SYSTEM DEMONSTRATION 
INTERACTIVE GENERATION AND 
KNOWLEDGE ADMINISTRATION IN MULTIMETEO 
Jos~. Coch, ERLI 
261, rue de Paris 
93556 Montreuil Cedex, FRANCE 
jose.coch@erli.fr- 
1. Several important problems to solve for applying Generation to the real world 
We believe that it is important from the practical point of view to use natural language generation (NLG) 
in real world applications. The key benefits are outlined as follows: 
• Higher quality of generated texts 
• Cost effective maintenance and adaptability 
• Usability•(including acceptability by the users). 
Generally, it takes "exotic" (linguists, knowledge engineers) manpower to maintain, and adapt a NLG 
system. To avoid this problem, we have developed a Knowledge Administration • station, which is usable 
by the target population (in our project, weather forecasters). 
On the other hand, the system is designed to help the forecasters and not replace them. It is able to adapt 
to each forecaster's style and manage enhancements• they wish to bring to their texts. With that in mind, 
the Interactive Generation environment was designed to allow forecasters to modify generated texts in 
their native language, and then generate weather forecasts in several foreign languages based on those 
modifications. 
Interactive• Generation is a viable alternative to Automatic Translation. 
2. MultiMeteo 
2.1. General motivation 
The European weather forecast market is no longer limited to the national level. Travelling is growing 
rapidly, with an increasing number of people going abroad, for both personal and professional reasons. 
Markets for meteorological forecasts tend to overlap borders and thus need multilingual tools. 
The volumes involved are high, and time constraints for producing fresh usable information are stringent: 
several hundreds of bulletins for each Weather Office, produced several times a day. 
End users have also become• much more demanding about quality and tailoring, which here are realised 
by different styles of weather forecast. 
1.2. Goal and characteristics of MultiMeteo 
The g0al of the MultiMeteo projectis to build an automatic multilingual short-report generation system, 
and to use it in the Meteorology field for producing weather forecasts from structured data I. 
i Other weather-forecast generation systems are FoG \[Goldberg et al. 94\], Weathra \[Sigurt et al. 92\], MET90 \[Sigurt 
et al. 96\], and METEOVIS \[Kerpedjiev & Noncheva 90\] 
300 
I 
I 
I 
I 
ii 
il 
I 
iI 
This system will allow European forecasters to automatically produce texts in English, French, German, 
Spanish, and Dutch. 
MultiMeteo is a 3-year project funded partially by the Language Engineering programme of the European 
Commission, and partially by European weather offices. It involves around 40 people in France, Spain, 
Austria and Belgium. There are around 15 pilot sites in Europe, which are representative of different 
meteorological characteristics (south, north, plain, mountain, sea, etc.). In each site, 3 or 4 styles willbe 
developed (local and regional general public, seaside general public, fun-board sport oriented, etc.). 
3. Interactive Generation and Knowledge Administration 
The Interactive Generation environment allows the user (the forecaster)to: 
• Read the generated text 
• Modify it in terms of concepts, modifiers, links, etc. 
• • Read the modified version 
The following example shows a generated text, and the menu proposed by the system when the user 
clicks on a clause (here "le ciel couvert devient peu nuageux") - "the grey sky becomes less cloudy" 
-, .~ ~ I.,,~t:~;~-.~ ", ---~ 
I UEII I 
METEO-FRANCE, centre departemental du Mans. Voici le bulletin du 16 Decembre a 7 hel 
pour la zone de la vallee du Loir. 
Ce jour : . 
Le temps est nuageux avec une pluie fine. vent d'Est faible ~ moder~ et temperatures voisines de 
Pour la nuit : 
Le ciel est changeant avec de la bruine, vent d'Est mod~re et temperatures voisines de 4 degres 
Pour demain • 
, ......................................................................................................................................................................................................... 
Concept N1 (PI) Devenant 142 
ConceptDvpt:D Eclaircies 
Ciel se D~gageant .. " . 
Concept Nuages Devenant Moins Nombreux 
Concept jOiel:!Devenant Moins Nuageux 
.... : -: .: .... .. : 
Concept :Dvpt ~De~mides Eclairc es 
• .:. ... . : . 
ConceptEclaircies Devenant Un Peu Plus Franches 
dewai(en~ I ii:"ii.-i: .;-...=,.: ..-- . . " 
== ........... p0 rrai0nt):' ' - u .............................................................. ................. ~ ..... 
pourrai(efi)t : :" " W 
W c0mmenoera(0nt) " ~~ ---------- ---" 
continuera(ont).. " " 
•encore 
toujours 
// l/ 
301 
In the above menu, alternative concepts are proposed, and several types of modifiers (modality, phase, 
etc.) can be added. 
For example, if. the concept "Eclaircies devenant un peu plus franches" (Sunny spells becoming more 
extensive) is selected, the System generates the following text: 
METEO-FRANCE, centre d~partemental du Mans. Voici le bull~ 
pour la zone de la vallee du Loir. 
Cejour" 
Le temps est nuageux avec unepluie fine. vent d'Est faible a mod~r 
Pour la.nuit • 
Le ciel . vent d'Est mod~r~ et temp 
Les eclaircies deviennent un peu p\]us franches,~rcec de !a bruine. Ic 
• ~. : .~:~ .: 
Once the forecaster saves the result, MultiMeteo generates the weather forecast in all the selected 
• languages (which are English, Spanish and German). Of course his/her modifications are taken into 
• account i n all of the languages. : 
In addition, MultiMeteo offers a Knowledge Administration station, which allows the administrator to 
modify the style of the texts to be generated, in terms of default structure, titles, order between 
paragraphs, available models of paragraphs, type of sentences used (telegraphic / non telegraphic), terms 
to be used, etc. 
In fact, there are two levels of administration: central and •local. Certain functions are specific to the 
central administration, such as the creation of a new style from scratch, or the management of the 
multilingual terminology. • 
4. Technical Characteristics 
MultiMeteo runs under PC/Windows 95 (or more), Windows NT and Unix station/Solaris2. The input is a 
numerical matrix, where columns are the meteorological parameters and rows are time steps. The output 
is an HTML or RTF file that contains one or more weather forecasts. 
302 
• i I 
! 
I 
The MultiMeteo software has 4 main modules: 
• The Front-End Module (FEM) 
• The Administration Server 
• The Generation Server 
• The Help facilities. 
The FEM, the Administration, and the Help are written in JAVA. 
The Generation Server is based on ERLI's AlethGen toolbox, which has already been used for generating 
texts, in particular for producing correspondence for a leading French mail-order company (see \[Coch 
1996a\] and \[Coch 1996b\]). AlethGen is divided in an engine (written in C++) and Knowledge Bases. The 
approach is inspired by the Meaning-Text Theory 
Acknowledgements 
The MultiMeteo project is partially funded by the European Commission's Language Engineering 
programme (LE 1612), and we are very grateful to the members of the EuroPean Commission's DGXIII 
for their support of this work. 

References 
\[Coch 1996a\] Coch Jos6,1996. "Overview of AlethGen". Proceedings of the International Workshop on 
Natural Language Generation (INLG-96). Herstmonceux, England. 
\[Coch 1996b\] Coch Jos6,1996. "Evaluating and comparing three text-producti0n techniques", in 
Proceedings of the 16th Conference of Computational Linguistics, Coling 96, Copenhagen, Danmark. 
\[Coch 1997a\] Coch Jos6, 1997. "Applications industrielles de la g6nfration : pourquoi et comment", in 
Proceedings of GA TOZ Grenoble, France. 
\[Coch 1997b\] Coch Jos6, 1997. "Quand l'ordinateur prend la plume: la g6n6ration de textes", in 
Document Numdrique Vol. 1, N°3, Hermes, France. 
\[Goldberg et al 1994\] Goldberg Eli, Driedger Norbert, Kittredge Richard, 1994. "Using natural-language 
processing to produce weather forecasts", in IEEE Expert, 9(2):45-53. 
\[Kerpedjiev & Noncheva 1990\] Kerpedjiev Stephan, Noncheva Veska, 1990. "Intelligent handling of 
weather forecasts" in Proceedings of the 13th Conference of Computational Linguistics, Coling 90, 
Helsinki. 
\[Sigurd et al 1992\] Sigurd Bengt, Willners Caroline, Eeg-Olofsson Mats, Johansson Christer, 1992. 
"Deep comprehension, generation and translation of weather forecasts (Weathra)" in Proceedings of the 
14th Conference of Computational Linguistics, Coling 92, Nantes. 
\[Sigurd et al 1996\] Sigurd Bengt, Lastow Birgitta, V~ivarghrd Tomas, 1996. "Computer generation of 
weather overviews" in Proceedings of the ECAI 96 Workshop Gaps and Bridges: New directions in 
Planning atut Natural Language Generation. 
