1. General design information
 

The SIMPLE Semantic Lexicon for German has been developed in the framework of the SIMPLE project, which started in April 1998. It aimed at adding semantic information on top of the German PAROLE morphological and syntactic lexicon. In assigning the relevant semantic information to the Parole units we followed the common model, linguistic specifications, DTD and other formats as laid down in the Linguistic specifications.

1.1. Lexicon population

The German part of SIMPLE consists of two parts - the data which were provided by MANNHEIM (MAN) (4/5^th) and those of GRAZ (GRA) (1/5^th). All initial data for encoding were extracted from the German - PAROLE lexicon by MAN and then made available to Graz. The frequency of the lexical items was taken into consideration and items with high frequency preferred. Due to the linking with the PAROLE-data Graz was dependant on the data provided by Mannheim which supplied us with a list of 327 noun lemmas, 153 verb lemmas and 637 adjective lemmas. As Mannheim was not doing any adjective encoding, this part of the encoding was left to Graz which explains why the amount of the adjective lemmas was considerably higher than the numbers of the other POS.

The Base Concepts have been selected for encoding when equivalents were present in the German PAROLE lexicon. However, no previsions were taken to chose mainly "words" which were particularly representative for the Eurowordnet Basic concepts as the basic concepts (BC) turned out to be biased by semantic divisions only representative for English and on the other hand not covering important semantic features. Some readings could not be associated with BCs and were therefore marked with the feature BC0.

The choice of the lemmas selected was primarily driven by two criteria:

presence of the item in German Parole lexicon (having a syntactic link) and providing a large range of synonyms.

1.2 Background resources and lexicographical tools

The tool for achieving the encoding-tasks was the MS-Access database developed by the Catalan partners of the Institut d'Estudis Catalans (IEC) in Barcelona. We would like to express our gratitude to our colleagues for the generosity and the support we received in the course of the project. In particular we would like to thank Teresa Sadurni for the development and the maintenance of the database and her help we received on many occasions.
 

We have also used background resources and extracted semantic groups from CD dictionaries and printed synonym dictionaries. One of the main sources was the LEXIROM, the Duden Synonym dictionary as well as Mackensen’s synonym dictionary. Most definitions have been extracted from the electronic versions of the Lexirom and other dictionary sources. All encoded words in our lexicon include a definition which were mainly based on the dictionary sources or - if for some reason inappropriate - we developed one of our own. For the encoding of predicates we used standard valence dictionaries like Helbig/Schenkel and other sources (see the bibliography).
 

1.3 Coverage and completeness

The encoding task has been organised in such a way that encoders first chose the lemmas which according to the dictionaries provided large groups of closely related entries. This strategy proved to be quite useful for the consistency of the encoding and the numbers of Semus.

It is important to point out that due to the structure of the German Parole lexicon, our list of basic lemmas consisted mainly of abstract nouns and verbs with cognitive content. The encoding of these items proofed to be quite challenging as there were only few template types and semantic features available in the given ontology. To cope with this would have meant to introduce a number of additional templates and possibly  requiring a  modification of the framework set out by the Specification Group. We therefore refrained from taking this step and restricted the encoding mainly to the mandatory and required features.

As most abstract nouns at our disposition were mainly derivational and deverbal by origin one of the effects was that a number of Dummy Usems had to be encoded as they were missing in the PAROLE lexicon.

The dummy elements are of two types:

• single word senses that did not exist in the PAROLE lexicon and therefore could not be semantically encoded;
• single verbs for predicates that did not exist in the PAROLE lexicon and therefore could not be encoded semantically.

1.4. Validation

In order to check the consistency of our encoded date we used the utilities and checking routines provided by the MS-Access database provided by BAR. All checkings were run and the reported errors as well as inconsistencies in the encoding removed.

1.5. Lexicon Contents

Factual data about the lexicon related to the Parole lexicons.

Table 2: TOTAL number of Dummys, Predicates, Synus addressed in SIMPLE
 

2. Semantic encoding

In the German SIMPLE lexicon the encoding of entries was performed using the encoding tool supplied by the Catalan site. The tool is an interface which allows to encode/check/browse the Simple data, stored in a relational database (MS-Access). It has been created by Marta Villegas and Teresa Sadurní and developed by Teresa Sadurní at the Institut d'Estudis Catalans (IEC), in Barcelona.

This tool has provided most valuable help for the encoding process. Its great number of very useful options for inserting entries, copying and maintaining existing data has allowed to perform an accurate and quick coding of entries and a continuous consistency checking of data.

2.1. Criteria for Syntax-Semantic linking

The syntactic-semantic linking is established on a one-to-one relation between SynUs and SemUs at the level of correspondence objects. As we were using the MS-Access database the choices for the linking were automatically proposed by the db and had to be chosen according to the particular features of the SEMU.

In the German PAROLE lexicon, entries may have one or more syntactic units. As a rule, all meanings of one lemma corresponding to the most general syntactic description were first encoded and the others left aside for eventual later treatment. Consequently, this one syntactic unit is linked to all the subsenses (semantic units) of the word in question.
 

2.1.1 One to many:
 

<SynU
id="USADJ1235"
example=""
naming=".A1"
description="DADJ17">
<CorrespSynUSemU
targetsemu="USem1948"
correspondence="ISObivalent">
<CorrespSynUSemU
targetsemu="USem1949"
correspondence="ISObivalent">
<CorrespSynUSemU
targetsemu="USem2262">
<CorrespSynUSemU
targetsemu="USem2263">
</SynU>

 2.1.2 One to one:

<SynU
id="USN7132"
example=""
naming=".N1"
description="DN2">
<CorrespSynUSemU
targetsemu="USem406"
correspondence="ISOmonovalent">

2.1. 1.Types of Correspondence

Subtypes of Correspondence used:
 

Correspondence	Description	Example	Comment
ISOmonovalent	Isomorphic mapping for unary predicates	Autor [author] gehen [to go]	Nouns Monovalent verbs
ISObivalent	Isomorphic mapping for bivalent predicates	ändern[to change]	Bivalent verbs
ISOtrivalent	Isomorphic mapping for trivalent predicates	überzeugen [to convince]	Trivalent verbs

RED2to1	reduced mapping for predicates with two arguments where Arg1 changes to Arg0	Aufdruck [impresssion] Ausbildung [education]	deverbal nouns derived from verbs with a direct object or prepositional object
RED3to2	reduced mapping for predicates with three arguments where Arg2 goes to Arg0 and Arg2 to Arg1	Rennen [run] Auseinandersetzung [conflict, quarrel, discussion]	deverbal nouns with two prepositional objects or two objects

 2.2. Criteria for assigning Domain Features

As recommended in Linguistic Specifications, we tried to assign to each Usem a specific semantic class and template type available in the Linguistic Specifications as provided by the Access database. Due to the internal structure of the database, it was not possible to assign more than one domain.

Assignment of the Domain categories has followed in most cases the recommendations given in the Linguistic Specifications D2.1. The main criteria for the assignment of the Domain values were the following:

• Domain values were primarily related to the subject field in which a given SemU is more likely to appear or to be concomitant with.
• e.g. the SemU "Sicherheitsbehörde" [security body, security authority/forces] has been assigned the Domain "TSVP_LAW_ENFORCEMENT_TS_domaine" as this is most likely to be connected with at present.

---


<SemU

id = "USem413"
naming = "Polizei"
example = "Sicherheitsbehörde"

weightvalsemfeaturel="TSVP_GROUP_NAMES_TS_classificateur_de_nom_C
TSVP_LAW_ENFORCEMENT_TS_domaine_D TSVP_PLUS_TS_COLLECTIVE_T
TSVP_PLUS_TS_HUMAN_T TSVP_PLUS_TS_PROFESSIONAL_T

---

 
• When a general and a more specific domain were available, we have generally chosen the more specific one. The SEMU "Postamt" was assigned the feature "TSVP_MAIL_TS_domaine_D" although it might just as well be assigned the feature "TSVP_SERVICE_INDUSTRY" which seems to fit equally well even though it is not yet considered to be part of an "industry" in many countries but part of the public services. However this domain is missing in the ERLI list.

---


<SemU
id = "USem414"
naming = "Post"
example = "Postamt"
comment = " "
freedefinition = "BC0"
formaldefinition = "Amt"
weightvalsemfeaturel="TSVP_BUILDING_TS_classificateur_de_nom_CTSVP_MAIL_TS_domaine_D
TSVP_PLUS_TS_LOCATIVE_T WVSFDimension3PROT WVSFTemplateBuildingPROT

---

 
• Problems with abstract nouns

 

 

In a considerable number of cases we encountered problems with abstract deverbal nouns like "Abbruch" [demolition] which has a relatively "concrete" meaning in connection with the demolition of house in the field of construction works. In this connection it can be assigned the domain TSVP_CONSTRUCTION_TS_domaine_D.The word has a even more general sense denoting any kind of stopping or finishing of an action which cannot be linked to any specific field. In this case the domain field was left unspecified which is equivalent to the domain class "General". Another illustrative case is the word "Anlass" [occasion, cause, reason] which- like many other expressions denoting cognitive processes or relations - does not allow the assignment of any specific domain.
 

Linguistic specifications by the specification group were the main criteria for assigning template type and semantic class information. The dictionaries of Helbig/Schenkel and Ballmer/Brennenstuhl were used for the encoding of verbs, in adjective and noun encoding valuable help was found in the dictionaries of Sommerfeldt/Schreiber.
 

2.3.1. Language specific typing

This gives an overview over a demonstrative selection of most common assingement types for template types:

1. General process verbs like"verändern" [change] "verarbeiten"[process] or"abservieren" [clear the table]which are either monovalent or bivalent with a direct object were assigned the template type "Change_state" or "Cause_Change_of_State".
2. Event verbs like "verschwinden" [vanish] or "fortbringen" [remove] which imply a change of location were assigned thetemplate type "Cause_Change_Location".
3. Process verbs like "austeilen/verteilen"[distribute] which imply a change of ownership were assigned thetemplate type "Change_of_Possession
4. Causative verbs like "veranlassen" [cause, bring to action] were assigned the template type "Activity".
5. Factitive verbs like "tangieren" [concern, effect] which imply an obligatory patient role were assigned the template type "Cause_Motion".
6. Modality verbs like "verbessern" [improve] which imply a modal adjective complement were assigned the template type "Change_of_Value"
7. Cognitive verbs like "erinnern" [remember] and "überlegen" [reflect, think about] (which are also event verbs) were assigned the template type "Cognitive_Event"
8. Nouns denoting cognitive states like"Geistlosigkeit" [mindlessness] or regular (stative) processes like "Einfluss" [influence] were assigned the template type "Cognitive_Fact".
9. Nouns denoting social events like "Feier" [celebration] were assigned the template type "Convention"
10. Cooperative verbs describing general social activities like "treffen" [meet] or "helfen" [help] which require cooperative action were assigned the template type "Cooperative_Activity".
11. A considerable number of nouns (mostly of deverbal origin) like "Informierung" [briefing], Engagement [commitment] which imply some intentional process or activity are assigned the template "telic".

2.3.3. Criteria for encoding Semantic Relations

In the SIMPLE model semantic relations between SemU’s are intended to increase the descriptive power of the model. The encoding of semantic relations between different entries was done on the basis on choosing the linguistically relevant distinctive semantic relations, even if these point to dummy SemU's.

The dummy Usems collected in the course of the encoding arose from the expansion of lexicon population mainly by virtue of the many abstract derivational nouns we had to deal with. The other source was the already mentioned wish to cover the relevant parts of the semantics of the words encoded in a comprehensive way.

Examples to illustrate the assignment of dummys are
 

• USem "Engagement" [engagement, commitment]
Has as target "Activity - DUMMYMitarbeit" as any kind of "engagement" in an activity requires "Mitarbeit" [assistance, collaboration, cooperation].
• USem "Erinnerung" [recollection, memory]
Has as target "Activity - DUMMYVorstellung" as any kind of activity in the brain is necessarily bound to the "imagination" of what is thought about.

• USem "Weltoffenheit" [open mindedness] is "Agentive" and solely linked to the entity "Mensch"
• USem "aktiv" has as sematic relation "Constitutiveactivity" and as target "Manager" as any manager has "activity" as inherent feature.
• USem "Artikel" has as sematic relation "Contains" and as target "Information" as this is the purpose of an article. Etc. etc.

Polysemic relations have not been implemented which is again due to the fact that we mainly had to deal with abstract word senses.

Synonimy was only encoded with some adjectives.

Hyponimy was not encoded.

Each predicative entry contains information in respect to:

• the predicate to which it is linked,
• the type of link with its predicate,
• the number of arguments the predicate has,
• semantic role of the arguments of the predicate, and

selectional restrictions for the arguments linked to the predicate.

The same predicate is used

• for different lemmas, when a derivation relation holds between them
• e.g. deverbal nouns are linked to the predicate of the verb they are derived from
• deadjectival nouns are linked to the predicate of the relevant adjective
• for different syntactic units of the same lemma if the reading of the predicate is the same, i.e. if the syntactic units are linked to the same semantic unit.

• master: when the lemma and the lexical predicate are the same: verbs e. g."schauen > schauen", deverbal nouns: "Abbau" [deconstruction] <> "abbauen" [deconstruct, decompose etc.]
• agentnominalisation: for nouns that are derived from verbs and denote the agent of the event "Zuschauer" [spectator];
• verbnominalisation: for nouns derived from verbs denoting the event itself: "Anfrage" [request] < "anfragen" [to request]
• adjectivenominalisation : for nouns derived from adjectives: "Wirklichkeit" [reality] < "wirklich" [real].

• verbs
• adjectives
• deverbal nouns (agent and verb nominalisations),
• deadjectival nouns.
• kinship nouns