Japan Advanced Institute of Science and Technology

Japan Advanced Institute of Science and Technology

JAIST Repository

https://dspace.jaist.ac.jp/

Title

開発

Author(s)

Citation

Issue Date

Type

Text version

URL

Rights

Description

from a Cooking Recipe

Hideki Uematsu

Scho ol of Information Science,

Japan Advanced Institute of Science and Technology

February 15, 2001

Keywords: the domain of cooking tasks, 3D animation, intermediate expressions,

dictionaries.

In recentyears, asthe performance of hardware isimproving,3D graphicstechnology

isprogressing remarkably. Althoughthose graphicswereused invariouselds, theywere

not necessarily required in the usual life. However, the researches which relate graphics

withlanguageprocessingincreasegradually. The reasonwhythissortof researchismade

is that people can understand further contentsof language information by visualization.

In order to construct a system which generates graphics from language expression, we

haveto clariesthe following three problems.

What kind of information should beextracted fromlanguage information?

In addition to language information, what kind of information should be comple-

mented?

What kind of graphicdata should be prepared inadvance?

We aim toclarify characteristic of the language expressions inthe domain of cooking

tasksand toconstructasystem whichgeneratesgraphicanimation ofcooking tasksfrom

naturallanguageexpressionsin"recipe". Inthispaperwediscusstheissuesforconstruct-

ing suchsystem. After analyzing cooking recipes, we found that "graphic elements"and

"cooking actions"in recipes have to b e specied for constructing the graphicgeneration

system. Graphicelementsare language expressionswhichhaveto be visualizedand they

can be categorized. The categories of "Graphic elements" are "food", "action", "instru-

ment", "position". However, in a recipe, these elements are not necessarily described

completely, accordingly we found that the dictionaries ab out these elements have to b e

held inthe system.

Copyright c

2001byHidekiUematsu

onthebasisofthecontinuationofacookingaction,"actionwhichcanbespecieditsend"

and "actionwhichcan't bespecieditsend". Inasentence whichincludes"actionwhich

can't be specied its end", at least one adverbial phrase exists generally to clarify the

duration of anaction. Moreover, these adverbial phrase can be divide intothe following

three categories.

concrete adverbialphrase accompaniedby anumerical value

abstractadverbialphrase whichmeans the degree of action

adverbialphrase expressed by the form change offood

Furthermore,since cooking tasksconsist of the sequence of nite cooking actions,we

analyzed the relation of two consecutive cooking actions. Consequently, the way of the

visualization of two consecutive cooking actions divides into fourpatterns. Consider the

caseof connectingtwoactions "A,B", the systemcouldgenerate thefollowingfourtyp es

of co oking animations.

1. animation of A +animation of B

2. play animation of B during playinganimation of A

3. play only animationof A

4. insert new animation of Cbetween animation of A and animation of B

Based on the above analysis results, we designed the system which generates graphic

animationof cooking tasksfromcooking recipesasaninput. First, withthe resultof the

morphologicalanalysis andparsing ofcooking recipes,graphicelementsincooking recip e

can be specied referring to the dictionaries of foo d and an instrument. Then, the tags

which represent foo d and an instrument are given to these graphic elements in cooking

recipes. Next,on the basis of a verb, each cooking procedure divide into a cooking step.

Then, the group of a cooking step is replaced with a case frame represented inthe verb

dictionary. Using the tag information incooking recipes, the parameters in a case frame

is identied. To satisfy preconditions of a cooking action, a new step is added to each

cooking step if necessary. After this process, "intermediate expressions" which include

necessary information, graphicelementsand co oking actions, is nally generated.

In order togenerate graphicanimation,weprepared graphicsforgraphicelementsby

using 3DCG modeling tool. In the domain of cooking tasks, food sometimes changes its

formincontactwith aninstrument. When aCG objectfor"food"iscreated, wemade it

haveinternalstructure inconsiderationofthe changeof itsform. Inconnection withthis

change, to create animation of cooking tasks, we prepared scripts whichexpress mutual

action b etween graphicelements.

With the above descriptions in mind, we constructed the experimental system which

generates animation of cooking tasksfrom cooking recipes as aninput.

"graphic elements" must be expanded. Furthermore, more graphics must be prepared.

For example, graphics for fo od whose form is more complicated, such as noodles and

liquid materials, must be prepared.