MULTIMETEO is developing a system for automatically generating weather forecast information in several EU languages, drawing on basic data from the meteorological institutions involved in the project. Besides facilitating convenient rapid access to weather information services via phone, fax and electronic networks, the system will provide information in the preferred language, geared to the specific needs of groups such as transport services or sports organizers. From a semantic meteorological model, a text in the desired language will be tailored to the style and detail required by the group concerned.
During the first year MultiMeteo has carried out a market and cost benefit analysis, leading to selection of priority installation sites and weather forecast styles for generation. Styles for forecasts have also been formalised after study of 17 corpora and forecasting guidelines from a range of partner sites. Analysis and specification of user requirements has defined the organisational and technical environments and the quality criteria to be applied. As an aid in these areas an early weather forecast generating prototype has been installed and tested at a user site. This experience was invaluable in focusing the users on their real requirements in terms of characteristics of the texts to be generated and the desired user interface.
The project has refined the overall quality plan with a specification of quality criteria and priorities, a quality process and quality. A fresh prototype interface showing the functionality of an improved demonstrator with multi-lingual interactive generation and administration has been developed to further refine user requirements. An extensive validation test plan for the final demonstrator has been agreed with users.
The Market Analysis
The market analysis studied the existing French and Spanish markets for different styles of weather forecasts at six sites. The French styles studied were: 'rivage' (sea-side resorts), inland tourism, fun-board oriented (water sports), road conditions, mountain ranges, regional and smaller 'departmental' general-public, general-public for Paris, regional and national press. The Spanish styles studied were: 'predicción general para España' (national general-public), 'predicción para una Comunidad Autónoma' (regional general-public), and 'predicción provincial' (departmental general-public). Madrid uses all the styles, the other Spanish sites have the regional and departmental general-public styles.
The market analysis lead to the identification of priority demonstration sites in Spain (Madrid, Malaga, Santander, Valencia, Palma de Mallorca, Las Palmas de Gran Canaria) and France (Toulon, Lyon, Grenoble-St. Martin d'Hères, Paris, Brest, Strasbourg), together with the associated styles for forecasts; in Spain, national, regional and provincial forecasts for the general public, in France kiosk based for the general public, coastal (holiday resort), and specialised water sports forecasts. In addition to Spanish and French there is also a tourist, expatriate and water sports market for English and German language forecasts. Météo-France will use Multimeteo to realise a new market in very frequently updated and local or 'zone' forecasts for the general-public, which is not presently feasible.
Organisational and technical environment
Presently forecasts are produced in two steps:
- 'visualisation', where a forecaster gains an appreciation of the current meteorological data from Radar, Satellites, Mathematical Models, and others sources.
- 'forecasting', actual writing of the weather forecasts, sometimes in several styles (local or regional level, mountain or sea-side oriented, etc.).
Weather forecasts are produced from one to three times a day depending on the style. Using MultiMeteo the production takes three steps, the first is as before then:
- a dialogue takes place between the forecaster and software for capturing relevant data for the weather forecast.
- the MultiMeteo software generates the forecast in selected languages and with the selected styles. There is an opportunity for the forecaster to modify the MultiMeteo parameters before the generation process, like geographical scope, input source, temporal scope, and, to edit a forecast after generation.
Development of the first prototype
A weather-forecast generation prototype in French for Météo-France has been developed. Weather forecasts are currently generated in a general-public style only, with a geographically and seasonally limited vocabulary.
The prototype has been tested at a meteorological centre, integrated with the site data bases, with a favourable report. Other meteorological centres (including Madrid) are already testing it. It has been found that some observed phenomena are not feasible using the fully automatic generation approach.
Development of a new prototype
Assessment of the first prototype has led to requirements for improvement. The richness of texts generated needs to be improved and the system needs to be positioned as a tool for forecasters to use, not a replacement for their services.
A new interactive generation approach is proposed as a better solution. In the interactive approach, the system performs the following steps:
- reading of the forecast meteorological data from the external data bases or files as before,
- generation of a first version of the text associated to the weather forecast,
- the user interface then allows the forecaster to enter directives for composing the final version of the text,
- generation of the final version of the text in multiple languages.
Ergonomics will be crucial in specifying the interactive-generation user interface. Based on GSI-Erli and La Redoute studies of semi-automatic text production, a friendly interface can reduce time for producing a single forecast by two thirds. A poor interface will simply mean that the system is abandoned by users.
The Way Ahead
User analysis and specifications will be completed in the first quarter of 1997, development of the demonstrators can then begin. The first available full demonstrator (French zone-kiosk style) is planned for July 97. The rest of the developments will be completed by the end of August 97. All French and Spanish demonstrators will be available at that time.
Whilst the software developed in the project will first be applied to weather forecasts, it is also relevant for producing other types of more or less short and stereotyped information documents, for example: official bulletins, short news items (stock exchange, sport, economic, air quality, road conditions, etc.), bank and insurance reports and short technical reports. The demonstrator will provide a basis for exploring these domains later.
The European weather forecast market is no longer limited to the national level. Travel within Europe is also developing, with increasing numbers of people going abroad for varying lengths of time, for both personal and professional reasons. Also, with the development of Multimedia and the liberalisation of Telecoms, markets for meteorological forecasts transcend borders and thus need multilingual tools.
Volumes involved are high and time constraints for the production of fresh usable information are stringent: tens to several hundreds of bulletins for each meteorological office, produced once to three times a day. This prohibits the use of classical translation processes.
The input to the application is structured data containing weather information. The formalization of this information depends on the source country. In order to use the same kernel for all the countries, a knowledge base will describe the data structure conceptually.
The kernel of the generation process will use GSI-Erli's AlethGen toolbox, which has already been used for generating texts, in particular for producing correspondence for a leading French mail-order company.
Progress and results
The main results of the MultiMeteo project are the following:
- User need definition and Quality criteria
- System external specifications according to users' definition
- Development of conceptual and linguistic resources
- Adapting of the AlethGen engine
- Integration and verification of several demonstrators
- Validation at different real sites.
The MultiMeteo consortium will participate in several EC events, disseminating all available information on the project and progress.
Automatic generation of weather forecasts will use different media and be made available to a variety of customers. Timescales vary according to these different applications:
Text output (fax or videotext) will be implemented immediately after the termination of the project (or a few quarters later to allow for initial training of personnel) in Central and most of the Regional offices.
Multilingual applications will be proposed for posters in tourist station information offices; hotels; camping sites - and on videotext as necessary.
Multimedia applications could be developed in parallel with Project activities. One can expect that a distribution through client servers and networks will be available by the end of the Project., or shortly after. This depends on partner agreements. Professional applications will be developed as soon as possible.
Vocal applications are a major issue for the three user participants. As this kind of application has a high priority, exploitation will follow within a few quarters of the successful termination of the project.
Funding SchemeCSC - Cost-sharing contracts