl Miinncrn but 
Itascn 'hares', dat pl Hascn). 
49 
Figure 1: hierarchy of inflectional schemata 
(nil_sort 
s_gen~sg~sort n_dat_pl_sort l)lur_infl~sort n_obl_sg_sort n_gen~sg_sort 
s_pl-sort strong_pl_sort n_pLsort 
schwa_pl~sort erq)l_sort umlaut_pl~sort 
3 Representation of inflectional 
morphology 
Various proposals have been made for the represen- 
tation of inflectional morphology within constraint- 
based frameworks like HPSG (cf. Pollard (1994)). 
We neither adopt a word-syntax al)proach like that 
of (Krieger and Nerbonne, 1.993) assunfing lexical 
entries tbr inflectional affixes as well as roots, nor do 
we make use of lexical rules, as (Meurers and Min- 
nen, 1997) do. 
Instead, we follow (Riehemmm, 1.998) in formu- 
lating hierarchically structured schemata of the kind 
she has developed for derivational morphology l)ut 
apply them here to inflection and thus carry out a 
kind of inflectional analysis without lexical rules as 
projected by (Erjavec, 1996). Our schemata cap- 
ture inflectional paradigms and can be regarded as 
rela.tional constraints that relate stems, affixes, and 
inflected lexical forms. 
Figure 1 shows our hierarchy of inttectional 
schemata, wlfile figure 2 illustrates a concrete 
schema, namely that tbr the schwa plural of inflec- 
tional (:lass NS. Ill figure 2 l;he attribute ftype stands 
fbr the intlectional class. The attributes flex, surf, 
and base represent strings, namely the inflectional 
ending, surface (i.e. inflected) tbrm, and base form 
respectively. The symbol @ denotes the reduced 
vowel \[o\] (schwa), and - designates negated values. 
Lexical entries are assmned only for basic lexical 
signs (i.e. uninflected but possibly derived or com- 
pounded). Inflected lexical signs result from the in- 
teraction of these lexical entries and the inflectional 
sehenlata. Figure 3 gives the basic lexical sign (with 
the onfission of feature specifications that are irrel- 
ewmt for this discussion) for Hund 'dog', which is 
of class NS, followed by the inflected lexical sign tbr 
Hunde 'dogs', in which the value of the attribute 
moph (i.e. morphophonology) is an extension of the 
schema for schwa phlral given in figure 2. 
The inflectional classes assigned to basic lexical 
signs are modelled as formal types in the hierarchi- 
cal structure specified in figure 4. Note that the 
leaves of this tree correspond exactly to the inflec- 
tional classes of German nouns as described above 
in {}2. 
Morphophonemic and morlfltographemic alterna- 
Figure 2: schema for schwa-phn'al (sehwa_pl_sort) 
"schwa_pl 
ftype ns 
flex © 
surf atom 
base atom 
basic_Is J stem moph basic 
agr \[case-avml 
case |ctxt ~ daq 
agr kgen top J 
hum plu 
gend - ,fcm 
?ers pcr.s 
(;ions as shown in nominative plural Zcit-cn 'tinles' 
but Gabel-n 'forks' are also covered ill our descrip- 
tion. Here (he real(sat(on of the plural ending -'n 
del)ends on the shape of tlm noun steul (nalnely, 
whether or not it ends in a schwa syllable). In 
agreement with (Bird and Klein, 1994) and (Er\]avec, 
1996), we capture such alternations declaratively in 
a one-level model without recourse to transducers. 
Our treatment of umlaut adopts part of the tech- 
niques of (%'ost, 1993). 
4 Processing unknown words 
In our al)proach linguistic prol)erties of unlcnown 
words are inferred fl'om their sentential context as 
a byproduct of parsing. After parsing, which re- 
quires only a slight modification of sl;andard lexi- 
cal lookup, lexical entries are al)propriately updated. 
One of our key ideas is a gradual, information-based 
concept of "unknownness", where lexical entries are 
not unknown as a whole, but may contain unknown, 
i.e. potentially revisable, pieces of information (cf. 
Barg and Walther (1998)). This allows a uniform 
treatment for tile full range of lexical entries from 
completely known to maximally unknown. As dis- 
cussed in (Barg and Walther, 1998), our system has 
been implemented in MicroCUF, a derivative of the 
tbrmalism CUF of (DSrre and Dorna, 1993). 
50 
Figure 3: feature sl~ruci;ures for \]I'lmd and Hv.nd(; 
ba.sic_l.s 
-basic 
ftype n.5 
surf KI h.uml 
base E\] 
a.q'~' 
moph rca.sc'_avm \] 
case /ctxt" go.,| 
agr Lgen r:(~.sc j 
IIU Ill ,siw\] 
gend ?tt, a,sc 
pers third 
synsem ... 
-inJlccted_Ls 
"sch, w(t_l~l 
ftype \[\] 'n,u 
flex ~) 
surf h,'u, mh~ 
base \[\] h,u'l~d 
agr \[ 
ca.sc_avm/,j 
case ctxt " d(~ 
agr Lgen (:(l,s~: j 
hum plu 
gend ~ 'ma,sc 
:)ers \[\] tll, ird 
moph -ba.~ic_l,~' 
-ha.sic 
ftype \[~ 
surf \[\] 
base \[\] 
s\[em 
synsem  ... 
moph 
~gr 
synsem \[\] 
a~\]r 
\[ca,sc_av'm. \] 
case ctxt ~ gen 
Lgen case j 
hum ,slug 
gend 
)ors ~i~ 
I{evisable intbrmagion is fllrl;her classitied an spc- 
cializable or generalizable, where the tbrmer can only 
become 111or0 special, and I;he lal;l;er only lllOrO gen- 
eral, with furl;her cont;exts. Spe(:ializable ldnds of 
infbrmal;ion include seman(;ic t;ype of nouns, gen- 
(ler, and intleci;iolml class. Among the generalizable, 
kinds of intbrmaCion a.re i;he selecl;ional rcsl;rictions 
of verbs and adjecl;ivcs as well as (~hc case of nOUllS. 
Both kinds of intbrmaI;ion roger;her wil;h nonrevis- 
able (i.e. st, ri(:t) iutkn'mal;ion can cooccur in a single 
ent.ry. 
The overall approach in compatil)le wit;h sl;andard 
consl;railll;-1)ased analyses and makes only a few ex- 
tra demands ()11 the grmnmar, tlere, l;he revisable in- 
tormal;ion musl: 1)e e, xl)lMIly lnarkcd as such. Since 
ore' model is sii;uai;ed wii;hin (;he framework of (;yl)cd 
feagurc-based formalisms (of. Carpentx~r (1992)), 
revisable information is expressed lit terms of fin'- 
real tyt)es. The iniIJal values fin' revisable intbr- 
mat ion arc, specified with (;wo dist;inguished 1;ypes 
u_.s and u-9 for specializable and generalizable in- 
formation, resi)ect;ively. Type tmiticai;ion can be 
employed for the combination of sl>ecializable inf'or- 
mat.ion, whereas generalizable illtbrmatioll requires 
l,ype lllli()ll. 
The (lirecl; combim~l;i(m of revisable informal;ion 
(luring parsing is mffeasible for various reasons dis- 
cussed in (Barg and \Vail;her, 1.998). It; conscquenl;ly 
is carried oul; in a selmrai;e st('~ 1) after ghe curreni; sen- 
(:ell(:(; has heel/. \])arse(t. The gralnmai;ical amdysis 
i(;self I;hus remains coml)lei;(;ly declaral;ive and only 
makes use of mfiticalJon. In order ix) achieve |;his 
sel)aral;iou of analysis and revision we inl;roduce Lwo 
at;I;ribul;es for generalizable informal;iol h namely gen 
and ctxt, where ctxt receives l;he information inferred 
from l;he seni;enl;ial contexl;, and gen the polxmtially 
re, visable inforlnai;ion wil;h I;11(; inil;ial value u_ 9. 
Parsing l;hus proceeds in an entirely COllVt;lll;iollal 
lnamlel', excepi; thai; lexical look-up for a word wil;h 
tlIlI{IIOWII orl;hogralflly or 1)honology does noI; fail but. 
iustead yields an mMersl)eciiied canonical leM(:al Oll- 
I:ry. The Ul)(lal;ing after parsing (:Ollll)ares Ihe feal;m'e 
st ruct.ure of (;he origiual lexical entry with that. ill- 
f'errcd conl;exl;ually. The sl)ecializable infl)rlnal;ion of 
(;11(; forlller in replaced wil;h the (:orr(;slxmding values 
ot:' (;he lal;lxn'. Moreover, usiug the at.tribut.es gen and 
ctxt inl;l"Odut:ed above, the new gen value for general- 
izable intbrmal;ion is compul;e,d by t;he l;yl)c UlriCh of 
l;hc gen value front/;lie old lcxical elli;ry (initialy 'u_9) 
with the ctxt value resulging from (;he l)arse. Actual 
re, vision nal;urally in only carried ouI; when n conl;ex(; 
in fact; provides new informal;ion. 
5 Incremental inference of 
inflectional information 
In order to process llllklloWll word forms, we posl;tl- 
late canonical lexical entries which are ret;urned by 
lexical lookup if a word is hog recorded in the lexicon. 
For nomls, Ichis enI;ry corresponds 1;o an mlderspeci- 
fled basic lexical sign in which l;he inflectional class, 
case, number, and gender are specitled with revis- 
able types, i.e. the information can be acquired and 
updaix'~d. Figure 5 shows (;he basic lexical sigll for 
German norms (with the Olllissioll of tbai;m'e Sl)eciti- 
cal;ions ~haI~ arc irrelevanl, fi/r l;his (tiscussion). 
Whereas intleci;ional class (ftype), number (num), 
51 
Figure 4: hierarchy of inflectional types 
nouiL\[lox 
NA typieal_tlex 
strong_flex nasM_tlex 
NS NU NR NM weak_flex 
NWN NWS 
and gender (gend) are specializable, case is general- 
izable and hence contains the features gen and ctxt. 
Note that the initial values for specializable infor- 
mation consist of a disjunction (;) of the value u_s 
and the most general appropriate value for the corre- 
sponding feature. This ensures the identification of 
specializable infornlation (via ~t_s) on the one hand, 
and the correct specializations on the other. 
\Vhen a sentence containing an tlnknown noun is 
parsed, infbrmation about the noun conies from dif- 
ferent som'ces: while the surrounding context lnay 
supply agreement information, the word fornl itself 
together with morphol)honological constraints may 
restrict the possible inflectional class. 
As an examt)le we can suppose that the rather in- 
frequent noun Sund 'sound', 'strait', which like It'und 
'dog' belongs to class NS but is unfalniliar to inany 
German speakers, is not recorded in a given lexicon. 
The class NS contains both masculine and neuter 
nouns, and these differ ill none of their inflected 
forms. Thus, only agreement information from a 
context, such as dcr cnge £'und 'the narrow strait' 
(nonfinative), call establish the gender of S.w(td as 
being masculine. 
Figure 5: feature structure for the underspecified 
lexical entry 
-basic_Is 
-basic 
ftype (noun_flex;u_s) 
surf \[\] atom 
base \[\] 
"czcjl ~. 
moph \[ case_aw~ 
case /ctxt case 
~gr kgen u_g 
num (num;u_s) 
gend (9cnd;u_s) 
)ers third 
synsem ... 
Even in isolation, the forln Sund must be singu- 
lar since its final shape is not coml)atible with any 
phlral inflection (i.e. it ends neither ill -s nor ill a 
schwa syllable). Moreover, the morphoplionological 
constraiuts on stems allow only three possibilities: 
S'und is 
• femiuine (and then tile class is NA, NU, or NM 
and tile case is underspecified) 
• nonfenfinine and weak (i.e. (:lass NWN or 
NWS) (and tlmn the case must be nominative) 
• nonfbminine and nonweak (and then the case is 
not genitive) 
These hypotheses are captured in the three feature 
structures depicted in figure 6. 
As we have seen, when a word is parsed in context, 
this provides additional information. If we know, for 
exalnple, that S~tnd is lnasculine, the first hypothesis 
is excluded, and the gender specification of the re- 
maining two hylmtheses can be specializ&l to masc. 
If we additionally encounter S'und ill dative singu- 
lar, which is impossible for weak nouns (which nmst 
have a final -n), then only the third hypothesis re- 
mains. Finally, if the plural form S'undc occurs the 
system can specialize the inflectional class exactly to 
the type NS. The other morphological information 
cammt be further generalized or specialized, and we 
have the final lexical entry fbr Sund. 
Things are not always this easy. In particular, 
there may be a number of alternatives both fbr the 
segmentation of a form into a stem and an inflec- 
tional ending and ibr the ~ssignment of a stein to a 
lexeme. Moreover, these alternatives may depend on 
each other. Thus, the form Lcincn may be assigned 
to any of the lexemes Lein 'flax' (masc, NS), Leine 
'rope' (fern, NM), or Leincn 'linen' (neut, NS); even 
in a context, e.g. F'ritz verkauft Leinen 'Fritz sells 
ropes/linen', it may be impossible to disambiguate 
the form. While the nouns Band 'book volume' 
(mase, NU), Band 'strip' (neut, NR), Band 'bond' 
(neut, NS, archaic and rare in singular), Band 'music 
band' (fern, NA), and Bande 'gang' (fern, NM) may 
be unlikely to occur all in the same context, they 
ilhlstrate the dimension of the t)roblems of segmen- 
tation and lexical assignnlent, which in turn coil- 
52 
Figure 6: hyl)othescs fi)r ,5"u'nd 
-basic_Is 
basic 
ftype (ns;nu, nm;u_s) 
surf \[\] sund 
base \[\] 
ayfl" 
moph \[ casc_avn\[ 
case /ctxt (-a,sc 
tgr Lgen 'u_q 
hum (,siny;u_,s) 
gend (fcm;u_~) 
~ers third 
synsem ... 
-basic_Is 
basic 
ftype dnw;,l_,~') 
surf \[\] su'nd 
base \[\] 
moph 
agr 
synsem ... 
agr 
case_arm 
case /ctxt (:as,: 
Lgen (~om;,u_(j) 
num (sing;v_s) 
gend (" fem;v_s) 
~ers I, hi'rd 
-basic_Is 
-basic 
ftype (" nw;u_s) 
surf \[\] sund 
base \[\] 
moph 
J agr 
sy|\] Se Ill ... 
agr 
\[ casc_($~IIII 
case ctxt C(ISC. 
Lgen (" gen;'a_g) 
num (sin q;u_s) 
gend (~ fl'm;u_s) 
~ers third 
stitute part of the more general 1)robleni of disam- 
Mguation in natural  processing. \Ve have 
no magic solution f'or the latter, but in our approach 
such examples must be handled with disjmwtive rep- 
resentations until the context 1)rovides the necessary 
disambiguating infornmtion. 
6 An alternative model using 
finite-state techniques 
Alternatively, the incremental identifieation of in- 
flectional types can be modelled within the Dame- 
work of finite-state automata (cf. Sln'oat (1992)) 
without recourse to unification-based grammar for- 
malisms. A FSA can be defined that has an all)ha- 
1)et consisting of vectors specifying the stem shal)e 
and ending (and thus the segmentation) as well as 
tim agreenlcnt inforniation of possible word forms. 
Starting in an initial state corresponding to the con- 
straints that apply to all unknown words, the FSA 
is moved by successive forms of an unknown lexeme 
together with their agreement information into suc- 
cesso," states that capture the incrementally accrued 
inflectional intbrmation. The FSA may reach a final 
state, in which case tile intlectional class has l)een 
uniquely idenl;ified, or it nlay renmin in a nonfinal 
state. A lexic.on would siml)ly recoM tile latest state 
roached for each ll()llll. 
Imlflcnlentation of t.his model is greatly compli- 
cated by the problems of (lisambiguation just dis- 
cussed in {i5. In general, the states of the FSA must 
capture disjmmtions not only of intlectional classes, 
lint also of segmentation and gender alternatives. 
The application of automatic induction techniques 
to corpora appears to be essential, and we are cur- 
rently f)ursuing possibilities for this. 
7 Conclusion 
We have taken the inflec.tion of German nouns to il- 
lustrate a general tyl)e-based at/1)roach to handling 
ultkltowll words alia the illcrelllental accrual of their 
lexical information. The techniques can be al)l/lied 
not only to other classes of inflected words and to 
other s, 1)ut also to other aspects of lexical 
informal;ion such as the valency of verbs. This may 
allow practical systems for natural  process- 
ing to be enhanced so as to utilize input infornlation 
that otherwise is discarded as noise. 

References 

Petra Barg and Markus Walther. 1998. Processing unknown words in HPSG. In Proceedings of COLING-ACL'98, pages 91-95, Montreal. 

Steven Bird and Ewan Klein. 1994. Phonological 
analysis in typed feature systems. Computational 
Linguistics, 20:455-491. 

Michael R. Brent. 1991. Automatic acquisition of 
subcategorization frames from untagged text;. In 
Proceedings of 29th, ACL, pages 209-214, Berkeley. 

L3mne Cahill and Gerald Oazdar. 1999. German 
noun inflection. Journal of Linguistics, 35:1-42. 

Robert Carpenter. 1!)!)2. The Lwic of ~J)\]ped Feature 
St'l'~tCt'~t~'t'.S. Cambridge University Press. 

Harald Clahsen. 1999. Lexieal entries and rules of 
: A multidisciplinary study of German 
inflection. Behavioral and Brain Sciences. 

Jochen DSrre and Michael Dorna. 1993. CUF - 
a ~brmalism for linguistic knowledge representa- 
tion. In Joehen DSrre, editor, Computational As- 
pects of Constraint-Based Linguistic Description. 
IMS, Universitft Stuttgart. Deliverable R1.2.A, 
DYANA-2 - ESPRIT Project 6852. 

Karen Ehrlich and William J. Rapaport. 1997. A 
computational theory of vocalmlary extension. In 
Proceedings of the i9th Annual Conference of the 
Cognitive Science Societal, pages 205-210. 

Toma~ Erjavec. 1996. Unification, InheTqtance and 
Paradigms in the Morphology of Natural Lan- 
guages. Unpublished doctoral dissertation, Uni- 
versity of Ljubljana. 

Udo Hahn, Manfi'ed Klemmr, and Klemens Schnat- 
tinter. 1996. Learning from texts - a terminolog- 
ical meta-reasoning perspective. In S. Wermter, 
E. Riloff, and G. Scheler, editors, Conncetionist, 
Statistical, and S?/mbolic AppTvaehcs to Learning 
Jbr Natural Language Processing, pages 453 468. 
Springer, Berlin. 

James Kilbury, Petra Barg, mid Ingrid Renz. 1994. 
Simulation lexikalischen Erwerbs. In Sascha Fc- 
lix, Christopher Hal)el, and Gerd Rickheit, ed- 
itors, t(ognitive Linguistik: Repriiscntation ~tnd 
P'rvzesse, pages 251-271. Westdeutscher Verlag, 
Opladen. 

Jean-Pierre Koenig. 1999. Lezieal Relations. CSLI 
Publications. 

Hans-Ulrich Krieger and John Nerbom~e. 1993. 
Feature-based inheritance networks tbr comlmta- 
tional lexicons. In Ted Briscoe et al., editor, In- 
heritance, Defaults, and the Lczicon, pages 90 
136. Cambridge University Press, Cambridge. 

Detmar Meurers and Guido Minnen. 1997. A computational treatment of lexical rules in HPSG as 
covariation in lexical entries. Computational Linguistics, 23:543 568. 

Martin Ned. 1998. A case study in declarative 
morphology: German case inflection. In Wolf- 
gang Kehrein and Richard Wiese, editors, PhonoI- 
ogy and Mo~'phology of the Germanic Languages, 
pages 219-240. Max Niemeyer Verlag, Tiibingen. 

Carl Pollard and Ivan Sag. 1994. Itcad-Drivcn 
Ph, rase Structure GrammaT: Chicago University 
Press. 

Susanne Riehemmm. 1998. Type-based derivational 
morphology. Journal of Comparative Germanic 
Linguistics, 2:49-77. 

Richard Sproat. 1992. Morphology and Computa- 
tions,. MIT Press, Cambridge, Mass. 

Harald Trost. 1993. Coping with derivation in a 
morphological component. In Proceedings of 6th 
EACL, pages 368-376. 

Markus Walther and Petra Bart. 1998. '\[bwards in- 
cremental lexical acquisition in HPSG. Ill Pro- 
ceedings .\]oint ConJcrencc on For.real Grammar, 
Head-Driven Ph, rasc Str'ucturc Grammar, and 
Catcgorial Grammar; Saarbriicken. 

Dieter Wunderlich. 1999. German noun plural re- 
considered. Manuscril)t, University of Diisseldorf.  
