Tenets for an Interlinguai Representation of 
Definite NPs 
Montserrat Meya 
CDS Barcelona 
Siemens SA, Spain 
e-mail: Montserrat Meyallopart@eurokom.ie 
1o Introduction 
The main goal of this paper (as in Keenan 
and Stavi 1986) is to characterize the 
possible determiner denotations in order to 
develop a computational approach that 
makes explicit use of this information. To 
cope with the constraints that languages 
impose when generating determiners, a 
computational model has to follow the laws 
that map definiteness to structures and 
strings and viceversa. 
In the following proposal I distantiate from 
K. Btihlers Deixis Theory and Weinrichs 
(76) proposal where indefinites suggest 
subsequent information, while definite point 
out facts from the previous intbrmation. This 
very general position is insufficient if we 
want to formalize NP-definiteness. 
The semantics of NP definiteness must be 
captured adequately in computational 
frameworks for such tasks as answering 
quantified NL-- questions, or in a MT system 
to convert NPs from one language into 
another,. 
In the first part of this paper I draw a 
typology of definiteness; later I reflect on the 
definiteness of NPs in an IL-representation. 
The major result is given by the determiner 
generators. Definiteness should be evaluated 
in a Q-A system and in MT. The extensive 
functionality of definiteness is first 
elaborated in the parsing and results in an IL- 
representation; finally the determiner 
generators create correct morphological 
determiners and right determiner structures. 
2, Parsing definite NPs 
Weinrich (1976) distinguishes three types of 
anaphoric and cataphoric information: context, 
situation and code. Code is the paradigmatic 
meaning that speakers find in the lexicon 
(standard definitions). 
The NPs parsing strategy- that copes with 
these three information types- comes down to 
these main questions: 
• the question of building adequate 
semantic interpretation for the functional 
features fixed during analysis 
• The question of reference resolution for 
definite NPs 
To achieve that goal the Np grammar must 
work with a sufficient: 
® determiner subcategorisation 
• strategy about constraints depending 
upon the semantic features of the nouns, 
and on the context (PPs, Rel-S, 
compounding, etc.) 
This strategy, of course, can not rely on the 
presupposition of existence and uniqueness, 
i.e. the "accommodation" model of Lewis 
263 
(1979) or the Russellian analysis of "the" so 
closely related to the analysis of "only", that 
would fail for sentences like: (MacCawley 
81. p.177): 
1. "The restaurant on the Clark street 
is excellent" 
because Russell's analysis did not allow 
truth value gaps or interactions between 
logical form and context (MacCawley. 
t).177), if there is more than one restaurant 
on the Clark Street, sentence 1 expresses a 
false proposition. 
Determiner Subcategorisation 
The list of determiners is surprisingly long 
(Keenan 86) and there is even the possibility 
to build complex determiners by adding 
modifiers to determiners. According to 
keenan and Stavi there are 3 types of 
determiners: cardinal, logical and non 
logical. 
Dets are taken into consideration according 
to the fact of being terminal or pre-terminal 
nodes. This subcategorisation entails the 
following additional feature-value pairs. 
Terminal determiners are subcategorized in 
the lexicon according to the features: 
• definite (logical) vs. indefinite (non- 
logical) 
• demonstrative vs. possessive 
• type of quantification {collective, 
distributional, numeral, amount, 
partitive, negation} 
• comparatives 
Pre-terminal determiner features 
When NP structures are built during parsing, 
the intermediate nodes must come up with 
values for the features: 
• definiteness {proper name, bounded 
NP} 
• negation 
• topicalization 
• genetics 
• head-specifier relation 
• apposition {Titel .... } 
• genitives {possessive, part-of .... } 
• predication {equality vs. copula} 
• type of sta'ucture {PP, Rel-S,...} 
• semantics of the NP {mass vs. count 
noun, temporal, local, argument .... } 
Each feature configuration corresponds with a 
NP structure with full decoration (in a MT 
system), or a compositional formula 
representation (in understanding systems); in 
both cases it is built up rule-to-rule. 
The outgoing object of the parser is a IL- 
representation. Later on the generators will 
define, according to the idiosyncratic laws of 
each language, the morpheme assignment for 
this interlingua (IL). 
Determiner functionality 
The semantic value of the tags {genetics, 
definiteness, topic} for NPs is obtained during 
analysis and it is the result of taking into 
consideration: 
• the whole NP structure {attribution, 
apposition, genitives, head-specifier 
relation in compounds is relevant for 
definiteness. 
• the NP typology together with the 
sentence aspect and time is relevant for 
generics. 
• the semantic primitives tie the lexical 
item to an ontological class and is 
relevant for the possible word readings 
• word order and negation is relevant for 
topic 
Generics 
Generics do not underly universal or nearly 
universal quantifiers. The reference of generics 
mirrors that of habitual propositions. They are 
treated as predication of individuals though 
264 
allowing the clauses in question to take 
different truth values at different times. 
Generics are expressed in the present tense 
with the aspect value habitual (usually, 
occasionally, always). The 3 different 
generic types reflect the different 
quantitication of the determiners in 
propositions like: 
1. "Men do not cry" 
(Quant.=distrib) (=usually every 
man does not cry) 
2. "Germans are bad husbands" 
(Quant.= distr) 
3. % man does not cry" 
(Quant.=mass)(whoever has the 
property of being) 
4. "the lion is a mammal" 
(Quant=coll) (The species...) 
5. "I like lamb" 
Contrary to MacCawley I think that 
indefinite plural generics as in (1) are 
basically references to sets of objects. 
Indefinite singular generic NPs are 
references to properties of individuals (3), 
and definite singular NPs with count nouns 
(4) or indefinite ( without article) mass nouns 
(5) are basically references to kinds. 
The final value "generics" on the NP is 
calculated from the verbal tense value (PR) 
and from the aspect value (habitual) together 
with the determiner of the NP, the semantic 
type of the noun, and the whole NP syntactic 
structure. 
6. "Wine is a drink" 
vs. "el vino es una bebida" 
\[$SUBJ mass\] 
7. "I drink wine" 
vs, "bebo (0) vino" 
\[$OBJ mass\] 
The same generic NP in examples 6/7 
behave differently in spanish according to 
the sentence argument. 
Both the generics and the syntactic role in the 
sentence have an effect on the article 
generation in different languages. 
Head-Specifier Structure 
The addition of features providing information 
about certain syntactic structures is very useful 
in reflecting the nominal definiteness behavior. 
The following structures assist in 
distinguishing (between general and specific 
reference) definite NPs: 
° acronyms and proper names: 
"The Union" 
• appositions: 
"The Siemens company" 
• complex proper names (dates, time: 
2 o'clock, 2.3.1990,...) 
° attribution (with adjectives, relative 
clauses, genitive constructions, 
prepositional phrases, compounding .... ) 
Here some examples with the value 
definiteness =True: 
"The second paragraph", "the cleanest 
water", "The man that I remember", 
Peter's hatred", "the truth about his 
past", "the picture on the wall", "the 
king's daughter",.. 
Genitives in whatever morphol\[ogical form are 
always definite constructs: 
• The king's daughter 
Die Koenigstochter 
la hija del rey 
• House door 
Haustiir 
la puerta de la casa 
Germanic languages only have one determiner 
in compounds, however the genitive relation 
must be captured to reflect the definiteness of 
the specifier in another language. 
During NP parsing, constituents are marked 
according to their head- specifier function 
together with their semantic type.In that way, 
265 
PPs, adjectival phrases and compounds are 
represented in a uniform way. Additional 
feature constraints on head and specifier 
regulate definiteness, depending on their 
semantic primitives: 
• Druckfestigkeit 
resistence a la compresion 
• Druckknopf 
bouton a presion 
• Einkommensteuer 
impot sur la revenue 
° Einfuhrsteue 
impot d'importation 
Directional and local compounds, or the 
corresponding PPs in other languages 
contain the feature definiteness because they 
are usually tied to some deictic situation (See 
Loebner 1985, semantic vs. pragmatic 
detiniteness). 
• im Palast 
en el palacio 
• zum Palast 
al palacio 
Topicalization 
The series "definite article + Lexeme + 
indefinite article + Lexeme" reflects a 
methodological progress in the information 
situation. If this direction changes, the 
direction of the predication also changes. 
That means that the noun (or NP) will be 
focused and therefore has a definite 
reference. 
8. the FBI employed American 
citizens 
(la FBI emplea 0 a ciudadanos) 
9. American citizens are employed by 
the FBI 
10. Los ciudadanos americanos son 
empleados por la FBI 
11. a: Dem Kind gab ich den Ball 
b: Es al nifio a quien di la pelota 
12. The man I love 
Topicalized NPs must be marked as such. 
Voice in example 9 topicalize the NP and 
demands in spanish definiteness for the NP. 
The bare topicalization in german in example 
(lla) corresponds a marked extraposition in 
spanish (11 b). Languages make a 
discriminating use of this feature when 
generating determiner requirements. 
Predicative constructions 
In predicative constructions the relevant 
criteria for the use of determiners are the type 
of copula (NP expresses a property (14-15), or 
a reference (16)), while with other verbs the 
relevant criteria is the aspect (process vs. 
accomplishment or achievement) 
A sentence involving a noun X with definite 
article in predicate position does not assert that 
the subject is X. We have to distinguish 
between a "be" of identity and that of copula. 
13. Felipe es (el) presidente de Espafia 
a: (t x: presidente(x,Espafia)) 
=(Felipe,x) 
14. Alex is an artist vs. Alex es artista 
15. Felipe es el presidente que prefiero 
a: (3x) (PE x, (Vy) 
(~=yx,~PEy),Felipe,x) 
English demands article for copula cases like 
(14), existing the opposite referential value for 
the article "the" (the teacher). Referential 
examples are treated the same in english and 
spanish for identity predicate. 
At the other hand arguments of predicates in 
accomplishments or achievements (not as a 
process) are (16) individualized and as such 
demand definiteness. The sole exception are 
lexicalized expressions and sentences with 
function verbs (17) 
16. "voy a la playa" vs. "I go to the 
beach" (achievement) 
"bese un tigre"(al tigre) vs. I kissed 
266 
17. 
the tiger (accomplish.) 
"me gusta besar tigres" vs."I like to 
kiss tigers" (process) 
"pongo la mesa" vs. "Ich decke den 
Tisch" 
"tengo razon" vs.Ich habe Recht" 
3. Interlingual representation 
of definiteness 
Definiteness is a means to satisfy coherence 
requirements. Definite NPs have a scope just 
like quantifiers and they may be involved in 
scope ambiguities as Russell suggested 
(MacCawley p.178). The most reliable test 
to decide about the definiteness of an NP is 
to use Peterson's (1979) test of there- 
insertion. There insertion requires that the 
existential quantifier be immediately above 
the clause into which there is to be inserted. 
If an existential context is possible then there 
is no definiteness. 
18. there are two men in the garden 
*there are the two men in the garden 
(no possibility of there) 
Definiteness in an extensional model it is a 
function from properties into sets of 
properties. 
NPs in the IL representation used in MT are 
flat tree structures with decorations 
containing sets of attribute-value pairs. 
These attributes were sketched above. 
Feature configurations are equivalent to well 
formed formulas after lambda conversion, so 
that this approach can be used in a NL 
understanding system as well. 
Below a rough outline of a PSR for 
nominals: 
1: NP <--. 2: DET + 3:NO 
conditions: ...... 
structure: 
1: nu = 2:nu 
1 :Quant --= !VRaspect & Qtype of 
determiner 
1:Generics = !VEtense & VB.aspect (i) 
1 :semtype= = !NO.semtype 
1 :definiteness = !{DET, NO.Compl} (ii) 
l:comparative = !NO.compl !PREP (iii) 
• (i) if tense = present and aspect = 
habitual and NO is plural then the 
quantification is distributional; if the 
tense is present but NO is singular then 
the quantification of the det is collective 
• (ii) if the determiner type is possessive or 
demonstrative, or the NP contains 
genitives, relative clauses, etc..the NP is 
definite 
• (iii) this feature keeps track for 
predicative constructions like: 
19. he works as a teacher 
vs. trabaja como 0 maestro 
The final value for definiteness and 
quantification is the result of the above 
mentioned calculus that is percolated to the NP 
nodes. 
4. Generators for definiteness 
It is straightforward to show that for any 
definiteness configuration there is a determiner 
assignment in a specific language. The 
generation grammar is the inverse process of 
the analysis. Below some rules: 
Rule 1 : 
if 
definiteness = T 
NU = Sg 
semtype = ~ABS 
focus = T 
then 
Ger. = definite DET + NO 
span = demonstrative DET + NO 
fran = demonstrative DET + NP 
"esta bolsa vs. die Trite" 
Rule 2: 
if 
generics = T 
267 
NU = PL 
Quant = distrib. 
then 
Germ. = no DET + NP pl 
eng. = demonstrative DET + NO pl 
spn. = Definite DET + NO pl 
"los hombres no lloran" 
Rule 3: 
if 
generics = T 
NU = sg 
then 
Germ. = Indefinite DET + NO sg 
Spn. = Indefinite DET + NO sg 
"un hombre no llora" 
The difference between rules 2 and 3 is that 
3 is a statement about the fact that something 
having the property of being a man does not 
CD' (is not allowed to cry). 
5. Summary 
A computational system must anaIyze and 
generate definiteness even for those cases 
where there is no unique referent to look for. 
Here I assume that definiteness is not the 
only function of definite determiners, and 
that it would be inadequate to use a brute 
force approach in looking for either 
uniqueness of referents nor uniqueness of 
roles as proposed by Bosch & Geurts 89. 
Definiteness as a semantic value of definite 
determiners is one of its possible values (the 
logical value of the quantifier "only"); more 
common values are distributional, collective 
quantification, or a surface phenomena 
within generic constructions, or in 
connection with the predicative value. 
Definiteness is unmarked when placed 
before proper names, and becomes an 
existential quantifier when accompanied by 
an abstract noun. NPs in directionals and 
locals usually have a pragmatic value (for 
more detail (Loebner 1985)). 
Romance languages have a higher use of 
"definiteness" than germanic languages. While 
the latter use no article the former use them; 
while the latter use definite determiners the 
former use demonstratives. 
The internal value, however, remains the same. 
Compare: 
20. Freedom is essential 
vs. la libertad es esencial 
21. Love does not exist 
vs. el amor no existe 
Or is "love" and "freedom" more "definite" in 
Romance languages? 

References 

Alonso,A . 1961 : Estilistica y gramatica del 
articulo en espa?ol,en Estudios LInguisticos. 
Madrid 

Barwise,J & R.Cooper: 1981. Generalised 
Quantifiers and Natural Language. Linguistics 
and Philosphy, 4. p.159-219 

Benthem van ,J &A.ter Meulen (eds) 
Generalised Quantifiers. Dordrecht 

Bosch,R & B.Geurts 1989: Proccessing 
definite NPs. IBM. Lilog Project. 

Gardenfors, P (ed) 1987: Generalised 
Quantifiers. Reidel.Dordrecht 

Geurts, Bart 1985 : Generics, in .:Journal of 
Semantics. vol.4, p.247 

Grimm, H.J 1987: Lexikon 
Artikelgebrauch. Leipzig 
zum 

Keenan,E. & J.Stasvi 1986: A semantic 
characterisation of Natural Language 
Determiners in: Linguistics and Philosophy, 4 
p.159-219 

Loebner Sebastian 1985 : Definites. In.Journal 
of Semantics. vol.4.p.279 

MacCawley,J.D 1981 : Everything that 
linguistis have always wanted to know about 
logic. Univ. Chicago Press 

Peterson, Ph. 1979 : On the logic of few, 
many and most. Notre Dame Journal of 
Formal Logic 20. p. 155o179 

Weinl:ich, H. 1976 : Sprache in Texten. Klett. 
Stuttgart 
