THE IMPLEMENTED HUMAN INTERPRETER AS A DATABASE
Gábor Alberti and Márton Károly
Department of Linguistics, University of Pécs, Pécs, Hungary
Keywords: Discourse semantics, Interpretation, Discourse analysis, Logical programming.
Abstract: In this paper we continue to publish the results of our work with eALIS, a new “post-Montagovian”
discourse semantic theory, demonstrating its functioning on a few classical semantic problems. We retain
mathematical exactness while simultaneously applying cognitive paradigm. Previously we wrote study
programs, for testing purposes, now we are building a (lexical) Prolog fact database. Although we are
implementing the grammatical analysis, too, it is important to note here that the whole process of
grammatical (phonological, morphological and syntactic) analysis is practically included in only one () of
the four basic functions while the other three describe various parts of semantics.
1 INTRODUCTION
The long-term goal of the eALIS project is
automated discourse analysis, preferably including
effective information extraction, text summarization,
machine translation (e.g. the translation of European
legal text) and analyzing NL queries.
To do all this, we provide the eALIS theory of
discourse semantics (Alberti et al., 2010b) as a
simultaneous extension of SDRT (Asher–Lascarides
2003) and LDRT (Alberti, 2000), which can be
regarded as a compromise between the total
rejection of formalization by cognitive semanticists
and the different kinds of higher order intensional
logic applied by formal semanticists, say, Pollard
(2007) or Kamp et al. (2011). Here, we provide a
basic data structure based on Prolog facts on which
the implementation of eALIS will be based.
2 EALIS
2.1 eALIS, Reciprocal and Lifelong
Interpretation System
We hereby summarize the main concepts of eALIS
which were introduced in Alberti et al. 2010b as a
“post-Montagovian” (Kampian) theory concerning
the formal interpretation of sentences constituting
coherent discourses (see also Asher–Lascarides
2003). eALIS has a LIfelong model of lexical,
interpersonal (“REciprocal”) and cultural knowledge
of interpreters, too. “LIfelong” also means that the
DRS-like structures of eALIS are continuously
being built, being able to reach an arbitrary degree
of complexity – much like the structures of “world
knowledge” in the human mind itself.
eALIS reconciles three objectives of formal
semantics: the exact formal basis itself (Montague’s
Thesis), compositionality (postulating the existence
of a homomorphism from syntax to semantics) and,
most importantly, eALIS’ own “discourse repre-
sentationalism”. The main difference between
Kampian (and its extensions) and eALIS’ DRT is
that the outer world, the states of the interpreters’
mind and the information structure of discourses
(which are also stored in the interpreter’s mind,
becoming part of it) are described by a unified
model (see detailed def. in Alberti 2011a:139-149).
To do this, the infons of Seligman–Moss (1997:245)
are used to describe information structure from the
view of the external world and the vertical hierarchy
of the so-called worldlets which are parts of the
internal (mental) world. Shortly, the information
state of any interpreter is depicted by eALIS’ own
DRS boxes. A part of the external world is projected
into them so that each interpreter’s unique
perception and knowledge about the world – or parts
of these – is represented by an internal world with
multiple worldlets. Embedded worldlets are used to
describe the hierarchy of fictiveness and
interpersonal knowledge (see Figure 4), too.
The principal theoretical difference between the
Montagovian semantics, the Kampian DRT and
eALIS is shown here:
379
Alberti G. and Károly M..
THE IMPLEMENTED HUMAN INTERPRETER AS A DATABASE.
DOI: 10.5220/0003639903790385
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 379-385
ISBN: 978-989-8425-80-5
Copyright
c
 2011 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: Montagovian model, (S)DRT and eALIS.
2.2 Explanation of eALIS’ Definition
To provide sufficient theoretical background to our
database, we give a short explanation of eALIS’
world model (Alberti, 2011a:139-149), see also pp.
151-179 for the detailed definition of its dynamic
and static interpretation), providing some additional
information to Alberti et al. 2010b. These structures
were (partially) implemented as data structures (as
shown in Section 4), the four internal functions form
the starting point of a future eALIS program.
U and the External World. eALIS’ world model
is  = U, W
0
, W where |U|=
0
, W
0
= U
0
, T, S, I,
D, , A, W  IT
m
 U[i], [i,t]

, [i,t]

, [i,t]

,
[i,t]

; T=T,, S=S,, I=I,, D=D,. First
of all we note that all sets are finite or countable, e.g.
time and space are based on Q. Secondly, U
0
(and
U) also contains elements that belong to the
structure of eALIS itself – because they, too, can
be referred. The set of possible time intervals (T) is
isomorphic with the set of Q’s intervals). Spatial
entities (S  Q
3
), interpreters (I) and physical
linguistic signs (D) are also included.
The relations over T, S and I (marked by , 
and ) are arbitrary, but  must contain Dis ( D as
a unary relation, containing the discourses as
linguistic entities) and Morph ( D
2
where d’, d 
Morph if d’ is a morph (in the linguistic sense) of the
discourse d. The set of morphs of any discourse
must be linearly ordered. The set of core relations
(  TU
0
*) of the external world must have a
compulsory
PERCEIVE element used by the infon
(Seligman–Moss 1997) ι (A) = 
PERCEIVE, t(T),
i(I), j(I), d(D), s(S).
U[i] and the Internal Worlds.
eALIS’ interpre-
tation actually defines the interpretation of discourse
d in ι relative to the external world W
0
and the
internal world W[i,t], the latter representing the
information state of interpreter i at the moment t.
We also suppose that i!t’,t” t’t” | t<t’t>t” 
[i,t], [i,t], [i,t], [i,t] (i’s internal functions) are
empty with t’ and t” being the time of birth and
death of interpreter i (~”lifelong” property).
The elements of U[i], where ij i,jI, ij 
U[i]U[j]=, are called referents. The referents
may be anchored to each other (reversibly; not
necessarily identified because identifying requires
accommodation). Any relations between the
referents are defined by the internal functions.
The two arguments of the eventuality function
[i,t]

: U[i]  U[i] are the eventual label and the
eventual referent. The label , is an ordered pair,
too. A few examples of  are: [i,t] (Pred,,e)=p
(resulting a predicate referent); [i,t] (Temp,,e)=t
(resulting t which is the temporal referent);
[i,t](Arg,
k
,e)=r
k
(resulting an argument as a
referent) etc. The first component of  is actually a
linguistic category, the second one represents a class
within it. Here,  is a (linguistic) category of event
structure,  describes the relation between speech
time, referred time and the event structure (e.g.
▼InRes describes the English present perfect tense,
▼ meaning t
ref
=t
speech
, and t
ref
is in the result stage of
e), while 
k
marks either the grammatical case, or,
according to Alberti–Kilián 2010, a generalized
thematic role in a polarized chain of influence.
In Figure 4, a different (simplified, timeless,
eventuality-based) notation is used. For example,
e
1
’: p
kill
r
witch
r
cow
means (Pred,,e
1
’) = p
kill
(functional notation); (Arg,
1
,e
1
’)=r
witch
and
(Arg,
2
, e
1
’)=r
cow
respectively, where  describes
the event structure of to kill (no preparation/result
phase, only cumulative phase: John was killed in a
few seconds.), 
1
is “Agent”, 
2
is “Patient”
according to Alberti–Kilián 2010, see Section 4.
The anchoring function works in a similar way:
an example is (Ant,Top,Gen,r
she
)=r
witch
. The
first element of  is the type of anchoring:
Arg
ument, Predicate, Adjunct, Antecedent,
Out
wards (meaning that the result of  is not a
referent [
i
U[i]] but an external entity of U
0
). The
second element of  is an ordered n-tuple
representing the (language dependent) factors that
legitimize the anchoring: Num
ber, Gender, Topic
retaining, Humanity etc.
The level function marks the worldlet which
contains the referent. Only fictive referents are
projected by , the root referents are not (but they
can be anchored “outwards” by means of  instead).
In general, r’=(
k
,
k
,i
k
,
k
,r) where 
k
{ [level-
changing feature, linguistic subordination], [level-
keeping, linguistic co-ordination]}{., ?, ![modal
markers ~kinds of sentences]}{supp, cons,
bel[ieve]
n
, int[ent]
n
, des[ire]
n
, dream, see, hear,
elab[orate], exp, nar[rative], back, conj, disj etc.
[modal markers marking the source of information]}
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
380

(not all combinations of 
k
are [linguistically]
possible). 
k
T
m
is the temporal component of the
level label, i
k
is the referent of the direct owner of r
where the transitive closure of (i
k
) anchors i
k
to a
certain iI interpreter. Finally, 
k
{+, 0, –} is the
polarity of the level label (positive, neutral,
negative), marking true, don’t know and false.
The level label system is defined in a such way
that one can reach the root referent i* by repeatedly
applying : r”=(
k-1
,
k-1
,i
k-1
,
k-1
,r’) … r
(s)
=(
k-
s+1
,
k-s+1
,i
k-s+1
,
k-s+1
,r
(s-1)
)… r
(k)
=(
1
,
1
,i
1
,
1
,r
(k-1)
)=
i* where (Out,…, i*)=i. The  series of the 
labels used in this formula is called the worldlet
index of r ( is empty if r is a root referent) and the
set of referents with a certain  worldlet index is
called a worldlet. For the root worldlet of any i
interpreter,  is empty. This results in Kampian DRS
boxes with SDRT-like (Asher–Lascarides 2003)
rhetorical information (Figure 4).
The cursor function is [i,t]: K  U[i]. K is a
finite set of pre-defined cursor labels and must
contain the following elements (among others):
- [i,t](Now/Here/Ego)=t
/s/i, (Out,…, t/
s
/i)=t/s(S)/i
- [i,t](ThisWay)=e
(speech situation)
- [i,t](Then/There/Eve)=t
’/s’/e’(the referred
time/spatial entity/eventuality)
- [i,t](You)=j
The temporary states of these four internal functions
above an interpreter’s internal universe serve as her
“agent model” in the process of (static and dynamic)
interpretation.
The interpretation of any “perceived” discourse
can be defined in our model relative to an external
world W
0
and internal world W[i,t]. (summary:
(Alberti et al. 2011b)), (details: (Alberti 2011a)).
Since the data structure that we intend to describe is
a mapping of the actual world model, and most
importantly, its four internal functions, we do not go
into the depths of the mathematical definition of the
interpretation. Indeed, we summarize the process of
the actual discourse analysis later on.
2.3 Illustration of the Apparatus of
eALIS
Let us consider a sentence (1a) with two meanings.
The figure in the center in (1d) shows the lexical
items which have been retrieved by the words of the
sentence (eventuality function ). The lexical
items (except for those of the piano) are shown as
eventualities more or less in the style of DRT. The
formulae in (1b-c) express the difference between
the collective and the distributive reading also in
DRT-style. The two figures with arrows show two
different anchorings () of referents belonging to
arguments and eventualities to each other. And what
is defined by these two anchorings exactly results in
the two formulae in (1b-c).
Example 1: Readings and anchoring function .
(1a) The boys lifted the piano.
(1b) C
OLLECTIVE READING:
[
e4
[
e5=e41
[
e2=e51
element r
21
r
22
][
e1=e52
boy r
21
]]
[
e42=e3
lifted r
22
r
32
]]
‘if referent r
21
is an element of group r
22
, it is a boy
(hence, r
22
is a group of boys); and this group r
22
lifted the piano (r
32
)’
(1c) D
ISTRIBUTIVE READING:
[
e5
[
e2=e51
element r
21
r
22
][
e4=e52
[
e1=e41
boy r
21
] 
[
e42=e3
lifted r
21
r
32
]]
‘if referent r
21
is an element of group r
22
, it is a boy
who lifted the piano (r
32
) (hence, r
22
is a group of
boys, each of them lifted the piano)’
(1d)  anchoring in the case of the collective
reading () / the distributive reading ():
Figure 2: The two possible readings of (1a). The lines
mark the anchoring function .
3 WHERE ARE THE POSSIBLE
WORLDS?
3.1 The Granularity Problem as a
Further Argument for eALIS
Let our starting point be Pollard’s criticism on the
mainstream Kripke/Montague-inspired possible-
worlds (PW) semantics: it is “a framework known to
have dubious foundations” (Pollard, 2007:1) because
of the granularity problem (2), among other
deficiencies (3); and hence “the idea of taking
worlds as a primitive of semantic theory is a serious
misstep” (Pollard, 2007:33). Another stubborn
problem in mainstream NL semantics concerns
distinct accessibility of certain referents in logically
equivalent sentences (4). In this area a promising
solution is offered by DRT, but at the cost (see (5))
e
1
: boy r
1
e
2
: element r
21
r
22
e
3
: lifted r
31
r
32
e
4
:  e
41
e
42
e
5
:  e
51
e
52
e
1
: boy r
1
e
2
: element r
21
r
22
e
3
: lifted r
31
r
32
e
4
:  e
41
e
42
e
5
:  e
51
e
52
e
1
: boy r
1
e
2
: elemen t r
21
r
22
e
3
: lifted r
31
r
32
e
4
:  e
41
e
42
e
5
:  e
51
e
52
THE IMPLEMENTED HUMAN INTERPRETER AS A DATABASE
381

of introducing an extra level of representation, that
of discourse structure (4c). Nor does (the Higher
Order Intensional Dynamic Logic of) current DRT
exceed the PW approach criticized above; although –
we claim – there would be an obvious way of using
(gigantic) DRSs as lifelong representations of
interpreters’ information states and embedded DRS
boxes (consisting of propositions) as PW-like ilks.
Following this way we have elaborated a “ReAL”
interpretation system (6a), which provides
straightforward solutions (6b-f) to (2-5), including
even the Hob-Nob problem (4d).
The essence of the granularity problem is that
having the same reference / meaning (see (2)) is not
a sufficient condition to allow replacement of one
name / sentence for another in a larger expression.
The other problem (Pollard, 2007:30-31) lies in “the
standard view, [viz.] reference is compositional”;
that is why “Frege had to resort to claiming that
utterances of sentences in certain contexts [e.g. S3 in
(3)] had the customary sense as the reference,”
which requires “sleight of hand.”
Example 2: The Granularity Problem.
(2) The ancients realized that [Hesperus was
Hesperus]
S1
/ [Hesperus was Phosphorus]
S2
.
Example 3: The problem of using the customary sense as
the reference (see fn36, Pollard 2007:31).
(3) [(Justin Timberlake knows that) [Paris Hilton
believes [snow is white]
S3
]
S2
]
S1
.
We follow Pollard in assuming that “worlds are
constructed from propositions ..., and not the other
way around” (Pollard, 2007:34), but intend to work
out this approach in a DRT-based framework in
order to account for phenomena concerning referent
accessibility (4a-d), at the same time. We claim,
however, that our system is devoid of DRT’s “extra
level” problem (5).
Example 4: The problem of referent accessibility.
(4a) [[A delegate arrived.]
S1
She registered.]
Discourse1
/
[[It is not the case that every delegate failed to
arrive.]
S2
*She registered.]
Discourse2
(4b) S1 and S2 are logically equivalent: x 
x
(4c) The representation of Discourse1/ Discourse2
in DRT: x “enclosed” is not accessible to y
(4d) Hob believes that a witch has killed Cob’s cow
and Nob thinks that she has blighted Bob’s sow.
(5) An Argument against DRT: DRS is an
illegitimate extra level of representation between
syntactic structure and the model of world in the
course of interpretation (Groenendijk–Stokhof,
1991).
Figure 3: Illustration to (4a).
3.2 eALIS as an Epistemic
Multi-agent System
eALIS is based on the idea that (gigantic) DRS-
like structures, set W, are suitable for serving as
lifelong representations of information states of
interpreters, i.e. the “agents” getting information.
Furthermore, as stated in Alberti et al. 2010b, both
static evalutation (Tru) and dynamic interpretation
(Dyn) can be carried out. Labeled tree systems of
worldlets can serve as parts of the world model (6f):
the interpretation of modal sentences is to be based
on certain worldlets instead of W (6c).
Summary 6: ReALIS as an epistemic multi-agent system
and its features.
(6a)  = W
o
, W, Dyn, Tru
(6b) (2): co-anchoring ((r’)=(r”)) does not
necessarily imply the identity of referents (r’ and r”);
identifying (and concluding, whose amount does
matter) requires accommodation
(6c) (3): S1 is true if S3’s eventuality referent can
be found in an appropriately labeled worldlet of JT
(containing JT’s thoughts on PH’s beliefs)
(6e) (4a-b): certain referents may be “enclosed” in
the worldlet structure
(6f) (4d): see Figure 4; what provides more
freedom in constructing worldlet structures in
eALIS is that DRT’s box structure depends on
logical factors whereas the worldlet structure is
affected by pragmatic factors (as well)
(6g) (5): DRS-like representations form no “extra
level” but turn into parts of the
eALIS world
model (enriched with descriptions of interpreters
themselves)
The content of worldlets can be enriched by (chiefly
pragmatics-dependent) accommodation, too:
identifying referents (6b, e) or drawing logical
inferences can be bound to pragmatic conditions.
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
382

Let the last figure serve as an illustration. The
representation to the left expresses the pure semantic
content of (4d). According to this, Hob’s witch is not
yet accessible to the referent of she. But then we can
consider some pragmatic factor, discussed by Zeevat
(2005:549) as follows: “Hob may have told Nob
about his belief [repr. to the right], [or] there may
be a rumour in the village about a witch that has
played a causal role in the formation of Hob’s and
Nob’s beliefs [repr. in the middle]...”
Figure 4: The Hob-Nob sentences in the eALIS model.
4 IMPLEMENTATION
4.1 Basic Principles of the Data
Structure and the Function 
In eALIS, the decomposition of syntax plays a
crucial role (Alberti, 2011b). Syntactic and semantic
information are stored together in the core lexicon or
produced by the lexical rules (Alberti et al., 2010a),
according to our ‘totally lexicalist’ approach. To
handle all these data, we use a Prolog engine which
is also able to handle non-determinism (see also
Figure 2: the sentence can be disambiguated after
having some more sentences of the discourse). For
now, theoretical consistency in the implementation
of the
eALIS model is more important – but in the
long run, the theory itself may be refined to improve
the speed of our program and/or reduce its
computational complexity: as the human brain has
its own limits revealed by psycholinguists, our
program does not need to surpass them either. (One
such limit is the depth of discourse analysis.)
Here we summarize the practical Prolog
equivalents of our concepts as parts of a (future)
Prolog fact database. In the program itself we
excessively use
assert and retract to manage it.
We note here that for now, we are working primarily
on Hungarian (a highly agglutinative language with
free word order) and English.
- Input: Step 1: morphologically tagged SL text;
Step 2: untagged SL text
- Output: discourse representation, Step 3: make
bidirectional use of the program possible (for e.g.
machine translation).
- Logic: linguistic/semantic data  Prolog facts
(Horn clauses)
o modal operators  -labels
o disjunction  different -labels
- Reversing basic Prolog mechanism: linguistic
data = facts, derivation of possible semantics = rules
(forward chaining)
- Lexicon: Prolog facts (later SQL), totally lexicalist
approach (Alberti et al. 2010a-b)
- Referents: variables/dynamic facts, values:
instances of any class (unique entities are also
regarded as classes, e.g. Paris Hilton)
- Problem: implementing second-order logic with
first-order tools
o reification: predicates as data
o use of extra-logical methods to extract predicates
if needed
Predicate Referents can only be handled by
reification. An example of a predicate template:
semrolefeat(111,’Agent’).
semrolefeat(112,’Patient’).
semrolefeat(113,’Beneficiary’).
ref
(20,p,‘R111^R112^R113^give(R111,
R112,R113)’).
The verb to give is reified and stored as data (string)
but we can rebuild it by using extra-logical
techniques. The roles of Agent, Patient and
Beneficiary keep their linguistic meanings. In
the case of active transitive verbs, Agent becomes
subject. A Hungarian exception: Mari
Mary
meg
Perf
-
THE IMPLEMENTED HUMAN INTERPRETER AS A DATABASE
383

híz
grow-fat
-ott
Past
. (Mary has grown fat.) The
intransitive verb meghízik has only one argument,
the subject, which is Patient.
Eventual Referents can be regarded as instantiated
predicate referents: Peter gave Mary the book. And
that makes me angry.
Analysis. We suppose that or input is already
morphologically analyzed and POS-tagged. So we
can (Step 1) assign potential referents to all nouns
(r), verbs (e), adverbs (e.g. t), adjectives (e, p, r) and
pronouns (typeless). Then (Step 2) we can determine
-relations by doing syntactical analysis. Here are
some facts that should be asserted during the
analysis of the two sentences:
ref(...,p,...). %see give above
ref(11,r,‘Peter’). ref(12,r,‘Mary’).
ref(13,r,‘book’).
%--calculate evref here
ref
(21,e,‘give(11,12,13)’).
ref(22,x,‘that’)].%type is not set
...
alpha(22,21). %see below
ref(23,e
,make_angry(22,EGO)).
Conclusion: make_angry(21,EGO).
Temporal Referents are handled by time intervals:
a week ago is described by date(past,
(0,0,1,0,0,0,0),(0,0,0,1,0,0,0)). The
first 1 is in the field weeks while the second one
marks the ‘precision’ of the linguistic expression –
about one day in the English language.
now is
regarded as a special temporal referent.
The words and their default morphology and parts
of their syntax and semantics are stored in the core
lexicon. This is basically the same as stated in
(Alberti et al., 2003) and (Alberti–Kilián, 2010) but
each feature (POS, type of morpheme etc.) is
described by a different fact. Lexical units search for
each other by offering and demanding features.
lexfeatval(301,3,’Nom-Subj’).
lexfeat(’Bob’,301,+7).
This means that the ‘Bob’ offers a grammatical case
of ‘NOM-Subject’ with rank +7. The same applies to
POS, referentiality/definiteness etc. If morphological
inflection takes place, the default features are
overridden: in Hungarian, Bob-ot
Acc
will get a new
lexfeat(’Bob’,302,+1) feature, which has a
“stronger” rank of +1.
Now we take a look into the “demand” part of
the database. Verbs search for their arguments:
semdemand(’GIVE’,’Agent’,1).
%1 would be 2,3 etc. for more args
synsem(’GIVE’,’en’,’give’).
synsem(’GIVE’,’hu’,’ad’). %etc.
%Note: syndemand is for English
syndemand(’give’,’Noun’,1,+2).
syndemand(’give’,’Nom-Subj’,1,+2).
syndemand(’give’,’Nei-3’,1,+2).
%1 is the same as in semdemand
%strings become ID’s in the real DB
As it is shown, the semantic field GIVE has several
language-dependent syntactical representations.
Here we only show the facts for the agent (subject)
of the English verb give (ad in Hungarian). Its
patient and beneficiary can be represented in a
similar way.
Nei-3 is a neighborhood rank of –3
meaning that Argument 1 must take place before the
verb with rank 3 but this requirement can be
overridden with any other fact with rank e.g. +1.
4.2 The Anchoring Function 
The function  anchors referents to each other.
Anchored referents are supposed to refer to the same
thing but anchoring is not necessarily permanent.
The relations behind  must be defined by a
background ontology: the legitimizing factors are
often extra-grammatical, based only on semantic
categories (e.g. when a parrot is mentioned as a
bird). In subsection 4.1 (
alpha(22,21)) this is not
the case: that can have a function of referring to the
eventuality – formally, (Ant,Eve,r
22
)=e
21
.
However,
alpha is restricted to mark antecedents
(Ant) in its present form without labels.
Multiple grammatical and semantic factors are
involved in  and not all analyses are correct. So we
plan to use a cost/weight metric for  to measure
discourse coherence. If it surpasses a certain limit
(which is set by the user), the discourse is
considered incoherent and ill-formed. (The same
applies for tolerance to grammatical errors and .)
4.3 The Level Function 
 assigns all referents to the specified worldlet(s) of
eALIS and describes mood and rhetorical
relations. As defined, all referents must carry all
level labels – as shown here:
Example 5: Storing the level labels of referents.
(6) If only I
ego
had
e2
a car
r1
! Sue
r3
would drive
e5
it
r4
, too. But now I
ego
only have
e6
a motorcycle
r7
.
ref(1,r,‘car’).
ref(2,e,‘have(EGO,1)’).
ref(3,r,‘Sue’). ref(4,r,‘it’).
ref(5,e,‘drive(3,4)’).
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
384

ref(6,r,‘motorcycle’).
ref(7,e,‘have(EGO,6)’). alpha(5,2).
lambda(1,[[sub,now,des,0,101,+1]]).
lambda(2,[[sub,now,des,0,101,+1]]).
lambda(3,[]).%Sue is a real entity
lambda(4,[[sub,now,des,0,101,+1]]).
lambda(5,[[sub,now,des,0,101,+1]]).
lambda(6,[]). lambda(7,[]).
The second argument of lambda is the above-
mentioned list of level labels. Each level label has
six parameters according to the definition: co/sub
marks how the actual worldlet is related to the next
one in the label chain, the first number (0 by default)
marks the level of belief etc. (where applicable), 101
is a placeholder for the interpreter’s ID, and, finally,
the polarity is marked by +1/0/–1 (believed etc. to be
true, “don’t know” or false).
 is set inherently by verbs (e.g. to believe, to
think, to desire), adjectives (alleged) and some other
words (e.g. negative words) and morphemes (modal
markers).
5 CONCLUSIONS
A full implementation of eALIS is yet to come but
our progress and partial results are going to be
published continuously. We demonstrated the
functioning of the eALIS model on a few classical
semantic problems, arguing that the cognitive
paradigm does not necessarily exclude mathematical
exactness. By now, we fixed most of the data
formats of the lexicon and database.
Although we plan to use external ontologies
and/or dictionaries, their integration into the
eALIS software is only the first step. Since
eALIS is a lifelong [self reference] interpretation
system, our database is designed to build itself (by
assertions) when analyzing discourses. So the
values of the three base functions – ,  and  (and
the cursor function  whose exact functioning is yet
to be determined) form an integrate part of the
database even if the relations discussed here are
analyzed on-the-fly.
ACKNOWLEDGEMENTS
We are grateful to SROP-4.2.1.B-10/2/KONV/2010/
KONV-2010-0002 (Developing Competitiveness of
Universities in the Southern Transdanubian Region)
for their contribution to our costs at KEOD 2011.
REFERENCES
Alberti, G. 2000. Lifelong Discourse Representation
Structures. Gothenburg Papers in Computational
Linguistics 00-5, 13-20.
Alberti, G., Balogh, K., Kleiber, J., Viszket, A. 2003.
Total Lexicalism and GASGrammars: A Direct Way
to Semantics. In Gelbukh, A. ed.: Proceedings of
CICLing 2003, Springer, Mexico City. 37-48.
Alberti, G. 2011a. ReALIS [in Hungarian]. Academic
Press, Budapest. To appear.
Alberti, G. 2011b. ReALIS, or the decomposition of
syntax [in Hungarian]. In: General Studies of
Linguistics. Academic Press, Budapest. To appear.
Alberti, G., Károly, M., Kleiber, J., 2010a. From
Sentences to Scope Relations and Backward. In Sharp,
B., Zock, M. eds.: Natural Language Processing and
Cognitive Science. Proceedings of NLPCS 2010.
SciTePress. Funchal, Madeira. 100-111.
Alberti, G., Károly, M., Kleiber, J., 2010b. The ReALIS
model of human interpreters and its application in
computational linguistics. In Cordeiro, J., Virvou, M.,
Shiskov, B. eds.: Proceedings of the 5th International
Conference on Software and Data Technologies Vol.
2. SciTePress. Funchal, Madeira. 468-474.
Alberti, G., Kilián, I., 2010. Polarized chains of influence
instead of argument structure lists – or the function 
of eALIS [in Hungarian]. In Proceedings of the 7
th
Hungarian Conference on Computational Linguistics.
Univ. of Szeged. Szeged. 113-126.
Asher, N., Lascarides, A., 2003. Logics of Conversation,
Cambridge University Press. Cambridge.
Dekker, P., 1999. Coreference and Representationalism. In
von Heusinger, K., Egli, U. eds.: Reference and
Anaphoric Relations. Kluwer. Dordrecht. 287-310.
Groenendijk, J., Stokhof, M., 1991. Dynamic Predicate
Logic. In Linguistics and Philosophy Vol. 14. Kluwer.
Dordrecht. 39-100.
Kamp, H., van Genabith, J., Reyle, U., 2011. Discourse
Representation Theory. In Handbook of Philosophical
Logic, Vol. 15. Springer. Heidelberg. 125-394.
Pollard, C., 2007. Hyperintensions. ESSLLI 2007.
http://www.cs.tcd.ie/esslli2007.
Reyle, U., 1993. Dealing with Ambiguities by
Underspecification. Semantics 10, 123-179.
Selingman, J., Moss, L. S., 1997. Situation Theory. In van
Benthem, J., ter Meulen, A. eds.: Handbook of Logic
and Language. MIT Press. Cambridge. 239-309.
Szabolcsi, A. ed., 1997. Ways of Scope Talking. Kluwer.
Dordrecht.
Zeevat, H. 2005. Overlaying Contexts of Interpretation. In
Mayer, E., Bary, C., Huitink, J. eds.: Proceedings of
SuB 9. Radboud University Nijmegen. Nijmegen. 538-
552.
THE IMPLEMENTED HUMAN INTERPRETER AS A DATABASE
385