PROXIMAL AND DISTAL DEMONSTRATIVES IN
DUTCH SPOKEN DIALOGUES
Robbert-Jan Beun and Paul Piwek
1
DENK: BASIC MODEL
Agent 2
Domain
Agent 1
Physical interaction
Symbolic interaction
Computer systemUser
2
SEARCH SPACES IN DENK
User:
private
beliefs
shared
beliefs
goals
discourse
‘Referring
expression’
C
o
m
p
u
t
e
r
pending
3
SEARCH SPACE FOR REFERENTS
Driven by ‘accessibility markers’
def articles, demonstratives, pronouns, …
A: bla bla bla bla bla bla
The driver drinks. The policeman observes him/himself.
cognitive state
(beliefs, goals, …)
4
PRIORITY SPACE FOR REFERENTS
A: Is this block heavy?
B: Yes
A1: Remove it/?that/?this!
A2: Remove that/?this one!
A3: Remove the/that/?this block!
search
space
accessibility marker
non-refl.
pronouns
refl.
pronouns
demon-
stratives
def.
article
sentence 2 1 3 4
dialogue 1 - 2 2
private - - - -
shared - - - 3
domain 3 - 1 1
5
Would we be able to refine the rules that guide
the search and the generation process for
demonstrative noun phrases, in particular the
difference between proximals and distals?
6
DEMONSTRATIVES
TRADITIONAL VIEWPOINTS
Demonstratives indicate relative distance of a referent in the
speech situation
Deictic center (origo): usually the speaker
Reference to nearby-faraway distance:
proximal: ‘dit/deze’ (‘this/these’)
distal: ‘dat/die’ (‘that/those’) faraway
54% of the languages express a two-way contrast
Distance- and person-oriented systems
Anaphoric use is derived from situational use
7
LANGUAGE AND PERCEPTION (KEMMERER, 1999)
Visual brain representation peripersonal space (within arm reach)
proximal?
extrapersonal space (beyond arm reach)
distal?
Conflicts three-way (or more) contrast languages
many other uses: This block is smaller than that block
non-situational use?
Essential difference between perceptual representations and abstract linguistic notions, however …
8
OTHER PROPOSALS
Based on notions such as:
importance, focus, given/new,
background/foreground, force to seek the
antecedent, intensity of indicating, accessibility,
prominence, distance, (degree of) attention,
familiarity, shared knowledge, contrast,
presupposed vs. predicted, ...
Hard to quantify!
9
Let’s take an example
10
TWO NOTIONS IN HUMAN INFORMATION
PROCESSING
Accessibility
‘the ease with which particular mental content comes to mind’ (Kahneman, 2003)
attention, perception, beliefs, environment, task, …
Importance
‘perceived relevance of actions, events, facts and objects for the attainment of goals/desires’
tasks, goals, preconditions, priority, intentions, …
11
If accessibility is low, then give more
force
If importance is high, then give more
force
Applied to demonstratives (Kirsner,
1979)
proximals strong indicating
distals neutral indicating12
INTENSITY OF INDICATING
FINDINGS BY KIRSNER (1979)
Accessibility
proximals used over longer distances
proximals are related with low accessibility
Importance
proximals more often used:
to refer to humans, (named) individuals
to refer to individual referents (as opposed to plural)
in subject position
proximals are related with high importance
13
CORPUS 1 (PIWEK, BEUN AND CREMERS, 2008, JOP 40)
Dutch dialogue
Nr. of words: ~5000
Domain of discourse: blocks world
Channel: spoken
Goal: make building like example
Builder: B, Instructor: I
14
BASIC SETTING
Agent 2Agent 1
Domain
15
EXAMPLE (ORIGINAL IN DUTCH)
I: Well J.
B: Yes
I: Let’s see …uh, do you have a red square?
B: Yes
I: Well, for a start, you have to … put that on the horizontal … beam v- uh, the bleu one, two by six
B: yes
I: in the front
B: over here?
I: yes, a little more to the front side
B: like this?
I: yes, like that, that is the first change, and then, uh, let’s see, what is the old one … (2.0) yes, then you have to put that block, in the middle, that one with yellow, that one has to be removed
B: this?
I: take it away
16
HYPOTHESIS
The use of a proximal vs. distal is related to the
intensity of indicating:
If accessibility is low use proximal
If importance is high use proximal
Else use distal
Accessibility
low high
Importanc
e
low proxima
l
distal
high proxima
l
proxima
l17
RESULTS
Accessibility and importance in terms of focus of
attention and task at hand (manipulation)
If low accessibility then proximal
yes (χ²=6.76, p<0.01)
If high importance then proximal
no significant result
18
SOME REMARKS
We investigated only situational reference
i.e. first reference to physical objects
Proximals were always (except one) accompanied
by pointing
Operational criteria of accessibility and importance
are disputable (also in Kirsner’s proposal)
19
SO, …
Intuitively a proximal refers to nearby, more
accessible information, but …
Observations showed that proximals refer to less
accessible information
reference over longer distances (Kirsner)
reference to objects that were not directly in the focus of
the speaker’s attention (Piwek, et al)
20
CORPUS 2
Dutch dialogue
Domain of discourse: computer, telephone, internet
Channel: spoken, phone
Goal: solving problems with computer
Client: C, Helpdesk agent: A
Nr. of words: ~30000
21
EXAMPLE (ORIGINAL IN DUTCH)
A: I understand from a colleague that you have a problem with
your phone
C: yes, I am now on the phone for almost one hour and I want …
A: hmmh
C: and I am phoning that one on the phone, a P. or a P. …
A: hmmh
C: from the helpdesk, that one was helping me, I had to reset …
that is what I did, and then, the phone, well, it
disappeared, the signal,…
A: ok
C: then I tried again, but couldn’t get him, I phoned several
times …
A: yes
C: and that didn’t work and now, I did, uhh, well, uhh …
A: ok, so you are phoning with a mobile phone now?
C: yes, …
22
BASIC SETTING
Agent 2Agent 1
Domain 2Domain 1
23
SOME NUMBERS
dialogue
lengt
h
(sec)
proximal distal
total
length
/
totalA K A K
1 442 - 1 19 27 47 9.4
2 559 4 1 24 15 44 12.7
3 219 1 - 11 3 15 14.6
4 451 1 4 35 13 53 8.5
5 169 1 - 9 8 18 9.4
6 1503 9 6 55 40 110 13.6
7 662 4 2 31 17 54 12.3
8 737 1 - 18 17 36 20.5
9 555 - - 18 14 32 17.3
10 802 3 - 38 24 65 12.3
6099 24 14 258 178 474 12.9
a demonstrative
every 13 seconds
24
REFERENCE TYPE
proximal distal total
exophoric (new) 6 (16%) 4 (1%) 10
cataphoric 2 (5%) 1 (0.2%) 3
anaphoric
left-dislocation - 21 (5%) 21
associative
(new)3 (8%) 18 (4%) 21
other 27 (71%) 392 (90%) 419
total 38 (100%) 436 (100%) 474
020406080
100
proximal
distal% o
f occurr
ence
25
EXOPHORIC (NEW)
7.84
K: Ik kan even geen ‘servers’ vinden, eh … , dan moet ik even
dit aanklikken
K: I cannot find ‘servers’, uh … , then I have to click this.
10.108
K: Dus, sluiten met die x in de rechterbovenhoek?
K: So, closing with that x in the upper right corner?
26
ASSOCIATIVE ANAPHOR (NEW)
1.36
A: Telefoneren is wel goedkoop, maar ik telefoneer zo dikwijls
dat ik dat weer kwijt ben aan mijn mobiele telefoon.
A: Phoning may be cheap, but I am phoning so often that I am
loosing that with my mobile phone.
4.17
A: U zult merken dat uw verbinding weer werkt als dat rode
lampje uitgaat.
A: You will notice that your connection works again if that
red light is off.
27
DISTANCE TO ANTECEDENT
proximal distal
distance
(in #turns)
rang
e0-30 0-212
mean 3.0 5.2
sd. 9.2 19.6
new 9 (24%) 20 (5%)
28
DISTANCE TO ANTECEDENT
proximal distal
distance
(in #turns)
rang
e0-30 0-212
mean 3.0 5.2
sd. 9.2 19.6
new 9 (24%) 20 (5%)
Difference of means is not significant!
29
DISTRIBUTION OF DISTANCE
distance
(in #turns)proximal distal total
0 10 (33%) 165 (40%) 175 (40%)
1 10 (33%) 119 (29%) 129 (29%)
2-10 7 (24%) 97 (23%) 104 (23%)
11-20 2 (6%) 13 (3%) 15 (3%)
21-80 1 (3%) 12 (3%) 13 (3%)
>80 - 8 (2%) 8 (2%)
total 30 (100%) 414 (100%) 444 (100%)
(‘new’ excluded from counting)
30
DISTRIBUTION OF DISTANCE
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
0 1 2-10 11-20 21-80 >80
proximal
distal
31
% o
f occurr
ence
No difference in distance distribution between proximal
and distal reference
32
No difference in distance distribution between proximal
and distal reference
But how then could we measure a difference in
accessibility in the first corpus?
33
No difference in distance distribution between proximal
and distal reference
But how then could we measure a difference in
accessibility in the first corpus?
Because people pointed to important and faraway
objects.
Pointing brings the object in focus and therefore nearby
use a proximal
Distinguish between the act of reference and the act of
focusing.
34
CATAPHORIC PROXIMALS
Exophoric (separated)
(7.84)
K: … dan moet ik even dit aanklikken, denk ik
K: … then I have to click this, I think
Discourse (6.120)
A: Laten we deze afspraak maken dat …
A: Let’s make this appointment that …
speaker activated
35
But suppose proximals are more near or in
focus, then why didn’t we measure a difference in
the distance distribution?
36
SUBSTRUCTURES IN DISCOURSE
6.15 K: I want to remove my email address
6.19 A: You cannot remove this email address
6.20 K: That is a problem
6.29 A: Do you use this email address?
37
substructure
substructure
SUBSTRUCTURES IN DISCOURSE
• tasks and subtasks
• interruptions
• rhetorical structures
• sub-questions
• …
38
K: Do x
Bla bla this x bla bla
K: Yes, but then … .
A: Well, …
…
PRONOMINAL VS. ADNOMINAL
proximal distal total
pronominal 15 (39%) 321 (73%) 336
adnominal 23 (61%) 115 (27%) 138
total 38 (100%) 436 (100%) 474
0
20
40
60
80
100
Pronominal Adnominal
Proximal
Distal
χ² = 20.1, p<.001
% o
f occurr
ence
39
If proximal would be near, focus or easy
accessible, then we would expect less description.
What happens?
40
DISTAL, DISTANCE AND FORM
0
10
20
30
40
50
60
70
80
90
100
0-3 4-10 >10
pronominal
adnominal
0-3/4-10:
χ²=50.3, p<.001
4-10/>10:
χ²=3.82, p≤.05
41
%
PROXIMAL, DISTANCE AND FORM
0
10
20
30
40
50
60
70
80
90
100
0-3 4-10 >10
pronominal
adnominal
0-3/>4:
χ²=2.48, p>.05
42
%
Proximals are relatively more ‘new’
proximals: 24%
distals: 5%
New introductions ‘receive’ more information
Relatively more proximals are in adnominal form
Other cases could be explained by discourse deixis
or contrast
43
CONCLUSION
The traditional classification of near and faraway could be
restored
We found no difference in distance (distribution) between
proximal and distal use in number of turns
Distance in turns or words is not (always) an adequate
measure for accessibility or nearness
degree of focus in perceivable world (situation)
related to structure in discourse (domain model, task)
A clear distinction should be made between the act of
bringing an object into focus and the act of reference
Conversational setting should be described very carefully
44