In ths paper we want to show that the possble analyses of ths problem wthn the framework of PSG are lmted by combnatons of the followng basc assumpton

Transcription

1 On Non-ead Non-Movement In: G. Goerz (ed.): KONVNS 9, Sprnger Verlag, 99, pp 8- Klaus Netter eutsches Forschungszentrum fur Kunstlche Intellgenz Gmb Stuhlsatzenhausweg, -00 Saarbrucken, Germany e-mal: The paper dscusses derent approaches to the analyss of nte verb poston n German wthn the framework of PSG. It argues that among three parameters (a) standard subcategorzaton and adjuncton, (b) unform structural representaton and (c) clause ntal nte verbs as heads, only two can be mantaned smultaneously, unless addtonal devces are ntroduced. We abandon (c) and present an approach whch allows to derve a bnary rght-branchng structure for the \Mttelfeld" also for V-ntal structures wthout modcatons of general prncples of the theory. The key to the analyss s the postulaton of an empty node as a functonal head whch takes the nte verb as a complement. e Arbet dskutert mehrere Analysen fur de Stellung des nten Verbums m eutschen m Rahmen der PSG. s wrd argumentert, da von den dre Parametern (a) Standard Subkategorserung und Adjunkton, (b) unforme strukturelle Representaton und (c) ntale Verben als Kopfe, ohne ene rweterung der Theore um spezelle Mechansmen nur zwe aufrechterhalten werden konnen. Wr geben (c) auf und dskuteren ene Analyse, de auch be V-ntalen Strukturen en bnar verzwegendes Mttelfeld erzeugt, ohne da generelle Modkatonen de Theore erforderlch werden. er Schlussel fur de Analyse st de Annahme enes leeren Knoten als funktonalen Kopf des Satzes, der das nte Verbum als Komplement subkategorsert. Introducton The ntal and nal poston of the nte verb n languages such as German or utch presents an nterestng problem for grammar frameworks such as ead rven Phrase Structure Grammar (PSG) and related ategoral Grammar style theores. Snce the nte verb as the head of a clause may ether precede or follow ts complements and adjuncts ()/(), the queston arses whether the two types of clauses can and should be assgned a unform structural representaton (modulo the poston of the nte verb and the complementzer). () [ Brngt ] [ er ] [ A morgen ] [ dem Kunden ] [ A punktlch ] [ de Fracht ] () da [ er ] [ A morgen ] [ dem Kunden ] [ A punktlch ] [ de Fracht ] [ brngt ] The research underlyng ths paper was supported by a research grant, FKZ ITW 900 0, from the German Bundesmnsterum fur Forschung und Technologe to the FKI project IO. We would lke to thank Sergo Balar, John Nerbonne, Make Partong and ans Uszkoret for dcussons and comments. The descrbed approach has been mplemented n the IO system on the bass of the type descrpton language TL. Throughout ths paper we wll abbrevate the abstract grammatcal functons head, complement and adjunct as, and A respectvely. In our dscusson we wll largely abstract away from the phenomena of V-second or topcalzaton and from word order varatons n the \Mttelfeld". Both of these phenomena we consder to be qute unrelated to the problem of the poston of the nte verb and for both a range of ndependent modular solutons can be envsaged. 8

2 In ths paper we want to show that the possble analyses of ths problem wthn the framework of PSG are lmted by combnatons of the followng basc assumptons: (a) The PSG treatment of complements and adjuncts s mantaned as t s expressed n the Subcategorzaton Prncple (SP) and the Semantcs Prncple (SemP). (b) V-ntal and V-nal structures are gven a unform structural representaton. (c) The verb n the ntal poston s analysed n stu as the head of the clause. We wll argue that xng two of the parameters wll automatcally lead to a rejecton of the remanng assumpton. Mantanng all three assumptons wll requre the ntroducton of addtonal devces requrng a modcaton and extenson of prncples and generalzatons of the framework. The rst approach we dscuss, by (Pollard 99), mantans the parameters (b) and (c) by postulatng a at structure for the verb and ts complements and adjuncts. We wll argue that ths approach not only complcates the treatment of adjuncton but also commts one to the assumpton of a phrase structure for whch there s lttle lngustc motvaton. As a varant of the rejecton of (a) we dscuss the possblty of a unform bnary rght-branchng structure whch nvolves a modcaton of the SP and whch has a techncally very unpleasng analyss as a consequence. Thrdly, we wll dscuss the eects of mantanng the assumptons (a) and (c) on the second parameter, above all the mplcatons of a non-unform phrase structural representaton for V-ntal and V- nal clauses. Fnally two solutons suggested by Borsley (990) and Kss/Wesche (99), and Olva (99) attempt to mantan all three assumpton at the cost of ntroducng addtonal devces. In secton we ntroduce our own approach and assume that assumpton (c) as the least motvated should be abandoned,.e., that a clause ntal nte verb s not derved as the clausal head n stu. We present an analyss whch s embedded n an approach to specers or functonal categores, such as determners, complementzers etc., whch analyses these categores as the functonal heads of the respectve consttuents. The analog to a complementzer n a V-nal clause wll be a clause-nal empty functonal head whch subcategorzes for the (ntal) nte verb.. Pollard: On ead Non-Movement Pollard (99) suggests to treat V-ntal structures n German as a more general case of the phrase structure schema for auxlary nverson n nglsh. Whereas n nglsh the schema s restrcted to apply only to phrases headed by an auxlary, t would apply n German to all nte verbs, such that the structure for V-ntal clauses wll be (). () [ [ Schenkt ] [ der Mann ] [ dem Knd ] [ en Fahrrad ] ] The poston of the nte verb s determned by a topologcal A feature INV markng nverson. All root sentences (or man clauses) are requred to be speced as [INV +]. Two LP-rules have to specfy that a verbal head marked as [INV +] has to precede all ts complements, whereas a verbal head marked [INV?] has to follow them. The obvous advantage of ths approach s that t not only provdes a soluton to the problem of nte verb poston but also gves us a relatvely smple handle on the order of complements n the \Mttelfeld" whch can be freely permuted. f. Pollard/Sag (99: 9), where the relevant phrase structure schema for nglsh s dened as \(Schema ) a saturated ([SUBAT h ]) phrase wth TRS value of sort head-comp-struc n whch the A-TR value s a lexcal sgn." In fact, n connecton wth a technque developed by nrchs/nakazawa (99), ths approach even allows the descrpton of some of the more ntrcate phenomena of partal VP-frontng. Some of the shortcomngs of ths analyss are addressed and neutralzed n Nerbonne (99). 9

3 owever, gnorng a certan number of lngustc counter-arguments, the man problem of ths approach s that t wll complcate the syntactc treatment of adjuncton and complementaton. If we want to mantan a unform at structure for the \Mttelfeld" and the assumpton of the nte verb as a head n n stu, we may be able to keep up the SP, but we cannot mantan the treatment of adjuncton as t s proposed n Pollard/Sag (99:). In Pollard/Sag (99) adjuncts are characterzed as non-head daughters whch select a head through an attrbute MO whose value s dented wth the SYNSM value of the A-TR. In the SemP the ow of semantc nformaton for head-complement structures and head-adjunct-structures () s dened such that n the latter case the ONTNT value of the mother node s not dented wth the one of the A-TR but rather wth the ONTNT value of the AJ-TR. Thus, whle the A-TR contnues to provde the syntactc head, the AJ-TR functons as the semantc head. () Phrase Structure Schema for ead Adjunct Structures h LO AT A ONT " h # LO AT A A h LO AT A MO ONT Sentences such as ()/() where several adjuncts are nterspersed wth complements present a serous problem for Pollard's approach for varous reasons. To begn wth, t wll be necessary to modfy the approach such that n a at head-complement structure the verb combnes not only wth all the complements but also wth all the adjuncts at the same tme. In a certan way ths renders the noton of a phrase structure obsolete, snce head, complements and adjuncts appear now as a lst of consttuents on a par. As a further consequence the value of the attrbute AJ-TR wll have to be a lst or set of feature structures rather than a smple feature structure. Snce each element of ths set may mpose selectonal restrctons on the head, we wll need unversal quantcaton n order to dstrbute the head daughters nformaton over all the elements. In addton, we cannot mantan the SemP anymore snce we would now have multple semantc heads, namely as many as there are adjuncts. Ths would agan make t necessary to use unversal quantcaton or a correspondng functon to collect the semantc content of all the adjuncts.. Inverson of the Subcategorzaton Prncple To avod n partcular the last problem addressed, one could go for an alternatve approach where the verbal head s combned wth only one complement or adjunct at a tme, yeldng a bnary branchng structure of a lexcal or phrasal head, and a complement or adjunct. Whle the head-adjunct rule could be mantaned as t s gven above, only one headcomplement rule s requred (). For the sake of smplcty we assume here that the SP s Ths problem s the central concern of Kasper (99) who also develops a proposal for the necessary modcatons. 0

4 reduced to a smple functon cons concatenatng the SYNSM value of the OMP-TR wth the SUBAT lst on the mother. () Phrase Structure Schema for a Bnary ead-omplement Structure LO AT ONT " A SUBAT # LO AT ONT " A SUBAT cons( ; ) # h LO The advantage of such a bnary dervaton s obvously that we can freely mx complements and any number of adjuncts wthout havng to worry about the dentty of the syntactc or semantc head at any gven tme. owever, accordng to the denton of the SP a head cannot precede a rght-branchng structure of complements (or follow a left-branchng one) snce the SP works \nsde out". The head ntroduces the subcategorzaton lst whch s then successvely reduced untl t has become an empty lst at the root of the doman of the head. Thus, t seems that the assumpton of the \Mttelfeld" as a unform rght-branchng structure and the ntal nte verb as a head n stu could only be mantaned f we nverted the SP for the partcular case of V-ntal structures, such that t works exactly the opposte to standard dervatons. In such a denton, the SUBAT value of the mother node would have to be the sum or concatenaton of the SUBAT lsts of the A-TR and the OMP-TR. Ths could be mplemented wth a lot of trcky hackng but t s easy to magne that t would also cause an enormous complcaton and prolferaton of rules and prncples just for ths one partcular case.. Non-unform Structural Representatons If we want to assume a bnary structure, due to the problem of adjuncton and mantan that the verb n ntal poston s derved as the head, the standard SP wll enforce two derent drectons of branchng for V-ntal and V-nal structures, namely left-recursve and rght-recursve respectvely (). As one eect of these assumptons standard c-command relatons wll be nverted for derent head postons. If c-command should play a role n determnng the scope of quanters or adjuncts, we end up n a stuaton where we would have to accept derent scopng relatons for the same sequence of complements and adjuncts, dependng on whether the verbal head appears n a nal or ntal poston. Smlarly, the smpled mplementaton of the SP above, whch explots the order of elements on the SUBAT for determnng the order of complements, wll now gve us reversed orderngs of complements for a left- and a rght-branchng structure. As a consequence we have to reverse the order of elements on the SUBAT-lst for the two derent An approach n categoral grammar whch goes n ths drecton s epple (990) who suggests to apply type rasng to the verbal head as well as to the complements, such that all complements can be combned wth each other to form a consttuent before they combne wth the head.

5 verb postons,.e., assume two derent lexcal entres. Furthermore, under ths account we are forced to stpulate consttuency for substrngs for whch there s no lngustc evdence whatsoever. In partcular, n a smple unmarked V{S{O structure we would have to clam that the verb rst of all forms a consttuent wth ts subject before t combnes wth an object. () Left- and Rght-Recursve Bnary Branchng h h h h A A ; ; brngt Peter morgen de Ladung Peter morgen de Ladung brngt In short, the only mleage that we can get out of a left-branchng bnary structure s that t smples our treatment of adjuncts, whereas t does not ncrease n any respect the lngustc adequacy or elegance of the descrpton.. Introducton of xtra Mechansms To resolve the obvous conct between a rght-branchng structure and the precedng verbal head several proposals have been made to \transport" the subcategorzaton frame from the \external" poston of the head to an \nternal",.e., clause nal poston, such that the SP can be appled n ts standard way. One proposal along ths lne s by Kss/Wesche (99) who apply an approach by Borsley (989) to German and who use a non-local mechansm, another suggeston s by Olva (99) who explots the ead Feature Prncple (FP) for ths purpose. Kss/Wesche's proposal nvolves a specc non-local feature \ouble Slash" SL n analogy to the standard non-local feature SLAS used to percolate dslocated W-Phrases. The nte verb n ntal poston s derved by a lexcal rule and subcategorzes for a VP carryng ths feature. Ths lexcal rule also condexes the value of the SL feature wth the LOAL attrbute of the orgnal verb such that SL also contans the orgnal subcategorzaton frame (). At the bottom of the tree SL s condexed wth the LOAL value of an empty head such that the SUBAT lst can now be reduced from rght to left. The weaknesses of ths approach are not only that t requres n addton to an empty head a lexcal rule dervng a second type of nte verb. Above all, t makes use of a mechansm devsed for non-local relatons, n order to express a dependency whch s clearly lmted to \head domans" whch could be dened through the FP. Whle genune non-local dependences n German, such as topcalzaton or dslocaton of W-elements, are not lmted by clause boundares, the relaton between a nte verb An approach along these lnes has been suggested n oeksema (98) for categoral grammar.

6 () Introducton of SL (ncludng Lexcal Rule) " LO AT h NON-LO SL # h LO AT h h NON-LO SL " V n h # LO AT h: : : n ntal poston and a correspondng empty termnal node s clearly bound to a doman whch s even smaller than the doman of a clause, f one assumes that a clause may contan several (non-nte) verbal heads n addton. As a consequence the Non-Local Prncple governng the dstrbuton and ow of non-local nformaton has to be artcally constraned for ths partcular case n order to percolate SL features only along a path of heads. At least n ths latter aspect the approach taken by Olva (99) appears to be preferable nsofar as he explots the FP drectly for establshng a relaton between the subcatorzaton of an ntal nte verb and a clause nal empty node. The structures he proposes are very smlar to (), the derence beng that not the entre LOAL value of the verb s condexed wth the SL value, but only the SUBAT value s condexed wth a A feature A-SUBAT appearng on ts complement. Ths subcategorzaton s percolated by the FP and taken o agan at the bottom of the tree where t becomes the SUBAT value of an empty node. owever, n order to acheve ths percolaton eect Olva has to splt up the FP and the SP nto several subprncples whch regulate when the A-SUBAT may be dented wth SUBAT value and when not,.e., agan some very general prncples have to be moded. Non-ead Non-Movement Before we now outlne our own approach, we shortly summarze some of the basc assumptons and generalzatons that we consder to be lngustcally relevant. { There s no evdence for assumng derent types of lexcal entres for nte verbs n a V-nal and a V-ntal poston. { A rght branchng bnary structure for the \Mttelfeld" presumably smples the descrpton of phenomena such as adjuncton, scopng and focus, and coordnaton. { The most general way to characterze word order n German s that categores whch can be marked as [V?],.e., nouns and prepostons, take a head-ntal poston, whereas those whch are marked [V +],.e., verbs and adjectves, are consstently head-nal. { A relaton between an ntal nte verb and a clause nal empty head s not a non-local relaton but should be restrcted to head domans. In order to mantan these generalzatons we wll adopt one assumpton from the movement analyss for V-ntal, namely that there s a clause nal empty node whch functons as the head of the clause. owever, to establsh the relaton between ths empty node and the nte verb we wll make use of the exstng subcategorzaton mechansm of PSG rather than ntroducng new or modfyng exstng prncples. e

7 Before we go nto the detals of our analyss we outlne some background assumptons concernng some steps n the drecton of a theory of functonal heads outlned n Netter (99). In ths paper we gve an analyss for German noun phrases based on the vew that the determner functons structurally as the A-TR of noun phrases. One of the crucal hypotheses n ths analyss s that the tradtonal A features can be splt nto two subsets, whch are organzed under two features MAJOR and MINOR. Under the MAJOR feature nformaton assocated wth \major" categores (nouns, verbs, adjectves, prepostons) would be encoded, such as some categoral features, agreement, tense and mood features, etc. The MINOR feature contans nformaton concernng \mnor" or functonal categores such as complementzers, determners, conjunctons etc. We then dene the category of maxmal phrasal projectons, such as nomnal phrases, sentences, etc., by requrng that they have to be fully saturated (.e., have an empty SUBAT lst) but that they also have to be marked as \functonally complete" (f-complete). Functonal completeness s taken to be a property whch can ether be ntroduced through a functonal category, or whch can be assocated drectly wth a lexcal category. For example, sngular count nouns are marked n the lexcon as beng f-ncomplete [FOMPL?], whch means that they rst have to combne wth a determner n order to form a maxmal phrase. Mass nouns and bare plurals on the other hand are unmarked n ths respect, such that they can ether combne wth a determner or they can project nto a maxmal nomnal phrase drectly. Functonal categores, such as determners or complementzers, are assumed be the head of the phrase and to subcategorze for a major category whch has to be marked as f-ncomplete. The MAJOR feature of the complement s \rased" and uned wth the MAJOR attrbute of the functonal category. Accordngly, an unspeced and a speced phrase wll have the same MAJOR feature, but could der n ther MINOR features. (8) Lexcal ntry for etermner (SYNSM value) " # N + MAJOR A V? h MINOR FOMPL + LO AT * SUBAT AT A MAJOR h + MINOR FOMPL? SUBAT h np det ow does the noton of f-completeness for nouns and determners carry over to nte verb constructons? To begn wth, f we compare nte wth non-nte constructons we see that nte verbs are ntrnscally f-ncomplete n a way that non-nte verbs are not. For example, whle there are plenty of constructons (9)/(0) where non-nte verbs project nto full maxmal phrase wthout any specers or structural reorgansaton, ths s never the case for nte constructons. (9) en Rasen ncht betreten (0) Saarlandscher Fnanzmnster zuruckgetreten Whenever we nd a nte verb n German nvolved n the formaton of a maxmal phrase, be t a man or a subordnate clause, t s bound ether to the presence of some lexcal functonal head (f-head) or to some structural markng of the nte verb. The former stuaton s gven n the case of V-nal clauses, where we have to have a complementzer, A very smlar approach had been suggested by Gosse Bouma (988) for the framework of categoral grammar.

8 relatve or queston pronoun whch wll mark the sentence as f-complete. In all cases where such a lexcal f-head s absent, the nte verb has to appear n a clause ntal poston. As a natural concluson, we could assume that ts ntrnsc f-ncompleteness n ths case s compensated on a structural level. We presuppose accordngly that all nte verbs are ntrnscally marked n the lexcon as f-ncomplete,.e., [FOMPL?]. 8 The dervaton of V-nal nte constructons wth a complementzer then follows the pattern nvolvng a specer or determner n nomnal constructons. The lexcal entry of a complementzer wll be dentcal to the entry for determners (0)wth the derence that the MAJOR feature of the complement s verbal rather than nomnal. Gven the assumpton that verbal heads always have to appear n a head-nal poston the clause-nal poston of the nte verb follows. owever, for the dervaton of V-ntal structures there s clearly no such explct lexcal element whch could operate as the f-head markng the constructon as f-complete. To compensate ths lack we propose an empty f-head whose lexcal structure () wll be qute smlar to lexcally overt complementzers. Ths empty node has n common wth complementzers that t subcategorzes for a f-ncomplete verbal category, however n ths case the subcategorzaton wll be for a non-phrasal, lexcal category. xactly lke complementzers the empty head wll adopt the MAJOR features from the verbal category, whle provdng tself some MINOR head feature whch marks the clause as f-complete. 9 Under ths assumpton, the clause ntal nte verb wll not functon as the A-TR of the clause anymore but smply as a complement to the f-head. owever, all of ts relevant MAJOR propertes (e.g. tense, mood etc.) wll stll be passed on through the empty head to the maxmal projecton of the clause. The same holds for the semantc ONTNT of the verb. () mpty Functonal ead " # N? MAJ A V + h MIN FOMPL + 0 AT LO A MAJ h AT MIN FOMPL? SUBAT snoc LO SUBAT B LX ONT vn ONT eps ; A What stll requres some explanaton s how the subcategorzaton requrements of the verb are ncorporated nto those of the empty head. If the verb appears as a complement the empty f-head must be subcategorzed on the one hand for the nte verb tself, but 8 Whether non-nte verb forms should be assumed to be altogether unmarked n the lexcon we want to leave open here. The constructons whch have to be nvestgated to resolve ths queston nvolve \non-nte" complementzers such as um or anstatt. Alternatvely, t could also make sense to analyse the nntve marker zu as an f-head. 9 Note by the way that the f-head n V-ntal clauses does not necessarly have to be empty. It s easy to magne an analyss for separable verb prexes such as () where the prex replaces the empty f-head subactegorsng for the nte verb stem. () Stellt er das Problem dar

9 also for all the complements that the nte verb s subcategorzed for. Ths s expressed n () by specfyng the SUBAT value of the f-head as the value of a functonal constrant snoc (the \nverse" of cons). Ths functon appends a feature structure to the end of a lst of feature structures, yeldng as a result a lst of feature structures. In ths nstance, the functon wll have as ts value a lst whch s obtaned from appendng the feature structure of the nte verb tself to the lst whch occurs as the SUBAT value of the verb. We llustrate ths functon wth the example of a transtve nte verb whch subcategorzes for two nomnal complements. If we nstantate the rst argument of snoc wth such a verb the functon representng the SUBAT value of the empty node takes the followng form: () snoc 0 vn " # N? MAJ LO AT A V + h MIN FOMPL? SUBAT np-acc [ ]; np-nom [ ] As the result of the evaluaton of the functon we obtan a lst of three elements consstng of the two arguments of the verb and the verb tself: () * np-acc [ ]; np-nom [ ]; ; A " # N? MAJ AT A V + h + MIN FOMPL? SUBAT ; vn The mportant pont about ths SUBAT lst s of course that as soon as the empty head \consumes" one of the nomnal arguments the structure of ths argument wll be automatcally uned wth the varables on the SUBAT lst of the nte verb. Ths s relevant nsofar as only the lexcal entry of the verb wll \know" how to ntegrate or relate the semantcs of ts complements. The semantc content of the nte verb s then uned n the lexcal entry of the f-head () wth ts ONTNT feature, such that t wll enter nto the semantc representaton of the clause. () h h ; ; ; ; brngt Peter de Ladung e The only other devces that we need to complete the pcture are the ndependently motvated bnary head-complement and head-adjunct rules gven above. Also, we need

10 only one lnear precedence constrant to determne the branchng drecton of the clausal structure. Ths LP constrant wll requre that a A-TR speced as [V +] has to follow a OMP-TR or an AJ-TR. Gven these two rules together wth the head-nal LP constrant, we are now able to derve all of the above descrbed V-ntal sentences wth an unmarked word order of nomnal complements and an arbtrary order and number of adjuncts. As mentoned, verb-nal clauses wll be trval to derve f we specfy the complementzer as the f-head of the clauses. As a topc for further research we have to menton that our smplstc assumpton above, that the order of nomnal complements n the \Mttelfeld" can be encoded n the lnear order of the SUBAT lst, of course cannot be mantaned n a more serous approach to ths problem. Fortunately, there are already qute a number of proposals on the table whch show how precendence constrants over complements can be expressed even f these complements are not sster to each other but occur n a herarchcal structure. 0 Nevertheless, even f t wll have to be proven by further research, we very much expect that our approach wll turn out to be sucently modular to be mantaned under these extensons. Bblography [Borsley 989] Robert Borsley: \Phrase-structure concepton and the Barrers concepton of clause structure", n: Lngustcs, 8-8 [Bouma 988] Gosse Bouma: \Moders and specers n categoral uncaton grammar", n: Lngustcs, - [ngelkamp/rbach/uszkoret 99] Judth ngelkamp, Gregor rbach, ans Uszkoret: \andlng Lnear Precedence onstrants by Uncaton", n: Proceedngs of AL 99 [epple 990] Mark epple: \Verb Movement n German and utch", n:. ngdahl, M. Reape (eds.): Parametrc Varaton n Germanc and Romance, YANA elverable R..A, dnburgh [nrchs/nakazawa 99] rhard nrchs, Tsuneko Nakazawa: \Lnearzng Fnte AUX n German omplex VPs", to appear n: [Nerbonne et al. 99 ] [oeksema 98] Jack oeksema: \Wazdat? - ontracted Forms and Verb second n utch", n: J.F. Farlund (ed.): Germanc Lngustcs, IUL, Indana, - [Kasper 99] Robert Kasper: \Adjuncts n the Mttelfeld", n: [Nerbonne et al. 99 ] [Kss/Wesche 99] Tbor Kss, Brgt Wesche: \Verb Order and ead Movement", n: O. erzog,. Rollnger (eds.): Text Understandng n LILOG, Berln, edelberg, New York, -0 [Nerbonne 99] \Partal Verb Phrases and Spurous Ambgutes", In: [Nerbonne et al. 99 ] [Nerbonne et al. 99] John Nerbonne, Klaus Netter, arl Pollard (eds.): German Grammar n PSG, to appear as: SLI Lecture Notes, hcago [Netter 99] Klaus Netter: \Morpho{Syntax of German Noun Phrases", n: [Nerbonne et al. 99 ] [Olva 99] Karel Olva: \Word Order onstrants n Bnary Branchng Syntactc Structures", LAUS 0, Unverstat Saarbrucken, omputerlngustk [Pollard 99] arl Pollard: \On ead Non{Movement", n: Proceedngs of the symposum on scontnuous onsttuency, Appendx, Tlburg [Pollard/Sag 99] arl Pollard, Ivan Sag: Agreement, Bndng and ontrol: Informaton based Syntax and Semantcs, SLI Lecture Notes, hcago 0 f. e.g., ngelkamp/rbach/uszkoret (99), Olva (99)