I gave this presentation at the Cognitive Aspects of Computational Language Acquisition workshop in the European Association of Computational Linguistics conference in 2009.
It presents a freely available open source project written in Java for studying the computational modelling of child language acquisition. It supports plugging in different learning algorithms and evaluating them all within the same environment and with a standardised set of metrics.
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
A Collaborative Tool for the Computational Modelling of Child Language Acquisition
1. Kris Jack
A Collaborative Tool for the
Computational Modelling of
Child Language Acquisition
The Language
Acquisition
Toolkit
CEA LIST, Laboratoire d'ingénerie
de la connaissance multimédia multilingue,
FontenayauxRoses, France
2. Kris Jack Introduction
Introduction What is the Language Acquisition Toolkit (LAT)?
Background It's a program that helps you to model language
LAT acquisition. You can:
Discussion
Future
● create a miniature world in which your language
Conclusion learner lives;
● create comprehension and production tests for
your learner; and
● judge your learner's development over time.
You provide the learner, the LAT provides the
environment.
3. Kris Jack Background
Introduction What motivated the creation of the LAT?
Background ● Complicated problem → numerous computational
LAT models proposed.
Discussion
Future
● Useful to compare them but difficult due to
Conclusion different:
● assumptions;
● training;
● tests; and
● goals.
What is there to gain?
● Time saving benefits.
● Standardisation.
● Holistic models of longitudinal development.
4. Kris Jack LAT
Introduction www.langac.com
Background
Training
LAT
● Training
● Testing +
● Analysing
Testing
Discussion comprehension
Future
Conclusion production
Language Learner
Developmental
Data
Analysing
5. Kris Jack LAT
Introduction www.langac.com
Background
Training
LAT
● Training
● Testing +
● Analysing
Testing
Discussion comprehension
Future
Conclusion production
Language Learner
Let's see one
Developmental
possible world... Data
Analysing
6. Kris Jack LAT: Training (1/2)
Introduction Your world's framework:
Background ● miniature language acquisition paradigm (Feldman
LAT et al., 1990) – associate images with descriptions
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion ● world of geometric objects in which actions occur
and are described (action + description = event)
● objects (colours, shapes, sizes, 2D positions)
● actions (moving, growing, shrinking, appearing,
disappearing, pushing another object and pulling
another object, etc.)
● descriptions (defined by you)
7. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
8. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
9. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
10. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
11. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
12. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
det1 n1 det2 n2
a ye llow cir cle to the u pper right of the blue square
adj1 prep adj2
13. Kris Jack LAT: Training (2/2)
Introduction You must teach your language to the LAT
Background
LAT Before action (t=1) After action (t=2)
● Training
● Testing 3 3
● Analysing 2 2
Discussion 1 1
Future 1 2 3 1 2 3
Conclusion O1: square O1: square O2: circle
O1: blue O1: blue O2: yellow
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
det1 n1 det2 n2
a ye llow cir cle to the u pper right of the blue square
adj1 prep adj2
adj1 = {“ye llow”, “blue”, “red”, “green”, “pink”...}
14. Kris Jack LAT: Testing
Introduction www.langac.com
Background
Training
LAT
● Training
● Testing +
● Analysing
Testing
Discussion comprehension
Future
Conclusion production
Language Learner
Developmental
Data
Analysing
15. Kris Jack LAT: Testing (1/2)
Introduction What do you want to test?
Background ● Comprehension
LAT
● Training
● description in → meaning(s) out
● Testing ● Production
● Analysing
● meaning in → description(s) out
Discussion
Future
Conclusion Teach the LAT:
● what to test
● what results are correct
16. Kris Jack LAT: Testing (2/2)
Introduction Define testing pairs (description + meaning)
Background Before action (t=1) After action (t=2)
LAT
● Training 3 3
● Testing
2 2
1 1
● Analysing
1 2 3 1 2 3
Discussion
O1: square O1: square O2: circle
Future
O1: blue O1: blue O2: yellow
Conclusion
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
I want to test:
● all actions
17. Kris Jack LAT: Testing (2/2)
Introduction Define testing pairs (description + meaning)
Background Before action (t=1) After action (t=2)
LAT
● Training 3 3
● Testing
2 2
1 1
● Analysing
1 2 3 1 2 3
Discussion
O1: square O1: square O2: circle
Future
O1: blue O1: blue O2: yellow
Conclusion
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
I want to test: i.e. give the description
● all actions ● all shapes
square, it could be any
● all colours
object at any time
18. Kris Jack LAT: Testing (2/2)
Introduction Define testing pairs (description + meaning)
Background Before action (t=1) After action (t=2)
LAT
● Training 3 3
● Testing
2 2
1 1
● Analysing
1 2 3 1 2 3
Discussion
O1: square O1: square O2: circle
Future
O1: blue O1: blue O2: yellow
Conclusion
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
I want to test:
● all actions ● all shapes ● all NPs
● all colours
19. Kris Jack LAT: Testing (2/2)
Introduction Define testing pairs (description + meaning)
Background Before action (t=1) After action (t=2)
LAT
● Training 3 3
● Testing
2 2
1 1
● Analysing
1 2 3 1 2 3
Discussion
O1: square O1: square O2: circle
Future
O1: blue O1: blue O2: yellow
Conclusion
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
I want to test:
● all actions ● all shapes ● all NPs ● all NP1s
● all colours ● all NP2s
20. Kris Jack LAT: Testing (2/2)
Introduction Define testing pairs (description + meaning)
Background Before action (t=1) After action (t=2)
LAT
● Training 3 3
● Testing
2 2
1 1
● Analysing
1 2 3 1 2 3
Discussion
O1: square O1: square O2: circle
Future
O1: blue O1: blue O2: yellow
Conclusion
O1: x2 O1: x2 O2: x3
O1: y2 O1: y2 O2: y3
a ye llow cir cle to the u pper right of the blue square
21. Kris Jack LAT: Analysing
Introduction www.langac.com
Background
Training
LAT
● Training
● Testing +
● Analysing
Testing
Discussion comprehension
Future
Conclusion production
Language Learner
This is your job!
Developmental
Data
Analysing
22. Kris Jack LAT: Analysing
Introduction www.langac.com
Background
Training
LAT
● Training
● Testing +
● Analysing
Testing
Discussion comprehension
Future
Conclusion production
Language Learner
Developmental
Data
Analysing
23. Kris Jack LAT: Analysing (1/2)
Introduction What do you want to analyse?
Background
LAT
● Training
● Testing
● Analysing
Discussion
Future
Conclusion
24. Kris Jack LAT: Analysing (2/2)
Introduction How are the results presented?
Background ● Roundbased (results from a single round)
LAT
● Training
● Trialbased (results from all rounds across a single trial)
● Testing
● Modelbased (results across all trials)
● Analysing
Can take stages into account
Discussion
Future
Conclusion Holophrastic Late Multiword Post
Early Abstract
Prelinguistic Multiword Abstract
Reflexive Atomic Multiple Multiple units Multiple units
vocalisations, lexical units with syntactic with phrasal
cooing, units markings category
vocal play, (e.g. word membership
babbling order)
0 6 12 18 24 30 36 42 48 54
Time (months)
25. Kris Jack LAT: Analysing
Introduction www.langac.com
Background
Training
LAT
● Training
● Testing +
● Analysing
Testing
Discussion comprehension
Future
Conclusion production
Language Learner
Developmental
Data
Analysing
26. Kris Jack Discussion
Introduction What does the LAT give us?
Background ● Time saving benefits.
LAT
● Standardisation (let's play on fairly):
Discussion
Future ● training;
● testing; and
Conclusion
● analysing.
● Analyses for longitudinal development.
27. Kris Jack Future
Introduction What does the crystal ball show us?
Background ● More complex miniature language
LAT
Discussion ● More languages
Future ● A richer language representation (e.g. tones)
Conclusion
● More realistic symbolic meanings
● Accessibility for nonprogrammers
● Can we integrate paradigms that do not use a
miniature language?
● Wish list (www.langac.com)
It's collaborative so you can make these possible
28. Kris Jack Conclusion
Introduction Key points
Background ● The LAT allows you to:
LAT
Discussion ● create training data;
Future ● create testing data; and
Conclusion ● analyse your learning model's development
(comprehension and production).
● Standardisation
● let's speak the same language
● Work in progress
● Collaboration
● I'll show you mine if you show me yours
29. Kris Jack Thank you
Introduction www.langac.com
Background
Training
LAT
Discussion
Future +
Conclusion Testing
comprehension
production
Language Learner
Developmental
Data
Analysing
30. Kris Jack Results (1/2)
Introduction Stages in Production
100.00%
Background
LAT 90.00%
Discussion 80.00%
% of Requirements Met
Future 70.00%
Holophrastic
Early Multiword
Late Multiword
Abstract
Postabstract
Holo
Conclusion 60.00%
Early
50.00% Late
40.00% Abstract
End
30.00%
20.00%
10.00%
0.00%
0 15 30 45 60 75 90 105 120 135 150
Number of Events Observed
0 6 12 18 24 30 36 42 48 54
Time (months)