Cross-Domain Cue Switching Tiansi Dong tdong@uni-bonn.de AI - - PowerPoint PPT Presentation

Cross-Domain Cue Switching

Tiansi Dong tdong@uni-bonn.de AI Foundations Group Bonn-Aachen International Center for Information Technology (B-IT) University of Bonn

Translation Shared Knowledge in Spatial Description Approach to Shared Knowledge between Languages

Translation

Language Translation Example 1

How shall we translate white as snow into a native language in Benin (Natemba)?

white as snow very white white as pelican

Language Translation Example 2

How shall we translate you are my heart into Indonesian?

you are my heart you are the most important person to me you are my liver the liver is the most important

• rgan

the heart is the most important

• rgan

In different languages, the same meaning may be carried by words w i t h t o t a l ly d i f fe r e n t l i t e r a l

• meanings. Translation needs speakers

to select the right word in the target language which may have a distorted meaning structure from the source language

Language Translation Example 3

The Guugu Yimidhirr people only use absolute orientations: „on the southern edge of the western table“. How shall we translate it into the normal English expression?

N

I left it on the western table It is on the table on my left side

ß West East North South I left it on the western table

ß It is on the table on my left side

Left

ß West

West left

Shared Knowledge between Orientation Descriptions

the western table the table

• n my left side
• Newtonian

perspective Lebnizian perspective

Distortion

• Select: West Pole

Shared Knowledge: Distance Comparison

Select: Heart

Shared Knowledge among all Orientation Descriptions

The distance comparison

P:Aeiθ The orientation of P is the point

• n the unite circle which is

nearest to P. Q W

Shared Knowledge among all Orientation Descriptions

Distort the unite circle into a polygon Define a distance function between point and a line segment

front left right behind

Generate all Qualitative orientation framework in the literature

T.Dong and H.Guesgen „A Uniform Framework for Orientation Relations based on Distance Comparison“ ICCI’08.

Distance Relations

In UK, Object A is one foot away from Object B.

A B

In Germany, people use double-feet, or Elle Physicists use light: light-year is the distance travelled by light in one year; one meter is is the distance of light in vacuum in 0.00000000333564 second In ancient Egypt, China, people use Cubit, Chi(尺), Cun(⼨对)

Shared Knowledge among Distance Relations

The connection relation

A B

Select an object category (shape) as unit

The minimal number of elements in the object category, which can connect with the two objects, with the condition that they are connected with each other

Connection Relation

∀A, B[C(A, B) → ∀Z∃Z ∈ Z[C(A, Z) ∧ C(B, Z)]]

A B W V

• J. Piaget
• S. Carey

T.Dong „A Comment on RCC—from RCC to RCC++“, Journal of Philosophical Logic

Shared Knowledge for Spatial Relations (Summary)

absolute

• rientation

qualitative

• rientation

Distance comparison Distance Connection Select Distort

Shared Knowledge in Spatial Descriptions

A table is under the projector A tree leans against the bike A projector is on the table A bike leans against the tree

Commonsense knowledge of stability

Shared Knowledge in Spatial Domain

Object classification at basic-level category Commonsense knowledge of stability Object locations based on the connection relation

Suppose two persons in the same environment, what is their shared spatial knowledge? Supported by the case of L.E.

T.Dong „Recognizing Variable Environment-The Theory of Cognitive Prism“, Springer

Why Are W e Interested in Spatial Domain?

The first Domain that human encounters and understands It is the reference domain for the cognition of other domains A common research field for several disciplines

Approach to Shared Knowledge between Languages

Translation without understanding can be incorrect, sometimes impossible Descriptions about spatial domain in different languages or cultures have shared knowledge What is the knowledge shared between languages?

The Shared Knowledge between Languages of Bilingual Speakers

The Aggregated Language Model Separation between Meanings and Forms.

MacWhinney, B., Bates, E., & Kliegl, R. (1984). „Cue validity and sentence interpretation in English, German, and Italian“. Journal of Verbal Learning and Verbal Behavior, 23:127-150.

Cue Cue

Cue-Switching for Translation

cue switching Retrieve meaning by Cues of L1 Produce form by Cues of L2

26

Five Basic Cues in Language

1) Agreement

e.g. you are, I am

2) Animacy

e.g. a dog bites bones

3) Case

e.g. ich, mich, mir; I, me

4) Marker

e.g. 把，被，着，了

5) Word order

e.g. a dog bites a cat

German: 1) + 3) English: 2) + 5) Chinese: 4) + 5) Turkey: 3) Italian: 1)

Cue: the relation between form and meaning Five cues found in 165 Languages

27

165 Natural Languages with only 5 cues Meaning construction with understandable, but ungrammatical sentences There should be some algorithms, software framework to construct possible meanings only based on single word meanings.

W

• rd Salat Problem

cues used in German

Das Auto gefällt mir gut

cues used in Chinese

German-Chinese Translation based on Cue Switching

我很喜欢这⻋轧

Subject: das Auto Receptor: mir Descriptor: gut Frame: gefallen Actor: 我 Object: 这⻋轧 Descriptor: 很 Frame: 喜欢 Frame: gefallen Frame: 喜欢 Subject

• Receptor
• Actor
• Object

I:mir: 我 the car: das Auto: 这⻋轧 like: gefällt:喜欢 well:gut:很

Translation based on Cue Switching supported by Shared Knowledge

X on the western table

X 在我左 边的桌上

Object: X Location: table[West(table)] Frame: On Frame: 在..上 Object: X Location: 桌上[我左边(桌上)] cues used in English cues used in Chinese

ι ι

NRST(table,W-Pole)

NRST(table,I.Left)

NRST(I.Front,N-Pole) visual, GPS, dialog

C(I.Left, W-Pole)

On X 在..上 table 桌上 West ⻄覀⾯靣 桌上 ⻄覀⾯靣

/ ∈ dom

I.Left 我左边 West I.Left

verbal cue facial cue gesture cue voice cue

• lfactory

cue haptic cue visual cue auditory cue Each cue provides knowledge from one perspective about the physical world and/or the psychological world. General principles of the two worlds make the transition possible

Schematization Similarity Conjecture

To the extent that space is schematized similarly in language and cognition, language will be successful in conveying space. — B. Tversky & P. Lee

On-going Tasks

Acquire Shared Meaning Representation of German and Chinese sentences in textbooks Developing Software Systems for Machine Translation based on Cue-Switching Developing algorithms to solve for Word Salad Problem supported by DBpedia

Thanks