nternalcite[][nternalcite] 
KOMET: Grammar documentation 
Project komet 
gmdipsi 
Dolivostr.15 
6100 Darmstadt 
Germany 
First release, October 1992 
1 

TABLE OF CONTENTS 
1. Introduction 
1.1 Organization of grammar documentation 
1.2 Developmental notes 
1.3 Setting up the system 
2. The KOMET grammar 
2.1 Rank/Ranking 
2.2 Clause 
2.3 Nominal group 
2.4 Adjectival group 
2.5 Adverbial group 
2.6 Prepositional Phrase 
2.7 Verbal Group 
Bibliography 
APPENDICES 
I Graphs of regions 
II Spl sets and domain 
1 Introdction 
This document is the grammar documentation of the first release of the komet system for generation 
of German text, the methodological basis and goals of which have been defined in the komet 
department plans of gmd/ipsi. It builds on previous documentation about grammar prototypes 
[Teich1989a, Teich1990a, ?]. The komet system has been developed at gmd/ipsi, Darmstadt, 
Germany since the end of 1989. It takes the penman system [Penman Project1989] as linguistic work 
bench for the development of the grammar and the knowledge base (Upper Model and domain models 
[Bateman et al.1990, Bateman1990]) and uses the chooserinquiry interface for interaction between 
these as well as the sentence plan language (spl) [Kasper and Whitney1989] as the grammar input 
specification language. For descriptions of the overall komet architecture see [Bateman et al.1991]. 
This release comes with a version of the Upper Model we call the Merged Upper Model, which is 
a combined version of the formerly separate English and German Upper Models ([Henschel1992a]), 
with which als the English grammar, nigel, runs. 
1.1 Organization of grammar documentation 
For an introduction to systemicfunctional grammar in 
generation see [Matthiessen and Bateman1991] and The Nigel Manual of [?]. A short introduc 
tion to Systemic Funtional Linguistics (SFL) and the komet system will be given in Section 2. The 
present documentation is similar to Matthiessen's Lexicogrammatical Cartography [Matthiessen1995] 
and gives details of the coverage of the German grammar --- it is dissimilar in the sense that it is 
NOT an implementation of straight Hallidayan systemic grammar, but rather a conglomerate of the 
Fawcett direction ([Fawcett1980], IFG (Introduction to Functional Grammar) ([Halliday1985]) and 
other ideas within systemics (e.g., [Hudson1976]). The main source of description of grammatical 
2 

data is Helbig/Buscha's grammar of German ([Helbig and Buscha1988]) and the Akademiegram 
matik ([Heidolph et al.1980]). The size of the komet grammar is currently about 600 grammatical 
systems, 300 choosers and 450 inquiries. 
The komet grammar documentation is structured in the following way: The two organizing princi 
ples are rank and region. We provide a short description of every region of a rank, give at least one 
example of each, also referring and comparing to corresponding nigel regions. For the examples we 
will either provide English glosses or English translations (depending on what is more useful in a 
particular context). When we use glosses, we will indicate this by English gloss:. For an overview of 
regions in the komet grammar see Figure 1. With the examples we provide information about the 
realizational consequences of the features in a system contrastively, i.e., what happens when one is 
chosen over another, what happens if the other is chosen over the first etc. Wherever it seems to be 
of help for comprehension, we will also say something about the relation of grammatical features to 
semantic types of the Upper Model or textual relations of the text base. In addition, dependencies 
between features in the grammatical system network itself will be pointed out. The paragraphs 
headed A special case deal with some particularities of German (as opposed to English) or with 
a particular treatment of a phenomenon. In appendix I we attach graphs of the systems of every 
region. 
The version of the grammar documented here has been tested on a number of domain models; we 
include here a small domain model dealing with nontechnical medical descriptions in appendix II. 
Also in appendix II, two sets of spls are included: 
1. One set for the unfamiliar user which runs without a domain model. These are transitivity 
pattern spls; they can be used for getting familiar with the system. 
2. Another set of spls that run only with the domain model mentioned above and uses domain 
concepts. 
This grammar documentation is being revised and descriptions of newly covered areas are being 
added as grammar development proceeds. The grammar and the documentation can be referenced 
by `The komet grammar'. 
1.2 Developmental notes 
The German grammar has been written in a ``topdown'' fashion, i.e., the first systems implemented 
have been the major systems of clause rank. Moving down the rank scale, the next areas treated 
were the nominal group and the prepositional phrase. Part of these have been taken over from nigel 
in an experiment to investigate possible crosslinguistic relevance of grammar systems (across English 
and German). These have been adjusted to German. Consequently, the units of clause, nominal 
group and prepositional phrase are the best developed in the komet grammar. There is no verbal 
group yet because using a verbal group would force us to deal with longdistance dependencies, a 
topic in the area of word order, on which we have not concentrated yet in the komet project. This 
implies that statements of word order are not systematically worked out, but sufficient for the time 
being. Also, word rank and morphology rank do not exist yet. Germanspecific lexical features have 
been included in the word class hierarchy. The development of word rank and morphology rank is a 
topic for immediate future work. The lexicon is currently a fullform lexicon. The lexicon has about 
450 entries. 
1.3 Setting up the system 
Running the komet grammar requires setting up the penman environment first; details for doing 
this can be found in The Penman Reference Manual of [?]. For basic instructions about how to work 
with komet, e.g., writing spl, creating a domain model, adding to the grammar etc. we refer the 
reader to The Penman Primer of [?]. A list of files to be loaded for the komet grammar can be 
found under the `setup' directory. 
3 

ideational interpersonal textual 
rank: primary 
class: 
logical experiential (cohesion:) 
clause clause 
complexity transitivity 
circumstantial 
object 
insertion 
mood 
theme 
diathesis 
gr 
phr 
phr prep. 
ppspatio 
temporal 
ppother 
group nom 
inal 
like 
nom nominalgroup 
complexity 
classific. 
noun type 
epithet 
qualification 
deictic 
nominal 
person 
(determination) 
pronoun 
refer 
ence, 
ellipsis 
and 
substi 
tution 
adj modific. 
quality type 
adjectival 
comparison 
quant modific. 
quantity type 
adverb modific. adverbial 
verbal tense 
verb 
morphology 
complexes simplexes 
information 
unit: 
key information 
Functional Regions of the Grammar (German) 
crossclassified by metafunction and rank 
4 

2 Introduction to the komet system 
The overall domain of komet is Natural Language Processing (NLP); more precisely, its research 
topic is the generation of natural language texts from different kinds of domains for various types of 
readers. 
2.1 Theoretical and methodological basis of komet 
There are two basic tenets in the methodology underlying komet. First, natural language is con 
sidered multifunctional; second, natural language is a means of communication and interaction. 
The encoding of states of affairs and courses of events is not the only function language serves: 
it also serves to encode social relationships, subjective assessments of speakers and their affec 
tive and intentional states, foregrounding and backgrounding of information etc. This is what we 
understand by multifunctional. Considering language a means of communication and interaction 
means seeing language as always embedded in situational contexts and regarding all linguistic ex 
pression as ultimately determined by this context. These assumptions about language give us a bias 
towards working with functional linguistic theories, in particular Systemic Functional Linguistics 
(SFL) ([Halliday1978, Halliday1985]. 
In computational lingustic terms, komet is committed to a number of notions and features from 
state of the art theorizing in computational linguistics: modularity, reusability, nondirectionality and 
multilinguality. 
ffl Modularity is essential if larger scale systems are to be built up; just as does its ``mother'' 
system penman, komet seeks broad coverage of linguistic phenomena. 
ffl Reusability is vital when preparing for application in practical systems. 
ffl Supportive of these two features are declarative representations; which also result in the 
NLP system being neutral with respect to processing direction, i.e., it is potentially us 
able for analysis and generation. Work in this direction is well under way [Henschel1992b, 
Bateman et al.1992]. 
ffl The main objective of komet is the generation of German text; however, we also undertake 
contrastive linguistic studies in building up the German linguistic resources. komet is embed 
ded in a triangle of research groups (isi/usc, University of Sydney, gmd/ipsi) that deal with 
languages as different as Japanese and Chinese, on the one hand, and English and German, 
on the other hand. 
2.2 Systemic Functional Linguistics and Systemic Functional Grammar 
sfl belongs to the family of functional theories of language, which embraces a spectrum from 
Hjelsmlev ([Hjelmslev1943]) over Dik ([Dik1978]) to lfg ([Bresnan1982]). Speaking in terms of 
representation, sfl belongs to a family of grammatical frameworks that are based on formulations 
of the cooccurrrence potential of grammatical features, such as gpsg ([Gazdar et al.1985]), hpsg 
([Pollard and Sag1987]) etc. In this view, all lingustic specification is featurebased. Also, it is 
declarative and interpeted as constraints on potential feature combinations which ultimately define 
a grammatical structure ([Bateman1992]). 
There are two major notions that characterize systemic functional theory: system and function. 
System is the governing organizing principle, emphasizing the necessity to consider paradigmatic 
relations in describing language. The notion of function is manifested in the theory as function 
in context, where by context we understand context of situation. The functions of a language are 
always determined by a context of situation; and the functions in turn determine the organization 
of the linguistic resources. These functions fall into three generalized domains of meaning intrinsic 
to a language: the ideational, interpersonal and textual metafunctions. The ideational metafunction 
5 

encodes a language's propositional content. Its semantic side can be represented as what is known in 
AI (Artificial Intelligence) as taxonomies, hierarchies of concepts, such as processes, objects, qualities 
etc. Its grammatical side is reflected in the system of transitivity, which gives rise to participant 
role configurations such as actor, goal etc., the system of time and its realization in tense forms, 
the system of circumstantials, such as temporal and spatial adjuncts and others. The interpersonal 
metafunction encodes speakers' attitudes and evaluations. Major grammatical systems reflecting 
this are mood and modality. The semantic counterpart to this is what is called in AI and generation 
user and belief models. The textual metafunction encodes textual aspects of linguistic meaning, i.e., 
coherence and cohesion and the like. In the grammar it is primarily reflected in themerheme and 
givennew organization, which are realized through word order constraints and focus. The semantic 
side of the textual metafunction is a socalled text base, which can be conceived as a classification 
of primitive textual relations, such as e.g. rhetorical relations as they have been developed in rst 
([Mann and Thompson1987]). 
Orthogonally to the metafunctional organization, there is the dimension of stratification, 
reflecting the organization of the linguistic system into different degrees of abstraction 
([Halliday and Matthiessen1990]). Subject of linguistics is the stratum of form and the two ad 
jacent strata, which are semantics and phonology. Form consists of grammar and lexis; semantics in 
this view is seen as an interlevel between the context of situation and grammar. The relation between 
strata seen from semantics to phonology is one of interstratal realization. Within the grammar, there 
are two additional organizing conventions: axis and rank. Axis refers to the traditional linguistic 
distinction between paradigmatic and syntagmatic; rank refers to the constituency organization of 
syntagmatic units, e.g. for English and German these are clause, groups (nominal, verbal, adverbial 
etc) and prepositional phrase, word and morpheme. Again, the relation between axes and between 
ranks is one of realization, interaxis and interrank, respectively. Figure 2 provides a graphical display 
of the organization of systemic theory in strata and compares it to the komet architecture which is 
described in Section 2.3. 
In general, in sfl language is considered a resource for making meaning. Using language means 
making choices in this resource, where choice is governed by the situational context language is used 
in. This view strongly emphasizes the paradigmatic aspect of linguistic description: the system. All 
strata in the systemic model are thus paradigmatically represented as system networks. A system 
network is a directed acyclic graph with labeled arcs, the nodes of which are choice points, the 
systems. Each system has one or more output features, the terms, which represent minimal alter 
natives. Each alternative is only relevant in a particular functional or paradigmatic context --- this 
is encoded in the entry condition to a system. As opposed to a taxonomy, a system network allows 
crossclassification, so that simultaneous choice can be represented. This is absolutely necessary 
because systems from all the three metafunctions are simultaneously relevant. 1 
Choice in the grammatical network results in selection expressions (i.e., sets of features selected in 
one pass through the network). These are interpreted in terms of syntagmatic relations by realiza 
tion statements that for example specify the insertion of a constituent, its conflation with another 
constituent, its realization on a lower rank (preselection), ordering etc. Syntagmatic relations are 
represented in function structures, which do not specify linear ordering. It is only in the constituency 
tree structures that linear ordering is given. 
2.3 The komet architecture as an instantiation of SFL 
Our general research strategy, as motivated by the factors outlined in Sections 2.1 and 2.2, has lead 
to a specific realization of the characteristics attributed to language. The central components of the 
komet system are: 
ffl generator kernel: a systemic grammar of German [Teich1990b]; this is embedded in the 
chooser/inquiryinterface of the penman system, which establishes the interface from higher 
1 For a fairly concise introduction to systemic functional theory see [Bateman1992]. More extensive accounts of the 
theory can be found in [?, ?]. A representative description of the range of application of sfl is provided by [Butler1985]. 
sfl in generation [Matthiessen and Bateman1991, Matthiessen1989] and NLP in general [Butler1985, Teich1989b]. 
6 

Figure 1: SFL strata and komet architecture 
7 

levels (e.g. Upper Model) to the grammar. 
ffl Upper Model: the knowledge base, a version merged from the English Upper Model and the 
previous German Upper Model [Henschel1992a]. 
ffl Text base and text planner: the text base consists of a taxonomy of rhetorical relations 
(RST) [Maier and Hovy1991]. 
ffl Lexicalization component: A lexical network providing information about lexical cooccur 
rences as they can be found in Explanatory Combinatory Dictionaries (ECD) as developed in 
mtt ([Mel'cuk1988, Wanner1992]). 
Sensitivity to the situation (in the sense of context of situation) is provided by three parameters, 
the setting of which is the first step in the generation process in komet. These tree parameters are 
field, tenor and mode ([?]). Field specifies the content of the discourse; mode specifies the function 
of language in the discourse (e.g., expository, didactic etc.); tenor refers to characteristics of the 
discourse partcipants, i.e., speaker and hearer and their relationship. According to these, the text 
planner first selects concepts from the knowledge base to be expressed and then determines the 
rhetorical relations to be applied in the text to be generated. This results in a text plan which is 
the input to the lexicalization component which seeks lexical expression interpreting the concepts 
and textual information in terms of salience (for a more detailed account of generation in komet 
see [Bateman et al.1991]. The output of the lexicalization component is an spllike form that acts 
as input to the grammatical generation with the systemic grammar of German. For a display of the 
komet architecture see Figure 3. 
3 The komet grammar 
3.1 Rank/ Ranking 
The rank and ranking systems are the top systems of the grammar. They are the first to be 
entered when the grammar is accessed during the generation process. What is determined in running 
through them is the type of unit that is to be generated in that particular cycle through the grammar. 
Choosing for example the feature clause in the rank system makes available all the relevant features 
of the clause that are specified by the grammar, e.g., the regions circumstantial, transitivity, 
diathesis, mood (cf. Section 3.2). 2 
(system 
:name RANK 
:inputs START 
:outputs ((0.333 CLAUSE 
(INSERT PROCESS) 
(INSERT LEXVERB)) 
(0.333 GROUPSPHRASES) 
(0.333 WORDS)) 
:chooser RANKCHOOSER 
:region RANK 
:metafunction LOGICAL) 
The feature groupsphrases, on the other hand, is the input condition for a system that distinguishes 
between prepositionalphrase and other kinds of groups, such as nominal group, adverbial group etc. 
2 Some words about graphical conventions might be in place here: System names are given in small capitals; 
system features are given in italics and lower case; function names are given with the First Letter as Capital; region 
names are given in bold face on first occurrence in the text. Since we also use italics for general emphasize, we hope 
that the contexts are fully explanatory of the kind of emphasize we mean  general or system feature name  and 
remove potential ambiguity. 
8 

Figure 2: The komet architecture 
9 

This means that the rank and ranking systems contain features that stand for the basic constituency 
patterns we assume in systemic grammar and for which we in turn assume relatively disjunct sets 
of relevant grammatical features. 
3.2 Clause 
The top distinction on clause rank is the complexity distinction: There are simplex clauses and 
complex clauses in the Hallidayan sense (cf. [?], pp.192248). 
(system 
:name CLAUSECOMPLEXITY 
:inputs CLAUSE 
:outputs ((0.5 CLAUSESIMPLEX 
(INSERT THEME) 
(ORDERATFRONT THEME)) 
(0.5 CLAUSECOMPLEX)) 
:chooser CLAUSECOMPLEXITYCHOOSER 
:region CLAUSECOMPLEX 
:metafunction LOGICAL) 
Simplex clauses cover both independent and dependent clauses; complex clauses always consist of 
both an independent and a dependent clause. 
For independent, simplex clauses, there are five major regions on clause rank that in their real 
ization make up the constituency structure: Circumstantials, Diathesis, Mood, Theme and 
Transitivity. These will be described in sections 3.2.2, 3.2.3, 3.2.5, 3.2.8 and 3.2.9. 
3.2.1 Clausecomplex, clausecomplexity 
For types of complexes, the major types are as specified in [?] and as implemented in nigel. One 
of the types is elaboration, a subtype of expansion. 
(system 
:name EXPANSIONTYPE 
:inputs EXPANSION 
:outputs ((0.33333334 ELABORATION (INSERT ELABORATION)) 
(0.33333334 EXTENSION (INSERT EXTENSION)) 
(0.33333334 ENHANCEMENT (INSERT ENHANCEMENT))) 
:chooser EXPANSIONTYPECHOOSER 
:region CLAUSECOMPLEX 
:metafunction LOGICAL) 
Elaboration clauses are realized as relative clauses; they can either refer to a noun (thus being 
embedded in the nominal group; see Example (1)) or to a whole proposition (see Example (2)). The 
latter is referred to in the grammar of German as a particular type of `weiterfuhrender Nebensatz' 
(cf. [Helbig and Buscha1988], pp.672675). 
(1) Im Geschaftsjahr 1986/87 setzten die 40 Kasehandelsfirmen der Schweiz 
erischen Kaseunion insgesamt 79035 t Kase ab, was einem Ruckgang um 2,6 
Prozent entspricht. 
(English gloss: In the business year 1986/87 the 40 cheese trading firms of the 
Swiss Cheese Trade Union sold altogether 79035 tonnes of cheese, which means 
a recline by 2,6 percent.) 
10 

(2) Einen gunstigen Verlauf nahm der inlandische Tafelkaseabsatz, der um 3,9 
Prozent auf 22100 t stieg. 
(English gloss: A good outcome took the national cheese sales, which rose by 
3.9 percent to 22100 tonnes.) 
The difference between (1) and (2) in terms of rhetorical relations (RSTs) --- which are the basic 
textsemantic primitives the komet text planner uses --- lies in the subtype of the RSTrelation 
elaboration. In (1) the relation of domaindependentattribute is applied, in (2) the type of elabo 
ration type is interpretation. Option (1) is provided for in the grammar of the clause in a system 
elaborationofclause. Here, a constituent Event is inserted and preselected to be a dependent 
elaborating clause. In the second cycle through the grammar the structure of this Event constituent 
is built up. Dependent clauses share most of the information of independent clauses (except for 
order of the finite element). The relative pronoun in this case must be ``was'', which is classified in 
the grammar as antecedentlessrelative. ``Was'' does not carry any agreement features, it has case, 
however, which it is assigned according to the process type and according to which participant it 
realizes in the dependent clause (in example (1) ``was'' is subject and has thus nominative case). As 
mentioned above, relative clauses that refer back to a whole proposition belong to a type of linguistic 
construct that is called `weiterfuhrender Nebensatz' in grammars of German. 3 The `weiterfuhrende 
Nebensatz' is a subordinate clause syntactically, semantically it is interdependent on the main clause 
(cf. Example (3)). 
(3) Sie hat mich gestern besucht. Das hat mich sehr gefreut. --- Sie hat mich 
gestern besucht, was mich sehr gefreut hat. 
(English gloss: She visited me yesterday. That pleased me. --- She visited me 
yesterday, which pleased me.) 
Other types of `weiterfuhrender Nebensatz' are the following: 
ffl Whword plus preposition (pronominal adverbs), e.g., ``wohin'', ``wodurch'', ``womit'', which 
pronominalize adverbs and prepositional phrases. 
ffl Conjunctive clauses 
1. ``der'', e.g., Sie machte einen Versuch, der scheiterte aber restlos. (English gloss: She made a 
try, but that failed completely.) 
2. ``als'' (temporal ordering), e.g., Es war im August, da wanderte er nach Italien. (English gloss: 
It was in August, then he wandered to Italy.) 
3. ``als'' (anterior), e.g., Wir waren gerade angekommen, da regnete es auch schon. (English gloss: 
We had just arrived, then it rained already. 
4. ``wie'' , e.g., Er arbeitet fleissig, so scheint es. (English gloss: He worked diligently, so it 
seemed.) 
The generation of relative clauses embedded in nominal groups (NGs) proceeds the following way: 
A system processqualification in the NGgrammar accounts for the possibility of a relative 
clause within the NG. Here, too, a constituent Event is inserted and preselected to be a dependent, 
elaborating clause of type elaborationofthing (vs. elaborationofproposition). In the cycle through 
3 These have to be distinguished from appositions which are equally optional, can be omitted, exhibit reference iden 
tity with the antecedent, are reduced copula clauses and show case identity with the antecedent.However, appositions 
can act as attributes of nouns only. 
11 

the elaborating clause, given these preselections, the relative pronoun is determined to be of dtype 
(i.e, ``der'', ``die'', ``das''). 4 
Differences to English and nigel: Choosing for the right kind of relative pronoun in English works 
according to whether the entity referred to by the pronoun is a conscious being or not (``which'' vs. 
``who''), whereas in German there is agreement between the relative pronoun and the entity referred 
to in grammatical gender and number (but not in case). Also, the distinction between restrictive 
and nonrestrictive relative clauses does not have a formal reflex in German, whereas in English it is 
partly realized in the choice of ``that'' vs. ``which'' as the relative pronoun. 
3.2.2 Circumstantial 
Circumstantial is the functional label for adverbial. Circumstantials in German can be realized as 
adverbial groups or as prepositional phrases. The types of circumstantials are established according 
to the function they fulfill in a clause, e.g., time, space, manner etc. Since all of these are optional, 
the shape of the circumstantial network region is different from, say, the transitivity region, where 
there is not so much optionality. The circumstantial region takes the form of a number of parallel 
systems with two features, one standing for the occurrence of such a circumstantial, the other for 
the nonoccurrence in a given clause, e.g., 
(system 
:name TIMELOCATIONCIRCUMSTANCE 
:inputs OUTER 
:outputs ((0.5 TIMELOCATION (INSERT TIMELOCATIVE)) 
(0.5 NONTIMELOCATION)) 
:chooser TIMELOCATIONCIRCUMSTANCECHOOSER 
:region OUTERCIRCUMSTANCES 
:metafunction EXPERIENTIAL) 
Some types of circumstantials are realized as an adverbialgroup, others as a prepositional phrase, 
and for some both of these realizations are possible. This depends partly on the ideational type 
of circumstantial and partly on the preceding text. For example a temporal circumstantial can be 
realized as an adverbialgroup or a prepositional phrase (see examples (4) and (5)): 
(4) Gestern fuhr Helmut Kohl nach Berlin. 
(Yesterday Helmut Kohl went to Berlin.) 
(5) In der letzten Woche hielt Helmut Kohl sich in Berlin auf. 
(Last week Helmut Kohl was in Berlin.) 
This difference is reflected in a system temporallocationphoricity: 
(system 
:name TEMPORALLOCATIONPHORICITY 
:inputs TIMELOCATION 
:outputs ((0.5 PHORICTIME 
(PRESELECT TIMELOCATIVE ADVERBIALGROUP)) 
(0.5 NONPHORICTIME 
(PRESELECT TIMELOCATIVE 
PREPOSITIONALPHRASE 
LOCATIONPROCESS 
TEMPORALPROCESS))) 
:chooser TEMPORALLOCATIONPHORICITYCHOOSER 
:region CIRCUMSTANTIAL 
:metafunction EXPERIENTIAL) 
4 Choice between ``der'' and ``welcher'' is accounted for in a system, but a funtional motivation of the choice has 
not yet been formulated. 
12 

Such a phoricitysystem is provided for every type of circumstantial that can potentially be realized 
as an adverbialgroup or as a prepositionalphrase. 
Our treatment of circumstantials and participants on clause rank differs from that in nigel in that 
adverbial groups and prepositional phrases are functionally always circumstantials in nigel --- in the 
komet grammar prepositional phrases may also act as participants in transitivity structure, e.g., 
Luis Trenker stieg auf das Matterhorn. (English gloss: Luis Trenker climbed on the Matterhorn.). 
This difference is partly due to the fact that we are dealing with particularities of German and partly 
due to a slightly different view on transitivity patterns. These will be described in detail in Section 
3.2.9. 
3.2.3 Diathesis 
Diathesis refers to the way a language encodes the same situation experientially in different per 
spectives. It relates the notion of voice to such notions as ergative, middle, causative, impersonal. 
Diathesis specifies the correspondence between participant roles and syntactic functions. The region 
of diathesis consists of three systems: voice, relationchanges and referencestatus. 
(system 
:name VOICE 
:inputs CLAUSESIMPLEX 
:outputs ((0.5 ACTIVIZATION) 
(0.5 PASSIVIZATION (INSERT CHOMEUR) 
(PRESELECT CHOMEUR AGENTIVEINSTRUMENT))) 
:chooser VOICECHOOSER 
:region DIATHESIS 
:metafunction TEXTUAL) 
(system 
:name RELATIONCHANGING 
:inputs CLAUSESIMPLEX 
:outputs ((0.25 CAUSATIVIZATION) 
(0.25 DECAUSATIVIZATION) 
(0.25 DETRANSITIVIZATION) 
(0.25 NONRELATIONCHANGING)) 
:chooser RELATIONCHANGINGCHOOSER 
:region DIATHESIS 
:metafunction TEXTUAL) 
(system 
:name REFERENCESTATUS 
:inputs CLAUSESIMPLEX 
:outputs ((0.3 IMPERSONAL) 
(0.3 ARBITRARY) 
(0.3 REFERENTIAL)) 
:chooser REFERENCESTATUSCHOOSER 
:region DIATHESIS 
:metafunction TEXTUAL) 
The voice system distinguishes between activization and passivization; the relationchanging 
system distinguishes betweeen causativization e.g., Leo lasst sich die Schuhe putzen (Leo has his 
shoes cleaned.), decausativization, detransitivization and nonrelationchanging. In German, in the 
case of decausativization the verb is reflexivized, e.g., Sein Gestandnis verbesserte die Situation.  
Die Situation verbesserte sich. (English gloss: His admittance improved the situation. The situation 
improved.). The referencestatus system distinguishes between impersonal reference, e.g., Man 
13 

geht samtags gerne ins Kino. (English gloss: One likes to go to the movies on Saturdays.), arbitrary 
reference (e.g., Es handelt sich um eine gutartige Krankheit. (English gloss: It deals with a benign 
disease.)) and referential, i.e., normal reference behavior. For a sample analysis see example (6). 
Diathesis: activization vs. passivization 
(6) Wir ersteigen den Berg.  Der Berg wird erstiegen. 
We climb the mountain.  The mountain is climbed. 
Sample Analysis 
ffl activization 
Wir ersteigen den Berg. 
We climb the mountain. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Attribuant Process Location (nonexplicit) 
Subject Finite Direct Object 
ffl passivization 
Der Berg wird erstiegen. 
The mountain is climbed. 
Diathesis: passivization, nonrelationchanging, referential 
Theme (marked) 
Attribuant Process 
Subject Finite 
The grammatical difference between these two sentences lies in the value the clause feature diathesis 
takes: activization vs. passivization. This has implications for 
ffl voice in the verbal group 
ffl the conflation of the functions Theme and Subject with the transitivity participant roles (in 
the first case the Attribuant is the Theme (and Subject), in the second case the Location is 
the Theme (and Subject). 
A SPECIAL CASE: Diathesis and reflexivization 
In German there are basically two major types of reflexives with subtypes: 
1. reflexive verbs 
(a) true reflexives, e.g. sich beeilen (hurry) 
(b) reflexive constructions, e.g., sich berichtigen (correct oneself) 
2. reciprocal verbs 
(a) true reciprocals, e.g., Sie freundeten sich an. (They made friends (with one another).) 
(b) reciprocal constructions, e.g. Sie unterhalten sich. (They are talking to each other.) 
True reflexives are formally, not semantically reflexive. The reflexive pronoun is part of the lexical 
predicate and thus obligatory (e.g., *Er beeilt seine Mutter --- He hurries his mother.). With reflexive 
constructions there is real semantic coreference between the Subject and the reflexive pronoun. The 
reflexive pronoun is not obligatory, the reflexive is just another possibility for the transitive verb 
14 

and not part of the predicate (e.g., Er berichtigt sich. Er berichtigt den Schuler --- He corrects 
himself. He corrects the pupil.). True reciprocals appear mostly with a plural subject, substitution 
with einander (one another) is usually not possible (e.g., Sie freundeten sich an. *Sie freundeten 
einander an.). Reciprocal constructions, on the other hand, have alternative transitive and reflexive 
readings (e.g., Sie unterhielten sich. Er unterhielt das Publikum. Er unterhielt sich gut. --- They 
talked to each other. He entertained the audience. He had a good time.). Substitution with einander 
is usually possible in the type of Sie unterhielten sich. 
In the clause, the reflexive pronoun behaves like any object, i.e. it can change position in the clause. 
The same is true for the reflexive pronoun in reflexive constructions, the difference being that these 
can be put under contrastive stress (e.g., Sich hat er berichtigt, nicht seine Mutter (English gloss: 
Himself he has corrected, not his mother.)), whereas the reflexive pronoun of a true reflexive verb 
cannot be put under contrastive stress (e.g., *Sich beeilte er.). In all other cases of word order 
(dependent clause verbsecond, verbinitial with questions and imperatives), the reflexive pronoun 
behaves like an object: 
... dass er sich beeilt 
Beeilt er sich? 
... dass er sich berichtigt 
Berichtigt er sich? 
And the same is true for reciprocals: 
... dass sie sich anfreundeten 
Freundeten sie sich an? 
... dass sie sich unterhielten 
Unterhielten sie sich? 
Reflexive constructions can be derived by the interaction of transitivity and the diathesis option 
decausativization (see example of a gate below). For example with an agentcentered process, the 
Affected will be a reflexive pronoun, which can be preselected at this level. 
(gate 
:name ACTIVIZATIONDECAUSATIVIZATIONREFLEXIVEAFFECTED 
:inputs (AND ACTIVIZATION DECAUSATIVIZATION 
(OR IMPERSONAL REFERENTIAL) AFFECTING) 
:outputs ((1 ACTIVIZATIONDECAUSATIVIZATIONREFLEXIVEAFFECTED 
(CONFLATE AGENT SUBJECT) 
(CONFLATE DIRECTOBJECT AFFECTED) 
(CONFLATE PROCESS LEXVERB) 
(PRESELECT AFFECTED 
NOMINALGROUP 
REFLEXIVESPECIFICPRONOUN 
ACCUSATIVECASE) 
(PRESELECT VERBALGROUP ACTIVE))) 
:region DIATHESISTRANSITIVITYGATES 
:metafunction TEXTUAL) 
3.2.4 DiathesisTransitivityGates 
The region diathesistransitivitygates serves the actual conflation of syntactic functions with 
participant roles. See an example of such a gate below. 
The problem of specifying actual diathesis in gates is that the number of gates in this region explodes. 
To avoid this one possible solution is the introduction of an obliqueness hierarchy of participant 
roles as ageneralization over these roles. This is done simply by the introduction of a function 
15 

Mostinherentrole, which is the most oblique role in a ``virtual'' obliqueness hierarchy. This Most 
inherentrole can be determined for every process type. It is that participant role that is the least 
optional. For example, the Mostinherentrole for agentcentered process is the Agent role, for 
relational process it is the Attribuant role. Determining a function Mostinherentrole we obtain 
a generalization across process types that can be used in other systems of clause rank that have 
to operate with participant roles, e.g., conflation of participant roles with Theme and syntactic 
functions. The diathesistransitivity gates that otherwise render the clause grammar quite opaque 
will not be necessary any more. Instead, the conflation of participant role with syntactic function 
can be specified in the diathesis system directly. This generalized Mostinherentrole can also be 
used at other places in the clause grammar, e.g., if it turns out we want word order operations work 
on participant roles rather than, say, syntactic functions. 
(gate 
:name ACTIVIZATIONAFFECTEDCENTERED 
:inputs (AND ACTIVIZATION AFFECTEDCENTERED) 
:outputs ((1 ACTIVIZATIONAFFECTEDCENTERED 
(CONFLATE SUBJECT CAUSER) 
(CONFLATE DIRECTOBJECT AFFECTED))) 
:region DIATHESISTRANSITIVITYGATES 
:metafunction TEXTUAL) 
3.2.5 Mood 
Mood is the central system of the interpersonal metafunction at clause rank. It distinguishes between 
interaction types, such as indicative, declarative, interrogative and imperative. These are known in 
the grammar of German as `Satzmodus' (clause mode). The systems of the German mood region 
are similar to those implemented in nigel. However, the Mood element, i.e. Mood as a distinguished 
function itself consisting of Subject and Finite ([Halliday1985], pp.7178), is debatable for German. 
For English, the Mood element is that part of a clause that is arguable, i.e., that part that can be 
negated, modalized, emphasized (by `do') and that is replayed in tagquestions (e.g., He has arrived 
yesterday, hasn't he?). In German, the Mood element (if we assume one) can be replayed in the 
way English replays it, but this is a rather rare case, e.g., Er hat sich gestern ein Haus gekauft. 
--- Hat er? (as expression of surprise) (He has bought a house yesterday --- Has he?). There is no 
real equivalent in German to the English tagging in picking up the Mood element. Also, negation 
works quite differently: the locus of the negation particles is not the Finite in the Mood element, 
but rather negation can be placed right before the thing or fact negated, e.g., Ich habe kein Auto. (I 
don't have a car); Ich habe nicht Erich getroffen, sondern Leo. (I didn't meet Erich, but Leo.) The 
system of mood, in terms of the distinction in modes of interaction (indicative, imperative etc.), 
however, is the same for German as it is for English. These are reflected in 
ffl insertion of Subject and Finite in declarative clauses (when not elliptical); ordering of Subject 
before Finite in independent clauses --- ordering of Finite at end in dependent clauses, e.g., 
Sie fuhr gestern nach Berlin (She went to Berlin yesterday.) 
ffl insertion of Subject and Finite in interrogative clauses; ordering of Finite before Subject, e.g., 
Fuhr sie gestern nach Berlin? (Did she go to Berlin yesterday?) 
ffl no subject generally in imperative clauses, e.g., Bring mir ein Glas! (Bring me a glass!) 
(system 
:name MOOD 
:inputs CLAUSESIMPLEX 
:outputs ((0.5 INDICATIVE) 
(0.5 NONINDICATIVE)) 
16 

:chooser FINITECLAUSETYPECHOOSER 
:region MOOD 
:metafunction INTERPERSONAL) 
(system 
:name INDICATIVETYPE 
:inputs 
INDICATIVE 
:outputs 
((0.5 DECLARATIVE (ORDER SUBJECT FINITE)) 
(0.5 INTERROGATIVE (ORDER FINITE SUBJECT))) 
:chooser MOODTYPECHOOSER 
:region MOOD 
:metafunction INTERPERSONAL) 
(system 
:name FINITENESS 
:inputs CLAUSESIMPLEX 
:outputs ((0.5 FINITE (INSERT FINITE) (INSERT SUBJECT) 
(PRESELECT SUBJECT NOMINALGROUP NOMINATIVECASE)) 
(0.5 NONFINITE)) 
:chooser MOODCHOOSER 
:region MOOD 
:metafunction INTERPERSONAL) 
3.2.6 Objectinsertion 
Instead of having just two syntactic functions, Subject and Complement, as in nigel where they are 
part of Mood and Residue, we subspecify the type of complement further in Direct Object, Indirect 
Object, Object (for all kinds of oblique object). Also, we have the functions subject complement and 
object complement. The main purpose syntactic functions serve is as carriers of case information, 
i.e., we assign case via syntactic functions. The gates by which syntactic functions are inserted have 
very complex entry conditions with features from transitivity and diathesis. 
(gate 
:name DIRECTOBJECTINSERTION 
:inputs (AND OBJECTINSERTION 
(OR (AND ACTIVIZATION 
(OR (AND NONCENTRALIZEDLOCATION THIRDPARTYAGENCY) 
(AND CLASSIFICATORY THIRDPARTYAGENCY) 
(AND IDENTIFYING THIRDPARTYAGENCY)) 
EXISTENTIAL 
CENTRALIZEDLOCATION 
EFFECTING 
AFFECTING 
AFFECTEDCENTERED 
ASSOCIATIVE1 
PROCESSORORIENTED 
COGNITIVE 
TWOPHENOMENA 
RECEIVERORIENTED 
RECEIVERNONPLUSMESSAGE 
(AND NONMETAMESSAGE MESSAGEORIENTED))) 
17 

(AND IMPERSONAL 
(OR AGENTCENTERED 
(AND UNMARKEDTHEME ATTRIBUANTTHEME) 
(AND MARKEDTHEME LOCATIONTHEME) 
(AND MARKEDTHEME ASSOCIATEDTHEME))))) 
:outputs ((1 DIRECTOBJECTINSERTION (INSERT DIRECTOBJECT) 
(CONFLATE DIRECTOBJECT OBJECT) 
(PRESELECT DIRECTOBJECT NOMINALGROUP) 
(PRESELECT DIRECTOBJECT ACCUSATIVECASE))) 
:region OBJECTINSERTION 
:metafunction TEXTUAL) 
3.2.7 Tense 
The tense and time treatment for German is based on the Reichenbachian distinction between times: 
event time, reference time and speaking time. It is based on the time/tense treatment as it is im 
plemented in nigel. 
A descriptive account of German time and tense 
German has six grammatical tenses: present (Prasens), present perfect (Perfekt), past (Prateritum), 
past perfect (Plusquamperfekt), future I (Futur I) and future II (Futur II). Composite tenses are 
built with the auxiliaries haben (have) and sein (be), where 
haben is used with 
ffl transitive verbs (also in intransitive readings) 
ffl middle verbs, i.e. those which do not passivize (e.g., bekommen, erhalten --- get, receive) 
ffl reflexive verbs 
ffl impersonal verbs 
ffl intransitive verbs with Aktionsart of durative (e.g., sitzen, schlafen --- sit, sleep) 
sein is used with 
ffl intransitive verbs with Aktionsart of perfective (except for phase verbs, such as beginnen --- 
begin) 
ffl verbs of motion 
ffl sein and bleiben (stay) 
There are basically two exceptions to this: 
ffl When a certain view on an event (i.e., aspect) is to be expressed: in a more durative view, 
tense building is with haben; in a more perfective view it is with sein, e.g., Sie hat viel getanzt. 
 Sie ist durch den Saal getanzt. (She has danced a lot.  She has danced through the room.) 
ffl When there is a change in valency, e.g., with a nondirected action, i.e., an action without an 
Affected (or Actee) role: tense building is with sein (e.g., Er ist nach Dresden gefahren. --- 
He has gone to Dresden), with a directed action; it is with haben (e.g., Er hat einen Trabi 
gefahren. --- He has driven a Trabi.). 
18 

As described in Helbig/Buscha ([Helbig and Buscha1988], pp.137160), there is no onetoone corre 
spondence between tense and semantic time, because semantic time is not only expressed by tense, 
but also by additional lexemes (e.g., adverbials) --- which dominate over tense as indicators of the 
time expressed when they appear in a clause --- (e.g., Gestern geht er auf die Strasse und wird 
tatsachlich angefahren. --- English gloss: Yesterday he walks out into the street and is indeed hit by 
a car.) and because certain tense forms can also express modal meaning. 
There are two basic principles of tense choice: absolute and relative use of tenses. In its absolute use 
tense choice is only dependent on objective time, on speech act, on perspective of the speaker (e.g., 
Er kam nach Leipzig. Er besuchte uns. --- He came to Leipzig. He visited us.). In its relative use 
it is also dependent on other temporal events occurring in the cotext, e.g. Nachdem er in Leipzig 
angekommen war, besuchte er uns. --- After he had been in Leipzig, he visited us.. 
As pointed out above, there are three relevant temporal features: 
ffl event time (ET) 
ffl speaking time (ST) 
ffl reference time (RT) 
The tenses can be semantically described with the different relations of these three temporal 
features (ET, ST and RT), plus consideration of the modal factor (Modalfaktor [+ Mod]), the 
communicativepragmatic factor (kommunikativpragmatischer Faktor [+Colloq]) and lexical tem 
poral adverbials (lexikalische Temporalbestimmungen [+ Adv]) (cf.[Helbig and Buscha1988] pp.142 
145). Below, we give all possible uses of German tenses according to these parameters (temporal 
features and their relation, modal factor (Mod), communicativepragmatic factor (Colloq) and oc 
currence of adverbs (adv). 
Present (Prasens) 
1. Actual present (Aktuelles Prasens) 
ET = ST = RT,  Mod, + Colloq, + Adv 
The event or fact described in the present tense may have already begun in the past (e.g., Er arbeitet 
seit drei Jahren an seiner Dissertation. --- He has worked on his dissertation for three years.). If a 
modal meaning is to be expressed, a modal word has to be used (wohl (probably), sicher (surely)). 
Example: Ihre Tochter studiert jetzt in Berlin. --- Her daughter studies in Berlin now. 
2. Present denoting future 
ET = RT, ET and RT ? ST,  Mod, +Colloq, + Adv 
Is like the second variant of future I (except that the modal factor is lacking). With perfective verbs, 
the present tense has automatically future meaning (e.g., treffen (meet)). 
Example: Ich schliesse die Arbeit morgen ab. --- I will have finished with this work by tomorrow. 
3. Present denoting past 
ET = RT, RT and ET ! ST,  Mod,  Colloq, +Adv 
Here, the Adv is obligatory to make the present tense express past meaning. 
Example: Neulich treffe ich einen alten Mitschuler. --- The other day I met an old fellow student. 
4. Atemporal present (Generelles oder atemporales Prasens) 
19 

ST = RT, ET = ST and RT,  Mod, + Colloq,  Adv 
This use of the present describes general facts, which do not really have a temporal extension. 
Example: Silber ist ein Edelmetall. --- Silver is a precious metal. 
Past (Prateritum) 
For the past tense there is only one variant of usage: 
ET = RT, RT and ET ! ST,  Mod, + Colloq, + Adv 
Example: Sie arbeitete (gestern) den ganzen Tag. --- She worked all day yesterday. 
Present Perfect (Perfekt) 
1. Present Perfect denoting past 
ET = RT, RT and ET ! ST,  Mod, + Colloq, + Adv 
In this usage the perfect can be substituted by the past tense (''Vergangenheitsperfekt'') 
Example: Wir haben (gestern) den ganzen Tag gearbeitet. --- We worked all day yesterday. 
2. Present Perfect denoting a result 
RT = ST, ET ! RT and ST,  Mod, + Colloq, + Adv 
Implicitly encoded here: A result relevant for the speaking time. In this usage the perfect is not 
substitutable by the past tense and is restricted to transformative verbs (those that denote a trans 
formation to another stage (' Resultatsperfekt')). This usage corresponds to the usage of present 
perfect in English. 
Example: Peter ist vor ein paar Minuten eingeschlafen. --- Peter has fallen asleep a few minutes 
ago. 
3. Present Perfect denoting future 
ET ! RT, RT and ET ? ST,  Mod, + Colloq, + Adv 
This usage of perfect corresponds to the third variant of future II (except that there is no modal 
factor in the perfect  this can be expressed only with a modal word). 
Example: Bis zum nachsten Jahr hat er seine Dissertation abgeschlossen. --- By next year he will 
have finished his dissertation. 
Past Perfect (Plusquamperfekt) 
There is only one reading of past perfect: 
ET ! RT, RT ! ST,  Mod, + Colloq, + Adv 
This is used only when relative time is expressed (e.g., in temporal complex clauses). 
Example: Er war (gestern) bereits abgereist... --- He had already left yesterday. 
Future 1 (Futur I) 
20 

1. Future I denoting an event supposedly happening in the present 
ET = RT = ST, + Mod, + Colloq, + Adv 
With perfective verbs, there is reference to the future, not to the present. Optional adverbials of 
the type denoting present: jetzt (now), im Augenblick (at the moment) etc. 
Example: Er wird (jetzt) im Buro sein. --- He must be in the office now. 
2. Future I denoting future 
RT = ET, RT and ET ? ST, + Mod, +Colloq, + Adv 
This usage of the Future I corresponds to the second variant of the present. Adverbials that describe 
near future, such as: Morgen, nachste Woche (tomorrow, next week) etc. 
Example: Wir werden (morgen) ins Kino gehen. --- We will go to the movies tomorrow. 
There are two subvariants of this future I usage, one expressing intent and the other expressing an 
order. 
Future II (Futur II) 
1. Future II denoting an event supposedly having happened in the past 
RT = ET, RT and ET ! ST, + Mod, + Colloq, + Adv 
This usage of future II corresponds to the first variant of the perfect (difference: perfect does not 
have a modal factor; the modal factor can only be brought in by a modal word). The relation Future 
II/Perfect is the same as that of Future I/Present; this means that there is also a correspondence to 
the past tense (again without modal factor in the past tense). 
Example: Er wird (gestern) die Stadt besichtigt haben. --- He may have visited town yesterday. 
2. Future II denoting an event supposedly having happened in the past, resultative 
RT = ET, ET ! RT and ST, = Mod, + Colloq, + Adv 
Here, in analogy to the usage of the perfect, there is an implicit relevance of the result expressed 
for the speaking time; with transformative verbs. There is a correspondance in temporal meaning 
of this second variant of future II to the second variant of the perfect (again without modal factor 
in the perfect). 
Example: Er wird (schon) eingeschlafen sein. --- He may be asleep already. 
3. Future II denoting future 
ET ! RT, RT ? ST, + Mod, + Colloq, + Adv 
denoting a finished event in the future; correspondance to third variant of the perfect (Perfect:  
Mod, Future II: + Mod); obligatory adverbial in this reading. 
Example: Bald wird sie es geschafft haben. --- Soon she will have made it. 
Implementation 
In nigel, the three time features event time, reference time and speaking time and the relations 
between them are used to determine tense. All possible tenses are specified in five tense systems, 
where the choosers' inquiries establish the relations between ET, RT and ST. The modality factor 
21 

of certain tenses, the colloquiality and the occurrence of a time adverbial does not come into play in 
the Nigel computation of time. In English, the tenses have far more restricted usages as in German. 
For example future tense cannot explicitly be used to denote modality in the present. 5 Also, the 
English tense system is highly recursive, whereas the German is not. Furthermore, in nigel the 
continuous forms are seen as part of the tense system; their usage or their function, however, is 
additionally to express aspectual meaning and not only temporal meaning. 
What would have to be distinguished for German is the generation of tense (Zeitform) and the 
expression of semantic time, say, of a clause or proposition. Since, as already mentioned, there is 
no onetoone relation between time and tense, the language offers additional means to express a 
certain time. In German the pure tense is highly ambiguous. One tense has a number of usages and 
one time or one usage can be expressed by a number of tenses (and not only time, but also modality 
and colloquiality). A treatment of the cooccurrence constraints between tense, aspect, modality and 
adverbials requires modeling these first of all semantically, so that temporal reasoning is provided. 
For the time being, we have implemented these interacting areas only seperately in the komet 
grammar. 6 With regard to tense, this means that we have two tense systems: primarytense and 
secondarytense, where the secondary tense system is dependent on primary tense. As pointed 
out above, the German tense system is not recursive as the English one is: the only possible secondary 
tense type is past. 
(system 
:name PRIMARYTENSE 
:inputs TEMPORAL 
:outputs ((0.33333334 FUTURE (CLASSIFY FINITE FUTUREAUXILIARY) 
(CONFLATE TEMPO0 FINITE)) 
(0.33333334 PRESENT (CONFLATE FINITE TEMPO0) 
(CLASSIFY FINITE PRESENTFORM)) 
(0.33333334 PAST (CONFLATE FINITE TEMPO0) 
(CLASSIFY FINITE GPASTFORM))) 
:chooser PRIMARYTENSECHOOSER 
:region TENSE 
:metafunction LOGICAL 
:comments ''(FROM VGP.)'') 
(system 
:name SECONDARYTENSE 
:inputs (OR FUTURE PRESENT PAST) 
:outputs ((0.5 SECONDARY (INSERT TEMPO1) (INSERT TEMPO1DEPENDENT)) 
(0.5 NOSECONDARY)) 
:chooser SECONDARYTENSECHOOSER 
:region TENSE 
:metafunction LOGICAL) 
In all three cases of primary tense, ET precedes RT. For secondary tense we can have the following 
combinations: 
primary: present, secondary: past (pastinpresent; ET ! RT =! ST) 
primary: past, secondary: past (pastinpast; ET ! RT ! ST) 
primary: future, secondary: past (pastinfuture; ET ! RT ? ST) 
RT is established only when secondary has been chosen in the secondarytense system. 
5 ''...what would need to be demonstrated is that we choose socalled future tense for basically modal reasons rather 
than temporal reasons'' ([Matthiessen1984], p.77). This cannot be verified for English, but for German it can. 
6 This means we have to rely on semantically `correct' input from the spl. 
22 

3.2.8 Theme 
Thematic structure refers to the analysis of sentences into Theme and Rheme (see sample analysis 
below). 
Thematic structure: SubjectTheme (unmarked theme) vs. ObjectTheme (marked) 
(7) Man ersteigt den Berg. --- One climbs the mountain. 
Den Berg ersteigt man. --- English gloss: The mountain climbs one. 
Sample Analysis 
ffl Subjecttheme 
Man ersteigt den Berg. 
One climbs the mountain. 
Diathesis: activization, nonrelationchanging, impersonal 
Theme (unmarked) 
Attribuant Process Location 
Subject Finite Direct Object 
ffl Objecttheme 
Den Berg ersteigt man. 
The mountain climbs one. 
Diathesis: activization, nonrelationchanging, impersonal 
Theme (marked) 
Location Process Attribuant 
Direct Object Finite Subject 
In the first sentence the Attribuant is the Theme (in this case SubjectTheme, unmarked). In the 
second sentence the Location, conflated with the Direct Object, is the Theme. The German grammar 
currently operates with a theme notion that comes from the English grammar (subjecttheme is 
unmarked theme). This is not adequate as a definition of theme in German, which apparently can 
have other syntactic functions as unmarked theme, e.g., certain types of adjuncts. In principle, if we 
take Theme meaning the first position in the clause, not only immediate constituents of the clause 
can be in this position in German, but also parts of clause constituents, e.g., Entschuldigt hat er sich 
bei mir, aber wie ...! (English gloss: Apologized he has to me, but how ...!) Also, multiple themes 
are much more frequent and less marked in German. 
(system 
:name THEMEMARKEDNESS 
:inputs clausesimplex 
:outputs 
((0.5 UNMARKEDTHEME (CONFLATE THEME SUBJECT)) 
(0.5 MARKEDTHEME)) 
:chooser THEMEMARKEDNESSCHOOSER 
:region THEME 
:metafunction TEXTUAL) 
(system 
:name DIRECTOBJECTTHEME 
:inputs (AND MARKEDTHEME DIRECTOBJECTINSERTION) 
:outputs 
23 

((0.5 DIRECTOBJECTTHEME (CONFLATE THEME DIRECTOBJECT)) 
(0.5 DIRECTOBJECTNOTTHEME)) 
:chooser DIRECTOBJECTTHEMECHOOSER 
:region THEME 
:metafunction TEXTUAL) 
3.2.9 Transitivity 
Transitivity refers to the predicate types of a language and the participant roles they combine with. 
In komet we use a transitivity system that was designed by Steiner and is based on Fawcett's 
interpretation of systemic transitivity (cf. [Fawcett1980, Steiner et al.1988]). The least delicate 
system distinguishes between action, relational, mental and communicative processes, a contrast 
that the Hallidayan transitivity system also has. 
(system 
:name TRANSITIVITY 
:inputs CLAUSESIMPLEX 
:outputs ((0.25 RELATIONAL) 
(0.25 MENTAL) 
(0.25 COMMUNICATIVE) 
(0.25 ACTION)) 
:chooser TRANSITIVITYCHOOSER 
:region TRANSITIVITY 
:metafunction EXPERIENTIAL) 
The actiontype system has five subtypes: agentonly, agentcentered, affectedcentered, affected 
only and naturalphenomenon. The labels are largely selfexplanatory. The difference between 
agentcentered and affectedcentered mainly lies in the difference in mutual constraints with diathe 
sis: Affectedcentered can potentially undergo the decausativization alternation, agentcentered can 
undergo the detransitivization alternation (cf. Section 3.2.3). 7 
(system 
:name ACTIONTYPE 
:inputs ACTION 
:outputs ((0.2 AGENTONLY) 
(0.2 AFFECTEDONLY (INSERT AFFECTED) 
(PRESELECT AFFECTED NOMINALGROUP)) 
(0.2 AFFECTEDCENTERED (INSERT CAUSER) 
(PRESELECT CAUSER NOMINALGROUP) (INSERT AFFECTED) 
(PRESELECT AFFECTED NOMINALGROUP)) 
(0.2 AGENTCENTERED) 
(0.2 NATURALPHENOMENON 
(INSERT DUMMYAGENT) 
(CONFLATE DUMMYAGENT SUBJECT) 
(CLASSIFY DUMMYAGENT PRONOUN EMPTYSUBJECTES)))) 
:chooser ACTIONTYPECHOOSER 
:region TRANSITIVITY 
:metafunction EXPERIENTIAL) 
Transitivity --- Action Process, agentcentered vs. affectedcentered 
7 The alternation effectivemiddle (i.e., detransitivization) is not as pervasive in German as it is in English. However, 
it is possible and realized as reflexivization. A more productive pattern in German is that of decausativization (i.e., 
CauserAffected --- Affected) which we account for in the diathesis region (cf. Section 3.2.3). 
24 

(8) Luis leitet die Expedition. --- Luis leads the expedition. 
Hans bewegt das Pferd. --- Hans moves the horse. 
Sample Analysis: 
ffl agentcentered 
Luis leitet die Expedition. 
Luis leads the expedition. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Agent Process Affected 
Subject Finite Direct Object 
ffl affectedcentered 
Hans bewegt die Pferde. 
Hans moves the horses. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Causer Process Affected 
Subject Finite Direct Object 
The relationaltype system has six subtypes: existential, locational, classificatory, associative, 
identifying and changing. 
(system 
:name RELATIONALTYPE 
:inputs RELATIONAL 
:outputs ((0.17 EXISTENTIAL 
(INSERT EXISTENCEMARKER) 
(CONFLATE EXISTENCEMARKER SUBJECT) 
(CLASSIFY EXISTENCEMARKER PRONOUN EMPTYSUBJECTES) 
(INSERT ATTRIBUANT) 
(PRESELECT ATTRIBUANT NOMINALGROUP) 
(CLASSIFY PROCESS EXISTENTIAL)) 
(0.17 LOCATIONAL (INSERT ATTRIBUANT) 
(PRESELECT ATTRIBUANT GROUPSPHRASES) 
(CLASSIFY PROCESS LOCATIONAL)) 
(0.17 IDENTIFYING (INSERT ATTRIBUANT) 
(PRESELECT ATTRIBUANT NOMINALGROUP) 
(INSERT PIDENTIFIER) 
(PRESELECT PIDENTIFIER GROUPSPHRASES) 
(CLASSIFY PROCESS IDENTIFYING)) 
(0.17 CLASSIFICATORY (INSERT ATTRIBUANT) 
(PRESELECT ATTRIBUANT NOMINALGROUP)) 
(0.17 ASSOCIATIVE (INSERT ATTRIBUANT) 
(INSERT ASSOCIATED)) 
(0.17 CHANGING (INSERT ATTRIBUANT) 
(PRESELECT ATTRIBUANT NOMINALGROUP))) 
:chooser RELATIONALCHOOSER 
:region TRANSITIVITY 
:metafunction EXPERIENTIAL) 
Here, we will only describe two relational types in some detail: locational and changing. The changing 
subtype is rather domainspecific and was introduced to cover constructions of the type: 
25 

(9) Der Absatz steigt. 
(The sales rise.) 
(10) Der Absatz steigt auf 10000 t. 
(The sales rise to 10000 tonnes.) 
(11) Der Absatz steigt von 5000 auf 10000t. 
(The sales rise from 5000 to 10000 tonnes.) 
(12) Der Absatz steigt um 50 Prozent von 5000 t auf 10000 t. 
(The sales rise by 50 percent from 5000 to 10000 tonnes.) 
(13) Der Absatz steigt um 50 Prozent. 
(The sales rise by 50 percent.) 
The maximal frame of verbs of this type is as in (12). Generally, all of these roles are optional. The 
only variant that does not seem to be possible is (14): 
(14) ? Der Absatz steigt von 5000 t. 
( ? The sales rise from 5000 tonnes.) 
To account for the roles that could be realized, there are three more delicate systems sourcespecified, 
differencespecified and resultspecified. Since this process type is rather domaindependent we in 
cluded it as a concept only in the domain model as changingprocess with three subtypes. 
In general, for a number of process types in German, there is a tension between action and relation 
interpretation of a process, so that a clause: Wir steigen auf den Berg. (English gloss: We climb 
on the mountain.) can be interpreted as action (Agent, LocationCircumstantial) or as relational 
(Attribuant, LocationParticipant). Arguments for preferring the latter interpretation: 
ffl relative obligatoriness of the Location. 
ffl the direct opposition to clauses such as Wir ersteigen den Berg. or Wir besteigen den Berg. 
(We climb the mountain), where the Location role is strictly obligatory. In German, this is a 
way of expressing a locational relation where the relation is partly encoded in the prefix of the 
verb. 
Thus, we generally prefer the relational interpretation; see the following sample analysis for illustra 
tion: 
Transitivity: Relational Process, noncentralized vs. centralized location 
(15) Wir steigen auf den Berg. --- We climb onto the mountain. 
Wir ersteigen den Berg. --- We climb the mountain. 
Sample Analysis: 
ffl noncentralizedlocation 
Wir steigen auf den Berg. 
We climb onto the mountain. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Attribuant Process Location (explicit) 
Subject Finite Prepositional Object 
ffl centralizedlocation 
26 

Wir ersteigen den Berg. 
We climb the mountain. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Attribuant Process Location (nonexplicit) 
Subject Finite Direct Object 
The function structure of both of these sentences is Attribuant Process Location. In the first sentence 
the locational relation is said to be expressed explicitly, which is formally reflected in the realization 
of the role Location as a Prepositional Object. In the second sentence the locational relation is said 
to be implicitly expressed, which is formally reflected in two ways: The relation can be seen as being 
encoded in the prefix er of the verb ersteigen and in the realization the Location takes, i.e., being 
realized as Direct Object in accusative case. This is in fact a very productive pattern in German, 
going across process types where a relation is involved, e.g., ersteigen vs. steigen auf, erklettern vs. 
klettern auf, erarbeiten vs. arbeiten an, not just for the locational subtype. 
Another feature that distinguishes action and relational processes is that relational processes can 
potentially have a Thirdpartyagent, i.e., an additional participant who brings the action about, e.g. 
(16) Wir steigen auf das Matterhorn. 
(We climb on the Matterhorn.) 
(17) Luis Trenker geleitet uns auf das Matterhorn. 
(L.T. leads us onto the Matterhorn.) 
(18) Das Buch liegt auf dem Tisch. 
(The book is lying on the table.) 
(19) Peter legt das Buch auf den Tisch. 
(Peter puts the book on the table.) 
In (17) `Luis Trenker' is Thirdpartyagent; in (19) `Peter' is Thirdpartyagent. The lexical verb in 
these examples is different according to whether there is a Thirdpartyagent or not, but the pattern 
is general for all subtypes of relational process. 
The mentaltype system has five subtypes: processororiented, processoronly, phenomenon 
oriented, twophenomena, phenomenononly with the roles of Processor and Phenomenon. The 
difference between processororiented and phenomenonoriented lies in the position the roles have 
in the clause: With phenomenonoriented the Phenomenon takes the Subject position in an active 
clause with unmarked Theme), with processororiented the Processor takes the Subject position in 
an active clause with unmarked Theme. Semantically, we represent the difference as one between a 
mentalactive and a mentalinactive process. 
(system 
:name MENTALTYPE 
:inputs MENTAL 
:outputs ((0.25 PROCESSORORIENTED (INSERT PROCESSOR) 
(PRESELECT PROCESSOR NOMINALGROUP) 
(INSERT PHENOMENON) 
(PRESELECT PHENOMENON NOMINALGROUP) 
(CLASSIFY PROCESS MENTAL)) 
(0.25 PROCESSORONLY (INSERT PROCESSOR) 
(PRESELECT PROCESSOR NOMINALGROUP) 
(CLASSIFY PROCESS MENTAL)) 
(0.25 PHENOMENONORIENTED (INSERT PHENOMENON) 
(INSERT PROCESSOR) (CLASSIFY PROCESS MENTAL)) 
(0.25 TWOPHENOMENA (INSERT PHENOMENON1) 
(PRESELECT PHENOMENON1 NOMINALGROUP) 
27 

(INSERT PHENOMENON2) 
(PRESELECT PHENOMENON2 NOMINALGROUP) 
(CLASSIFY PROCESS MENTAL)) 
(0.25 PHENOMENONONLY (INSERT PHENOMENON) 
(CLASSIFY PROCESS MENTAL))) 
:chooser MENTALCHOOSER 
:region TRANSITIVITY 
:metafunction EXPERIENTIAL) 
The Phenomenon can either be realized as a nominalgroup or verbally, as a clause or as an infinitive. 
The latter we call a metaphenomenon in the Upper Model. 
Transitivity: Mental Process, phenomenonoriented vs. processororiented 
(20) Das Buch gefallt Hans. --- The book pleases Hans. 
Hans mag das Buch. --- Hans likes the book. 
Sample Analysis: 
ffl phenomenonoriented 
Das Buch gefallt Hans. 
The book pleases Hans. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Phenomenon Process Processor 
Subject Finite Indirect Object 
ffl processororiented 
Hans mag das Buch. 
Hans likes the book. 
Diathesis: activization, nonrelationchanging, referential 
Theme (unmarked) 
Processor Process Phenomenon 
Subject Finite Direct Object 
Communication processes can be quite similar to mental processes in structure. For example, the 
Phenomenon in mental processes and the Message in communication processes can be realized by a 
thatclause or a nonfinite depdendent clause, e.g., Communication: Er sagte, dass ihm die Ausstel 
lung gefallen hat. (He said that he liked the exhibition.), Mental: Er glaubt, dass seine Geldborse 
gestohlen wurde. (He believes that his wallet has been stolen.). 
A communication process can often be considered an externalized mental process, and a mental 
process an internalized mental process. The roles corresponding to Processor and Phenomenon of 
mental processes are Sender and Message for communication processes. There is no corresponding 
role to Receiver with mental process types, because mental processes do generally not come out 
into the open, so there is no addressee. Communication processes are subdivided into message 
oriented, where there is potentially a Message, receiveroriented, where there is at least a Sender 
and a Receiver, and messageonly, where there is only a Message. 
(system 
:name COMMUNICATIVE 
:inputs COMMUNICATIVE 
:outputs ((0.25 MESSAGEORIENTED (INSERT SENDER) 
(PRESELECT SENDER NOMINALGROUP) 
28 

(CLASSIFY PROCESS COMMUNICATION)) 
(0.25 RECEIVERORIENTED (INSERT SENDER) 
(PRESELECT SENDER NOMINALGROUP) (INSERT RECEIVER) 
(PRESELECT RECEIVER NOMINALGROUP) 
(CLASSIFY PROCESS COMMUNICATION)) 
(0.25 MESSAGEONLY (INSERT MESSAGE) 
(CLASSIFY PROCESS COMMUNICATION))) 
:chooser COMMUNICATIVECHOOSER 
:region TRANSITIVITY 
:metafunction EXPERIENTIAL) 
There is an additional role Prompt which stands with utterances in reported speech; this role can 
optionally occur with all subtypes of communication processes. 
(21) Der Schuler entgegnete (PROMPT), dass er krank gewesen sei. 
(The pupil replied that he had been sick.) 
(22) Der Schuler entgegnete auf die Frage (+PROMPT), dass er krank gewesen sei. 
(The pupil replied to the question that he had been sick.) 
(23) Der Schuler entgegnete dem Lehrer auf die Frage (+PROMPT), dass er krank 
gewesen sei. 
(The pupil replied to the teacher's question that he had been sick.) 
A SPECIAL CASE: empty subjects 
Looking at the action and relation process types again, it can be seen from the realization statements 
in the network that there are two features that are realized by an empty subject `es' (`it'). The 
action/naturalphenomenon accounts for sentences such as e.g., Es regnet. (It is raining.). These 
always have an `es'subject without a semantic role, so there is no participant at all in this type of 
clause. A different case is the type relational/existential, e.g., Es gibt einen guten Kase in Frankreich. 
(There is good cheese in France.), where there always has to be one participant (the Attribuant). 
This type of empty subject is licensed by transitivity type and must be distinguished from `es' 
as a means of focus, in e.g., Es fahren taglich Millionen von Autos auf Deutschlands Autobahnen. 
(English gloss: It drive millions of cars daily on Germany's highways.). 
NOTE: A set of spls with transitivity patterns is given in appendix II. 
3.3 Nominal group 
3.3.1 Nominalgroupcomplexity 
Nominal group complexity is described in two systems that distinguish between extending and elab 
orating nominal group complexes. Extending complexes can either be additive (`und' (`and')) or 
alternating (`oder' (`or')), elaborating complexes can either be exemplifying (`z.B.' (`e.g.')) or re 
stating (`d.h.' (`i.e.')). 
(system 
:name EXTENDINGNOMINALGROUPCOMPLEXITY 
:inputs 
NONWHNOMINAL 
:outputs 
((0.5 EXTENDINGNOMINALGROUPCOMPLEX 
(INSERT INITIATING) 
(PRESELECT INITIATING NONWHNOMINAL) 
(INSERT CONTINUING) 
29 

(PRESELECT CONTINUING NONWHNOMINAL) 
(INSERT EXTENDER) 
(ORDER INITIATING EXTENDER) 
(ORDER EXTENDER CONTINUING)) 
(0.5 NONEXTENDINGNOMINALGROUPCOMPLEX)) 
:chooser EXTENDINGNOMINALGROUPCOMPLEXITYCHOOSER 
:region NOMINALGROUPCOMPLEXITY 
:metafunction LOGICAL) 
(system 
:name ELABORATINGNOMINALGROUPCOMPLEXITY 
:inputs 
NONEXTENDINGNOMINALGROUPCOMPLEX 
:outputs 
((0.5 ELABORATINGNOMINALGROUPCOMPLEX 
(INSERT INITIATING) 
(PRESELECT INITIATING NONWHNOMINAL) 
(INSERT CONTINUING) 
(PRESELECT CONTINUING NONWHNOMINAL)) 
(0.5 NONELABORATINGNOMINALGROUPCOMPLEX)) 
:chooser ELABORATINGNOMINALGROUPCOMPLEXITYCHOOSER 
:region NOMINALGROUPCOMPLEXITY 
:metafunction LOGICAL) 
3.3.2 Classification 
Modification of the noun in the nominal group is implemented as specified in [Halliday1985] (pp.160 
167): epithet, classifier, qualifier. Accordingly, there are three regions classification, epithet and 
qualification. With modification by a Classifier, the Thing (noun) is classified as being part of a 
class or set (e.g., elektronische Datenverarbeitung  electronic data processing). 
(system 
:name PRIMARYCLASSIFICATION 
:inputs NOMINAL 
:outputs ((0.5 PRIMARYCLASSIFICATION (INSERT CLASSIFIER1) 
(CLASSIFY CLASSIFIER1 ADJECTIVE)) 
(0.5 PRIMARYNONCLASSIFICATION)) 
:chooser PRIMARYCLASSIFICATIONCHOOSER 
:region CLASSIFICATION 
:metafunction IDEATIONAL) 
3.3.3 Epithet 
With modification by an Epithet the Thing is attributed a quality, which is usually gradable (e.g., 
alte Manner  old men). Typical examples for epithets are size, colour, age. For an example system 
see below. 
(system 
:name AGEMODIFICATION 
:inputs NOMINAL 
:outputs ((0.2 AGEMODIFIED (INSERT AGE) (CLASSIFY AGE AGE) 
(CLASSIFY AGE ADJECTIVE)) 
(0.8 NOTAGEMODIFIED)) 
:chooser AGEMODIFICATIONCHOOSER 
30 

:region EPITHET 
:metafunction IDEATIONAL) 
3.3.4 Qualification 
With modification by Qualifier, further (restrictive) information is given on the thing. Qualifiers are 
posited after the Thing (e.g., das Haus im Wald  the house in the woods). In principle all the types 
of circumstances we find at clause rank can also act as Qualifiers at nominal group rank. See the 
example system below. 
(system 
:name PLACEQUALIFICATION 
:inputs NOMINAL 
:outputs ((0.1 LOCATIONQUALIFIED (INSERT LOCATIVEQUAL) 
(PRESELECT LOCATIVEQUAL 
LOCATIONPROCESS 
SPATIALPROCESS)) 
(0.9 NOTLOCATIONQUALIFIED)) 
:chooser PLACEQUALIFICATIONCHOOSER 
:region QUALIFICATION 
:metafunction IDEATIONAL) 
3.3.5 Countnumber 
The fact whether a noun is countable or a mass noun, is languagespecific and depends on a language's 
conceptualization of entities, i.e., it is dependent on the semantic system of entities. We can only 
give here the grammatical possibilities for German nouns, which are: 
ffl systemic feature: countablenoun, e.g., Haus, Baum, Film (house, tree, film) 
ffl systemic feature: uncountablenoun, e.g., Sand, Papier (sand, paper) 
ffl systemic features: pluralthing (nonsingular, plural), e.g., seine Papiere (his documents) 
ffl systemic features: singularthing (nonplural, singular), e.g., der Mond (the moon) 
(gate 
:name COUNTABLENOUN 
:inputs (AND (OR SINGULAR PLURAL) 
LEXICALTHING) 
:outputs ((1.0 COUNTABLENOUN (CLASSIFY THING COUNTABLE))) 
:region COUNTNUMBER 
:metafunction IDEATIONAL) 
(gate 
:name UNCOUNTABLENOUN 
:inputs (AND NONSINGULAR 
NONPLURAL 
LEXICALTHING) 
:outputs ((1.0 UNCOUNTABLENOUN (CLASSIFY THING UNCOUNTABLE) 
(INFLECTIFY THING STEM))) 
:region COUNTNUMBER 
:metafunction IDEATIONAL) 
(gate 
31 

:name PLURALTHING 
:inputs (AND PLURAL LEXICALTHING) 
:outputs ((1.0 PLURALTHING)) 
:region COUNTNUMBER 
:metafunction IDEATIONAL) 
(gate 
:name SINGULARTHING 
:inputs (AND SINGULAR LEXICALTHING) 
:outputs ((1.0 SINGULARTHING)) 
:region COUNTNUMBER 
:metafunction IDEATIONAL) 
3.3.6 Deictic 
The deictic region specifies agreement in gender and number and case concord between the noun 
and its dependents. In English, the nominal group does not exhibit agreement, so this area is com 
pletely Germanspecific. 
A SPECIAL CASE: Agreement in the nominal group 
For illustration of these agreement phenomena let us consider the following examples: 
singular vs plural 
(24) der schone Wald --- die schonen Walder 
(the beautiful forest --- the beautiful forests) 
masculine vs feminine vs neuter 
(25) ein schoner Wald --- eine schone Frau 
(a beautiful forest --- a beautiful woman) 
nominative vs genitive vs dative vs. accusative case 
(26) der schone Wald --- des schonen Waldes --- dem schonen Wald --- den schonen 
Wald 
identifiable vs notidentifiable 
(27) der schone Wald --- ein schoner Wald 
(the beautiful forest --- a beautiful forest) 
Number agreement 
In terms of implementation, one possible solution, e.g., with respect to number agreement would be 
to have multiplicityq (i.e., the inquiry that asks about the number of the concept to be expressed) 
posed to the semantics for each single element in the nominal group independently of each other. 
One could argue that number agreement in the nominal group is in fact a semantic phenomenon and 
that simply for all its elements choices with regard to a set of the same features have to be made 
because the referential object is the same. 
However, there are some counter arguments: 
ffl the Deictic for example does not have a corresponding ideational concept, its number cannot 
be inquired via semantic type. 
ffl one would lose the information that there is a grammatical relation between elements of one 
unit, i.e., that there is a morphological dependency (see example (24)). 
32 

Number agreement must be specified in the grammar itself. We describe number agreement between 
noun and determiner (Thing and Deictic) by gates that have as input conditions the number value 
determined for the noun and that systemic feature that results in the insertion of a determiner. 
With the output features we specify that the determiner has the same number value, e.g., if plural 
is one of the input conditions we write a realization statement about the determiner, e.g., classify 
Deictic gpl (where `gpl' stands for plural). See for an example the gate below. 8 
(gate 
:name DEICTICGENCASEPLURALSPECIFICATION 
:inputs (AND (OR DEMONSTRATIVESPECIFICPRONOUN 
NOMINALSPECIFIC 
DEICTICINDEFCLASSIFICATION) 
GENITIVECASE 
PLURAL) 
:outputs ((1.0 GENITIVEDEICTICPLURAL 
(CLASSIFY DEICTIC GGEN GPL))) 
:region PRONOUN 
:metafunction INTERPERSONAL) 
Gender agreement 
Gender agreement is a different case in that grammatical gender is an idiosyncratic feature of nouns 
as lexemes. The noun being (usually) the head of the NG, the dependents (determiner, adjective) 
reflect the gender agreement with the head in their morphological shape. 
The gender information is inquired about the nominalgroup head (i.e., the Thing) from the lexicon by 
two inquiries (genderneuterq and genderfeminineq) that are asked by the chooser of the lexical 
gender system: 
(system 
:name LEXICALGENDER 
:inputs (OR 
NOMINALTERMRESOLUTION 
RELATIVEWHICH 
RELATIVETHAT) 
:outputs ((0.33333 MASCULINEGENDER (CLASSIFY THING GMASC)) 
(0.33333 FEMININEGENDER (CLASSIFY THING GFEM)) 
(0.33333 NEUTERGENDER (CLASSIFY THING GNEUT))) 
:chooser LEXICALGENDERCHOOSER 
:region DEICTIC 
:metafunction AGREEMENT) 
Licensing this kind of agreement from the semantics is not possible because we are not dealing with 
natural, but grammatical gender. The actual agreement is then specified in the system deictic 
lexicalgender, the input condition of which is that a deictic be inserted for the deicticlexical 
gender system to be entered at all (this is encoded in the complex features as e.g., nominative 
deicticsingular). The parallel realization for Thing and Deictic (classify Thing gmasc, classify 
Deictic gmasc) then specify the actual agreement in gender between Thing and Deictic. 
(system 
:name DEICTICLEXICALGENDER 
8 Number agreement is specified in one and the same gate together with case concord, which will be described 
shortly. This is not on linguistic grounds, but just for reasons of redundancy reduction. 
33 

:inputs (AND NONEXTENDINGNOMINALGROUPCOMPLEX 
(OR NOMINATIVEDEICTICSINGULAR 
ACCUSATIVEDEICTICSINGULAR 
DATIVEDEICTICSINGULAR 
GENITIVEDEICTICSINGULAR 
NOMINATIVEDEICTICPLURAL 
ACCUSATIVEDEICTICPLURAL 
DATIVEDEICTICPLURAL 
GENITIVEDEICTICPLURAL) 
(OR NOMINALSPECIFIC DEICTICINDEFCLASSIFICATION)) 
:outputs ((0.33333 DEICTICTHINGMASCULINEGENDER 
(CLASSIFY DEICTIC GMASC)) 
(0.33333 DEICTICTHINGFEMININEGENDER 
(CLASSIFY DEICTIC GFEM)) 
(0.33333 DEICTICTHINGNEUTERGENDER 
(CLASSIFY DEICTIC GNEUT))) 
:chooser DEICTICLEXICALGENDERCHOOSER 
:region DEICTIC 
:metafunction AGREEMENT) 
Gender agreement between adjective (Classifier, Epithet) and noun (Thing) must then be described 
in a separate system, with those features as input condition that result in the insertion of an adjec 
tive. Furthermore, this has to be done separately for Classifier and Epithet, because they can occur 
together in the nominal group. A big problem here is how to deal with recursion, since there can be 
for example more than one classifier. 
Case concord 
Case is assigned by the verb to its dependent elements. In the komet grammar case is assigned 
to participants (and circumstantials) on clause rank (or PP rank) with preselection on the NG of 
the case a verb or a preposition governs. However, only the Thing (which we implicitly treat as 
head of the NG) inherits the case information. We are still left with the problem of having the case 
shared by the dependents. Whereas the case can be assigned to the lexical head already on clause 
rank (actually, it is assigned to the phrasal node), this cannot be done for its dependents because 
the information about which constituents the nominal group will have is not yet existent at clause 
rank. Again, one would need to express case information as being dependent on the lexical head, or, 
be able to state that whatever constituents the nominal group will have, these will share the same 
value of the case feature. The latter is representationally currently not possible, so we account for 
case concord in gates as for example the following (including number agreement): 
(gate 
:name DEICTICGENCASEPLURALSPECIFICATION 
:inputs (AND (OR DEMONSTRATIVESPECIFICPRONOUN 
NOMINALSPECIFIC 
DEICTICINDEFCLASSIFICATION) 
GENITIVECASE 
PLURAL) 
:outputs ((1.0 GENITIVEDEICTICPLURAL 
(CLASSIFY DEICTIC GGEN GPL))) 
:region PRONOUN 
:metafunction INTERPERSONAL) 
These gates have the following input conditions: 
34 

ffl features from systems the realization of which result in the insertion of some kind of Deictic 
(determiners, demonstratives, possessives) 
ffl the case preselected from clause or prepositional phrase rank 
ffl the number determined as feature of the Thing 
Agreement Determineradjectivenoun 
Yet another case is the above mentioned agreement between determiner, adjective and noun, where 
one would like to express that depending on the identifiability of the concept to be expressed, the 
morphology of the adjective will be different, i.e., declension will be weak or strong (see example 
(27)). This can also be done in gates to enforce a particular morphological realization of the adjective. 
There is the same problem here again with recursion. See an example of such a gate for a Status 
adjective: 
(gate 
:name DEICTICGENCASEPLURALSPECIFICATION 
:inputs (AND STATUSMODIFIED 
GENITIVECASE 
SINGULAR 
NOMINALSPECIFIC) 
:outputs ((1.0 STATUSGENITIVEPLURALIDENTIFIABLE 
(PRESELECT STATUS ADJGENITIVE ADJPLURAL ADJSTRONG))) 
:region PRONOUN 
:metafunction INTERPERSONAL) 
Since all adjectival modification of a noun is generated in a separate cycle (the cycle through the 
adjectival group, which is embedded in the NG), we have to pass this information by preselection 
to the adjectivalgroup (see realization statement in sample system: adjgenitive for case, adjplural 
for number, adjstrong for declension type). 
3.3.7 Nountype and Processualthingtype 
The region nountype covers all kinds of information about the head of the nominal group, e.g., 
whether it is a specific instantiation or not (which has some impact on determiner selection, e.g., 
das Haus vs. ein Haus --- the house vs. a house), whether it is a pronoun or a full noun (systems 
specificnominality and nonspecificnominality), whether it denotes a class or is an individual 
name (e.g., Haus vs. John) in the individuation system (which, again, has an impact on determiner 
(non)selection), whether it is processual or not (i.e., deverbal or not) in the system Thingtype 
(e.g., Absatz (sales)) etc. See some of the systems of this region below. 
(system 
:name SPECIFICNOMINALITY 
:inputs (or (AND NONMINIRANGEOFMINORPROCESS 
SPECIFICINSTANTIATION 
PLACEDETERMINED 
(OR NONQUANTIFIED 
RECOVERABLEQUANTITY 
PROPORTIONQUANTITY 
CARDINALQUANTITY)) 
(AND 
NONMINIRANGEOFMINORPROCESS 
SPECIFICINSTANTIATION 
35 

CLASSNAME 
(OR 
NONQUANTIFIED 
RECOVERABLEQUANTITY 
PROPORTIONQUANTITY 
CARDINALQUANTITY))) 
:outputs ((0.5 NOMINALSPECIFIC (INSERT DEICTIC) 
(CLASSIFY DEICTIC DEFART)) 
(0.5 PRONOMINALSPECIFIC)) 
:chooser SPECIFICNOMINALITYCHOOSER 
:region NOUNTYPE 
:metafunction IDEATIONAL) 
(system 
:name NONSPECIFICNOMINALITY 
:inputs NONSPECIFICINSTANTIATION 
:outputs ((0.5 NOMINALNONSPECIFIC) (0.5 PRONOMINALNONSPECIFIC)) 
:chooser NONSPECIFICNOMINALITYCHOOSER 
:region NOUNTYPE 
:metafunction IDEATIONAL) 
(system 
:name CLASSNAMETYPE 
:inputs (OR NONPROCESSUAL 
PRONOMINALNONSPECIFIC 
NONWHNOMINAL) 
:outputs ((0.5 PERSON) (0.5 NONPERSON)) 
:chooser CLASSNAMETYPECHOOSER 
:region NOUNTYPE 
:metafunction IDEATIONAL) 
(system 
:name INDIVIDUATION 
:inputs NONINTERACTANT 
:outputs ((0.5 INDIVIDUALNAME) (0.5 CLASSNAME)) 
:chooser INDIVIDUATIONCHOOSER 
:region NOUNTYPE 
:metafunction IDEATIONAL) 
The region processualthingtype contains frame information of (deverbal) nouns. The transitiv 
ity systems of the clause are duplicated in the grammar of the nominal group. Unfortunately, they 
cannot be shared across clause and nominal group directly. This is mainly because the realization 
statements for nominal transitivity are different and the roles which are obligatory in clause transi 
tivity are only optional for nominal transitivity. The only part that can physically be shared are the 
inquiries, so that the `sameness' of noun transitivity and clause transitivity is reflected semantically 
in the fact that the same Upper Model concepts are used. 
(system 
:name THINGTYPE 
:inputs NOMINAL 
:outputs ((0.5 NONPROCESSUAL) 
(0.5 PROCESSUAL (CLASSIFY THING NOMINALIZATION))) 
:Chooser THINGTYPECHOOSER 
:Selector CHOICEMASTER 
36 

:Region NOUNTYPE 
:Metafunction IDEATIONAL) 
(system 
:name PROCESSUALTHINGTYPE 
:inputs 
PROCESSUAL 
:outputs 
((0.25 ACTIONPROCESSUAL) (0.25 MENTALPROCESSUAL) 
(0.25 COMMUNICATIONPROCESSUAL) (0.25 RELATIONALPROCESSUAL)) 
:chooser PROCESSUALTHINGTYPECHOOSER 
:region PROCESSUALTHINGTYPE 
:metafunction IDEATIONAL) 
In analogy, the transitivity systems are made available for participles with adjective function. 
Again, the system and chooser are distinct and only the inquiries can be reused. 
3.3.8 Quantification and Quantitygroup 
Three types of quantification occurred in the texts of our sample domains: 
(28) 79035 t Kase (79035 tonnes of cheese) (numerative, measure, mass noun) 
(29) 2,6 Prozent (2.6 percent) (numerative, measure) 
(30) 40 Firmen (40 companies) (numerative, countable noun) 
`Kase', `Prozent' and `Firmen', respectively, act as heads in these nominal groups. In (28) `Kase' 
has a measureascription (which is a subtype of quantityselection in the Upper Model). In (30) 
`Firmen' also has a quantityascription, but its subtype is that of `exactly'. Only the first results in 
the generation of a quantifiergroup --- the quantification in (30) does not have a further structure, 
so it is realized in the same cycle. The quantifiergroup in (28) consists of a noun which is of the 
semantic type measure (`t' (tonnes)), which itself is further modified by a numerative. 
Problems with this treatment are that in (28) `Kase' is the semantic head, but not the syntactic 
head: 
(31) Die Kaseunion hat die 79035 t Kase abgesetzt. 
\Lambda Die Kaseunion hat den 79035 t Kase abgesetzt. 
(English gloss: The cheeseunion has sold the 79035 tonnes of cheese.) 
Further variants of (28) are (32) and (34) (partitive constructions). 
(32) 79035 t feinsten Kases (genitive) 
(33) 79035 t feinster Kase (nominative); 79035 t feinsten Kaese (accusative) 
(34) 79035 t von feinstem Kase (PP dative) 
In (32) and (34) `Kase' acts as postmodifier syntactically, but even in (33) it is the `t' that is the 
syntactic head. 
The actual quantitytype system in the region quantitygroup distinguishes four types: recov 
erable, cardinal, proportion and ordinal. 
37 

(system 
:name QUANTITYTYPE 
:inputs QUANTITYGROUP 
:outputs ((0.25 RECOVERABLEQUANTITY1) 
(0.25 CARDINALQUANTITY1 (INSERT HEAD) 
(CLASSIFY HEAD NUMERAL CARDINAL)) 
(0.25 PROPORTIONQUANTITY1) 
(0.25 ORDINALQUANTITY)) 
:chooser QUANTITYTYPECHOOSER 
:region QUANTIFICATION 
:metafunction IDEATIONAL) 
3.3.9 Nominalperson and Pronoun 
The region nominalperson is the entry condition to the region of pronoun. It is used to determine 
the grammatical person of pronouns, speaker meaning first person, speakerplus meaning first person 
plural, addressee meaning secondperson. Third person is included in the more delicate options of 
the noninteractant feature of the nominalperson system. 
(system 
:name NOMINALPERSON 
:inputs (OR NONELABORATINGNOMINALGROUPCOMPLEX 
EXTENDINGNOMINALGROUPCOMPLEX) 
:outputs ((0.5 INTERACTANT) (0.5 NONINTERACTANT)) 
:chooser NOMINALPERSONCHOOSER 
:region NOMINALPERSON 
:metafunction INTERPERSONAL) 
(system 
:name INTERACTANTPERSON 
:inputs INTERACTANT 
:outputs ((0.33333334 SPEAKER) (0.33333334 SPEAKERPLUS) 
(0.33333334 ADDRESSEE)) 
:chooser INTERACTANTPERSONCHOOSER 
:region NOMINALPERSON 
:metafunction INTERPERSONAL) 
All pronouns are accounted for in gates with conjunctive entry conditions that specify case, number 
and gender features of a specific pronoun, e.g., 
(gate 
:name SIE 
:inputs 
(AND pronominalspecific nonplural nominative FEMININE) 
:outputs 
((1.0 SiePRONOUN (LEXIFY THING sie))) 
:region PRONOUN 
:metafunction INTERPERSONAL) 
3.4 Adjectivalgroup 
The adjectival group is preselected from the nominal group when there is an Epithet or a Classifier 
inserted and from clause rank for predicative adjectives (subject or object complement). 
38 

3.4.1 Adjectivalcomparison 
The region adjectivalcomparison contains mainly information about scalability of adjectives 
(e.g., elektronisch (electronic)  nonscalable; schon (beautiful)  scalable) and distinguishes between 
nonscaled and scaled adjectives (e.g., sehr schon (very beautiful)  scaled). 
(system 
:name SCALABILITY 
:inputs ADJECTIVALGROUP 
:outputs ((0.5 SCALABLE) 
(0.5 NONSCALABLE (CLASSIFY QUALITY ADJECTIVE) 
(OUTCLASSIFY QUALITY DEGREE))) 
:chooser SCALABILITYCHOOSER 
:region ADJECTIVALCOMPARISON 
:metafunction IDEATIONAL) 
(system 
:name SCALINGTYPE 
:inputs SCALED 
:outputs ((0.5 COMPARATIVESCALING) 
(0.5 SUPERLATIVESCALING)) 
:chooser SCALINGTYPECHOOSER 
:region ADJECTIVALCOMPARISON 
:metafunction IDEATIONAL) 
3.5 Adverbialgroup 
3.5.1 Adverbial 
Adverbials can be subclassified according to metafunction: 
(system 
:name ADVERBIALTYPE 
:inputs 
ADVERBIALGROUP 
:outputs 
((0.33333334 INTERPERSONALADVERBIAL 
(CLASSIFY CIRCUMSTANCE ATTITUDINALADVERB)) 
(0.33333334 EXPERIENTIALADVERBIAL) 
(0.33333334 TEXTUALADVERBIAL (CLASSIFY CIRCUMSTANCE LINKER))) 
:chooser ADVERBIALTYPECHOOSER 
:region ADVERBIAL 
:metafunction IDEATIONAL) 
The komet grammar currently deals only with one type of experiential adverbial, e.g., Das Geschaft 
geht gut. (Business runs well.): 
(system 
:name ADVERBIALGROUPTYPE 
:inputs EXPERIENTIAL 
:outputs ((0.5 QUALITYADVERB 
(CLASSIFY CIRCUMSTANCE ADVERB QUALITY)) 
(0.5 NOTQUALITYADVERB)) 
:chooser ADVERBIALGROUPTYPECHOOSER 
:region ADVERBIALS 
:metafunction EXPERIENTIAL) 
39 

These can be intensified (just as adjectives can  this system is paralleled by adjective 
intensification in the adjectivalgroup), e.g., Das Geschaft geht sehr gut. (Business runs very 
well.): 
(system 
:name ADVERBIALINTENSIFICATION 
:inputs QUALITYADVERB 
:outputs ((0.5 ADVERBINTENSIFIED (INSERT INTENSIFIER) 
(PRESELECT INTENSIFIER ADVERBIALGROUP) 
(ORDER INTENSIFIER CIRCUMSTANCE)) 
(0.5 ADVERBNOTINTENSIFIED)) 
:chooser ADVERBIALINTENSIFICATIONCHOOSER 
:region ADVERBIALS 
:metafunction IDEATIONAL) 
3.6 Prepositional Phrase 
The prepositional phrase is divided into two regions: PPother and PPspatiotemporal. 
3.6.1 PPother 
The major types of prepositional phrases (PPs) are spatiotemporalprocess (which in itself 
builds the region PPspatiotemporal), instrumentalprocess, comparativeprocess, causalprocess, 
accompanimentprocess, matterprocess, roleprocess and changingprocess. 
(system 
:name MINORPROCESSTYPE 
:inputs PREPOSITIONALPHRASE 
:outputs ((0.2 SPATIOTEMPORALPROCESS) 
(0.1 INSTRUMENTALPROCESS) 
(0.1 COMPARATIVEPROCES) 
(0.1 CAUSALPROCESS) 
(0.1 ACCOMPANIMENTPROCESS) 
(0.1 MATTERPROCES) 
(0.1 ROLEPROCESS 
(CLASSIFY MINORPROCESS ROLE)) 
(0.1 CHANGINGPROCESS)) 
:chooser MINORPROCESSTYPECHOOSER 
:region PPOTHER 
:metafunction IDEATIONAL) 
The realizational consequence of each of these types is the choice of a particular preposition (the 
Minorprocess) which governs a particular case of the embedded nominal group (the Minirange). 
These are mostly specified in gates because they have multiple realization conditions, e.g., Bei 
Grippe handelt es sich um eine Virusinfektion. (Influenza is a viral infection.). 
(gate 
:name BEIMINORPROCESS 
:inputs (AND FULLMINORPROCESS NONFIGURATIVEMINORPROCESS MATTERPROCESS) 
:outputs ((1.0 BEIMINORPROCESS 
(LEXIFY MINORPROCESS BEI) 
(PRESELECT MINIRANGE DATIVECASE))) 
:region PPOTHER 
:metafunction IDEATIONAL) 
40 

Not all of these types are fully implemented yet; the best developed area is the region ppspatiotem 
poral. 
3.6.2 PPspatiotemporal 
Differences between English and German in the area of prepositional phrases (PPs) encoding a 
locational spatial relation mainly lie in the potential ambiguity of German prepositions between 
motionprocess, i.e., PPs with a dynamic meaning, e.g., in die See (into the sea), and restprocess, 
i.e., PPs with a static meaning, e.g., in der See (in the sea), in the area of spatial PPs. In English 
the difference between motionprocess and restprocess is reflected in the preposition itself (into vs. 
in); in German, the disambiguating factor is the case a preposition governs (usually accusative vs. 
dative). 
(system 
:name LOCATIONSTATUS1 
:inputs SPATIALPROCESS 
:outputs ((0.5 MOTIONPROCESS 
(PRESELECT MINIRANGE ACCUSATIVECASE) 
(CLASSIFY MINORPROCESS LOCDYN) 
(CLASSIFY MINORPROCESS PREPOSITION LEXICAL GACC)) 
(0.5 RESTPROCESS (PRESELECT MINIRANGE DATIVECASE))) 
:chooser LOCATIONSTATUS1CHOOSER 
:region RANKING 
:metafunction IDEATIONAL) 
Also, the case a preposition governs depends on whether the motion is away from or towards an 
object (e.g. Wir laufen vom Hafen (dative) ins Stadtzentrum (accusative) (We walk from the port 
to the center.). Again, case assignment is specified by preselection. 
(system 
:name MOTIONPROCESSTYPE 
:inputs MOTIONPROCESS 
:outputs ((0.5 AWAYFROM (PRESELECT MINIRANGE DATIVECASE)) 
(0.5 TOWARDS (PRESELECT MINIRANGE ACCUSATIVECASE))) 
:chooser MOTIONPROCESSTYPECHOOSER 
:region PPSPATIOTEMPORAL 
:metafunction IDEATIONAL) 
The region ppspatiotemporal is divided along two axes: extent/location and spatial/temporal: 
(system 
:name SPATIOTEMPORAL 
:inputs SPATIOTEMPORALPROCESS 
:outputs ((0.5 EXTENTPROCESS) 
(0.5 LOCATIONPROCESS)) 
:chooser SPATIOTEMPORALCHOOSER 
:region PPSPATIOTEMPORAL 
:metafunction IDEATIONAL) 
(system 
:name EXTENTPROCESSTYPE 
:inputs EXTENTPROCESS 
:outputs ((0.5 ABSOLUTEEXTENTPROCESS) 
41 

(0.5 RELATIVEEXTENTPROCESS)) 
:chooser EXTENTPROCESSTYPECHOOSER 
:region PPSPATIOTEMPORAL 
:metafunction IDEATIONAL) 
(system 
:name SPATIOTEMPORALTYPE 
:inputs (OR LOCATIONPROCESS RELATIVEEXTENTPROCESS) 
:outputs ((0.5 SPATIALPROCESS) 
(0.5 TEMPORALPROCESS)) 
:chooser SPATIOTEMPORALTYPECHOOSER 
:region PPSPATIOTEMPORAL 
:metafunction IDEATIONAL) 
A SPECIAL CASE: Pronominalized PPs 
Another grammatical alternative to express the interpretation type of elaboration (cf. Section 3.2) 
is in separate independent clauses that have a demonstrative referring to the previous proposition 
(see Examples (35)  (36)). 
(35) Damit ging der Absatz um 2,6 Prozent zuruck. 
(English gloss: With this, the sales reclined by 2.6 percent.) 
(36) Damit enspricht der Ruckgang 2,6 Prozent. 
(English gloss: With this, the recline corresponds to 2.6 percent.) 
Grammatically, `damit' is a pronominal adverb which stands for a prepositional phrase or adverb 
or even for a whole proposition (in which case it would better be called a pronominal proposition), 
e.g., 
(37) Mit diesem Resultat sind wir sehr zufrieden. 
We are very satisfied with this result. 
(38) Damit sind wir sehr zufrieden. 
(English gloss: With this, we are very satisfied.) 
Pronominal adverbs are prowords that are used instead of personal, demonstrative, interrogative 
or relative pronouns. This has the following pattern: 
ffl `da' for personal and demonstrative pronouns 
ffl `wo' for interrogative and relative pronouns 
For further details on the usage of pronominal adverbs we refer the reader to [Teich1991]. 
Currently the grammatical possibility of pronominalizing adverbials (i.e., circumstantials in general) 
is described in the grammar of the prepositional phrase. Parallel to the system minorprocess 
type there is a system prepositionalphrasedeicticity with two options fullminorprocess and 
pronominalizedminorprocess. The choice of one or the other lies in textual reasons. The realization 
statement for pronominalizedminorprocess is (conflate Minorprocess Minirange). The chooser of 
the system has an inquiry similar to emptynumberq for nouns, emptyq. A pronominalized PP can 
then either be pronominalized by a whpronominaladverb (e.g., `womit', `woher' or a dpronominal 
adverb (e.g., `damit', `daher'). See these two systems below. 
(system 
:name PREPOSITIONALPHRASEDEICTICITY 
:inputs PREPOSITIONALPHRASE 
42 

:outputs ((0.5 FULLMINORPROCESS (ORDER MINORPROCESS MINIRANGE)) 
(0.5 PRONOMINALIZEDMINORPROCESS (CONFLATE MINORPROCESS MINIRANGE))) 
:selector CHOICEMASTER 
:chooser PREPOSITIONALPHRASEDEICTICITYCHOOSER 
:region PPOTHER 
:metafunction IDEATIONAL) 
(system 
:name PRONOMINALIZEDMINORPROCESSTYPE 
:inputs PRONOMINALIZEDMINORPROCESS 
:outputs ((0.5 WHMINORPROCESS) 
(0.5 DMINORPROCESS)) 
:selector CHOICEMASTER 
:chooser PRONOMINALIZEDMINORPROCESSTYPECHOOSER 
:region PPOTHER 
:metafunction IDEATIONAL) 
The chooser of the second system uses questionq which is generally used in the grammar to de 
termine whether the pronoun to be generated is interrogative or not. For our example (35) we 
gather the following features for `damit': accompanimentprocess, pronominalizedminorprocess, d 
minorprocess. A gate with these input features leads to the realization of the pronominal adverb as 
`damit' --- if whminorprocess is chosen, `womit' would be generated. 
3.7 Verbalgroup 
As mentioned in the beginning, there is no ``physical'' verbal group in the komet grammar yet. 
However, there are some systems that refer to parts of a verbal group, such as the lexical verb, the 
finite etc. 
3.7.1 Verbmorphology 
One such region is that of verbmorphology which e.g., includes a system that deals with separable 
prefixes of verbs. The problem with separable prefixes is that they have to ordered on clause rank: 
they come at the end of the clause when the clause is independent and the tense is noncomposite 
(i.e., present or simple past). See example (39). 
(39) Die Grippe tritt hauptsachlich in Herbst und Winter auf. 
(Influenza occurs primarily in fall and winter.) 
In dependent clauses and with composite tenses the prefix is not separated from the stem (see 
example (40). 
(40) Wir wissen, dass die Grippe hauptsachlich in Herbst und Winter auftritt. 
(We know that influenza occurs primarily in fall and winter.) 
For the ordering of the prefix we include a system morphologyofverb1 on clause rank, the 
chooser of which poses inquiries to the lexicon to check whether a verb has a separable prefix or not. 
The prefix is then ordered accordingly. 
(system 
:name MORPHOLOGYOFVERB1 
:inputs PROCESSLEXICALIZATION 
:outputs ((0.5 SEPARABLEPREFIXVERB (INSERT PREFIX)) 
(0.5 NONSEPARABLEPREFIXVERB)) 
43 

:chooser MORPHOLOGYOFVERB1CHOOSER 
:region VERBMORPHOLOGY 
:metafunction EXPERIENTIAL) 
(gate 
:name ORDERINGOFPREFIX 
:inputs (AND SEPARABLEPREFIXVERB 
NONELABORATIONASSUBORDINATERELATIVECLAUSE) 
:outputs ((1.0 PREFIXATEND (ORDERATEND PREFIX))) 
:region VERBMORPHOLOGY 
:metafunction EXPERIENTIAL) 
References 
[Bateman et al.1990] John A. Bateman, Robert T. Kasper, Johanna D. Moore, and Richard A. 
Whitney. A general organization of knowledge for natural language processing: the penman 
upper model. Technical report, USC/Information Sciences Institute, Marina del Rey, California, 
1990. 
[Bateman et al.1991] John A. Bateman, Elisabeth A. Maier, Elke Teich, and Leo Wanner. To 
wards an architecture for situated text generation. In International Conference on Current Is 
sues in Computational Linguistics, Penang, Malaysia, 1991. Also available as technical report of 
GMD/Institut fur Integrierte Publikations und Informationssysteme, Darmstadt, Germany. 
[Bateman et al.1992] John A. Bateman, Martin Emele, and Stefan Momma. The nondirectional 
representation of Systemic Functional Grammars and Semantics as Typed Feature Structures. In 
Proceedings of COLING92, Nantes, France, July 1992. 
[Bateman1990] John A. Bateman. Upper modeling: organizing knowledge for natural language 
processing. In 5th. International Workshop on Natural Language Generation, 36 June 1990, 
Pittsburgh, PA., 1990. Organized by Kathleen R. McKeown (Columbia University), Johanna D. 
Moore (University of Pittsburgh) and Sergei Nirenburg (Carnegie Mellon University). 
[Bateman1992] John A. Bateman. Grammar, systemic. In Stuart Shapiro, editor, Encyclopedia of 
Artificial Intelligence, Second Edition, pages 583 -- 592. John Wiley and Sons, Inc., 1992. 
[Bresnan1982] Joan Bresnan, editor. The Mental Representation of Grammatical Relations. The 
MIT Press, Cambridge, MA, 1982. 
[Butler1985] Christopher S. Butler. Systemic linguistics: theory and applications. Batsford, London, 
1985. 
[Dik1978] Simon C. Dik. Functional Grammar. North Holland, Amsterdam, 1978. 
[Fawcett1980] Robin P. Fawcett. Cognitive Linguistics and Social Interaction. Exeter University 
and Julius Groos Verlag, Exeter and Heidelberg, 1980. Exeter Linguistic Studies 3. 
[Gazdar et al.1985] Gerald Gazdar, Ewan Klein, Geoffrey Pullum, and Ivan A. Sag. Generalized 
Phrase Structure Grammar. Blackwell Publishing and Harvard University Press, Oxford, England 
and Cambridge, Massachusetts, 1985. 
[Halliday and Matthiessen1990] Michael A.K. Halliday and Christian M.I.M. Matthiessen. Histori 
cal survey of systemic linguistics. In Fred C.C. Peng and J. Ney, editors, Current Approaches to 
Syntax. Benjamins and Whurr, Amsterdam and London, 1990. (Volume in preparation). 
44 

[Halliday1978] Michael A.K. Halliday. Language as social semiotic. Edward Arnold, London, 1978. 
[Halliday1985] Michael A.K. Halliday. An Introduction to Functional Grammar. Edward Arnold, 
London, 1985. 
[Heidolph et al.1980] K.E. Heidolph, W. Flamig, and W. Motsch. Grundzuge einer deutschen Gram 
matik. AkademieVerlag, Berlin, 1980. 
[Helbig and Buscha1988] Gerhard Helbig and Joachim Buscha. Deutsche Grammatik. Ein Handbuch 
fur den Auslanderunterricht. VEB Verlag Enzyklopadie, 1988. 
[Henschel1992a] Renate Henschel. A proposal for merging the english and german upper models. 
Technical report, GMD/Institut fur Integrierte Publikations und Informationssysteme, Darm 
stadt, West Germany, Darmstadt, Germany, 1992. 
[Henschel1992b] Renate Henschel. Towards a declarative representation for systemic grammar. Tech 
nical report, GMD/Institut fur Integrierte Publikations und Informationssysteme, Darmstadt, 
West Germany, Darmstadt, Germany, 1992. 
[Hjelmslev1943] Louis Hjelmslev. Omkring sprogteoriens grundlaeggelse. Akademisk Forlag, 
Kpenhavn, 1943. 
[Hudson1976] Richard A. Hudson. Arguments for a nontransformational grammar. Chicago Uni 
versity Press, Chicago, 1976. 
[Kasper and Whitney1989] Robert Kasper and Richard Whitney. SPL: A sentence plan language 
for text generation. Technical Report forthcoming, Information Sciences Institute, 1989. 4676 
Admiralty Way, Marina del Rey, California 902926695. 
[Maier and Hovy1991] Elisabeth A. Maier and Eduard H. Hovy. A metafunctionally motivated 
taxonomy of discourse structure relations. In Proceedings of the 3rd European Natural Lan 
guage Generation Workshop, Innsbruck, March, 1991, 1991. Also available as technical report of 
GMD/Institut fur Integrierte Publikations und Informationssysteme, Darmstadt, West Germany. 
[Mann and Thompson1987] William C. Mann and Sandra A. Thompson. Rhetorical structure the 
ory: a theory of text organization. In Livia Polanyi, editor, The Structure of Discourse. Ablex 
Publishing Corporation, Norwood, New Jersey, 1987. Also available as USC/Information Sciences 
Institute Technical Report Number RS87190. 
[Matthiessen and Bateman1991] Christian M.I.M. Matthiessen and John A. Bateman. Text gener 
ation and systemicfunctional linguistics: experiences from English and Japanese. Frances Pinter 
Publishers and St. Martin's Press, London and New York, 1991. 
[Matthiessen1984] Christian M.I.M. Matthiessen. Choosing tense in English. Technical Report 
ISI/RR84143, USC/Information Sciences Institute, Marina del Rey, CA, 1984. 
[Matthiessen1989] Christian M.I.M. Matthiessen. Text Generation as a Linguistic Research Task. 
PhD thesis, Department of Linguistics, UCLA, Los Angeles, CA, 1989. 
[Matthiessen1995] Christian M.I.M. Matthiessen. Lexicogrammatical cartography: English systems. 
International Language Science Publishers, Tokyo, Taipei and Dallas, 1995. 
[Mel'cuk1988] Igor A. Mel'cuk. Dependency Syntax: Theory and Practice. State University of New 
York Press, Albany, 1988. 
[Penman Project1989] Penman Project. penman documentation: the Primer, the User Guide, the 
Reference Manual, and the Nigel manual. Technical report, USC/Information Sciences Institute, 
Marina del Rey, California, 1989. 
45 

[Pollard and Sag1987] Carl Pollard and Ivan A. Sag. Informationbased syntax and semantics: vol 
ume 1. Chicago University Press, Chicago, 1987. Center for the Study of Language and Informa 
tion; Lecture Notes Number 13. 
[Steiner et al.1988] Erich H. Steiner, Paul Schmidt, and Cornelia ZelinksyWibbelt, editors. From 
Syntax to Semantics: insights from Machine Translation. Frances Pinter, London, 1988. 
[Teich1989a] Elke Teich. Komet grammar documentation part i. Technical report, GMD/Institut 
fur Integrierte Publikations und Informationssysteme, Darmstadt, West Germany, 1989. 
[Teich1989b] Elke Teich. Systemic linguistics in natural language processing. Master's thesis, Uni 
versitat des Saarlandes, Saarbrucken, West Germany, 1989. (Magisterarbeit). 
[Teich1990a] Elke Teich. Komet grammar documentation part ii. Technical report, GMD/Institut 
fur Integrierte Publikations und Informationssysteme, Darmstadt, West Germany, 1990. 
[Teich1990b] Elke Teich. A systemically based treatment of `raising' and `control' in the context 
of machine translation. In Eija Ventola, editor, Recent Systemic and Other Views on Language. 
Benjamins, Amsterdam, 1990. 
[Teich1991] Elke Teich. Komet grammar documentation part iii. Technical report, GMD/Institut 
fur Integrierte Publikations und Informationssysteme, Darmstadt, West Germany, 1991. 
[Wanner1992] Leo Wanner. Lexical choice and the organization of lexical resources in text gener 
ation. In European Conference on Artificial Itelligence, Vienna, Austria, 1992. Also available as 
technical report of GMD/Institut fur Integrierte Publikations und Informationssysteme, Darm 
stadt, Germany. 
46 

