Euromap Euromap Euromap
Euromap - your path to market for speech and language technologies  
The gateway to speech and language technology opportunities
European network of excellence in HLT
EC Programme fostering digital content on the global networks
   Natural language querying -EN
   MT in Bulgaria
   CAT Tools
   Information Access -Greece
   The Power Of Voice
   Language Tech in France
   Flanders speaking
   LT in Berlin - EN
   Beyond Simple Keywords -EN
   Crossing The Chasm
   Interview - Dr. Bartenstein -EN
   The Joy of Text
   How to HLT-Enable EN
   How to HLT-Enable NL
   Danish-Search Engines EN
   Austria - Way Ahead EN
   > Email Response - EN
     Email Response - DE
   Semantic Web
   Finnish Lessons
   Daily life-EN
 Success Stories
 Tech Transfer
 EU - HLT Fact Files
 HLT Policy Pages
 CEEC Info Centre
 Who's Who
 HLT Projects
Visit EUROMAP Language Technologies in these countries:

EUROMAP Article: Deutsch
Email Response Management Based on Language Technologies


email management

quote   A survey by Emagicon of Germany has found that it took an average of 95.38 hours for a [business] email reply to come back, if one was returned at all.  unquote

Information sources

? Emagicon AG
? Frost & Sullivan
? Email Management Benchmarking Association (EMMBA)

quote   ?Often hasty solutions have been developed, which reduce the costs for email responses,? says Uszkoreit, in whose opinion there are very few satisfactory solutions around today.   unquote
Professor Hans Uszkoreit
co-founder of XtraMind

quote   ?We have developed a DNA search engine with our finder which is proving to be very promising in tests. This product can be applied in all sectors with comparable data structures?.   unquote
Dr. Klaus Netter, CEO XtraMind

Related Products and News

? Bertelsmann Turns to Kana
CRM Magazine, 21-05-2002

? Tonxx: German provider of ERM software 'Responsio'

? RightNow: Email Response Management (Demo)

? YY Technologies: 'See YY' and 'Hear YY'

Related EC Funded Projects

? MIETTA II: A Multilingual Information Environment for Travel and Tourism Applications

? CLARITY: Cross-Language Information Retrieval and Organisation of Text and Audio Documents

The number of emails is soaring. The need for automated email processing systems is also growing rapidly. Currently available email response solutions, however, are less than perfect. A key problem can be seen in categorisation?the process of sorting emails automatically?where only a combination of different methods will lead to a satisfactory solution. A young Saarbr?cken-based company, XtraMind, has developed a promising software package incorporating language understanding, machine learning and professional workflow management to come up with a solution.

By , independent journalist

The request for email response management systems is booming as email emerges as a popular means of communication between customers and companies. Customers send their inquiries quickly and simply to suppliers and ... wait and wait. A survey by Emagicon AG of Germany has found that, in the summer of last year, it took an average of 95.38 hours for a reply to come back, if it was returned at all. At the same time, the International Data Corp (IDC) documented about 10 billions emails per day worldwide for the year 2000. This number is expected to rise to 35 billion a day by the year 2005. According to Frost & Sullivan, American companies alone will invest more than two billion US dollars in 2007 to control this influx of email.

Customer relationship and cost reduction

As a result of this email boom, a new field has recently emerged in customer relationship management ? called email response management (ERM). ERM is developing automated methods, based on language technology processing techniques, to speed up email handling as well as making it more efficient. Already 150 to 200 incoming emails a day could justify a properly deployed ERM system, from the perspective of retaining customer loyalty and reducing personnel costs.

?The software industry has risen to this challenge as well as it could,? says Hans Uszkoreitco-founder and Board Member of XtraMind and professor of computer linguistics at the Saarland University in Saarbr?ckenin a statement about the market situation. ?Often hasty and less-than-perfect solutions have been developed, which reduce the costs for email responses, at least to some extent.? But in Uszkoreit's opinion there are very few satisfactory solutions around today. ?The big challenge lies in the correct categorisation of email, for which humans use their language skills.?

The basic problem lies in categorisation

The categorisation of most ERM systems is currently based on either statistical-mathematical techniques, or pattern-based ('pattern matching') procedures. Using the statistical approach, a system 'learns' the statistical characteristics of the different categories through an analysis of example texts. The relative frequencies of single words play a central role in this approach. This is why the procedure usually requires large amounts of text for 'training' the system, to provide acceptable results. This is not the case, however, with pattern matching categorisers, where searching by keywords is much more relevant. Keywords are either specified in advance or can be acquired through a machined learning procedure. Their relative frequencies are, in principle, of no relevance.

It turns out that one successful strategy leading to acceptably high quality of results is through the use of combinations of different language technology techniques. A young Saarbr?cken-based company?XtraMind Technologies?is one German company working along these lines. In addition to an integrated ERM solution, XtraMind is offering single technologies as part of a construction tool set, which improves different functions within the ERM.

Speech recognition improves ERM

The categorisation software XM-XtraClass is at the core of Xtramind's modular product range. In addition to the statistical and pattern matching procedures already described, it integrates three language recognition methods:

  • the vector space procedure which maps sample texts of one category as a high-dimensional space vector. Through the state-space separation of the sample quantities, rules are derived to categorise new examples into this space
  • the ngram-model which uses the language-specific statistic distribution of word- and punctuation-sequences to determine the specific features of thematic categories
  • the boost-end rule learning procedure which derives classification rules from large quantities of data. From these, sensible categorisations are established, which then can be used for optimising the data.

Where several methods are equally suitable, these are made to cooperate with each other, leading to better results through mutual reinforcement. With its optimal blend of language technology and statistical procedures, XM-XtraClass claims to achieve the enviable combination of precision, abstraction, robustness and dynamic adaptation.

XtraMind owes its development head-start in part to the know-how of its founders, who come from such diverse backgrounds as language technology, software engineering, and IT consulting. Another unique factor contributing to the company's success is the fact that it is the first commercial spin-off of the Deutsches Forschungszentrum for K?nstliche Intelligenz (DFKI) - the high-achieving German Research Centre for Artificial Intelligence. The results of ten years of top-level research at DFKI have, therefore, been incorporated into the development of Xtramind's products, ensuring a sound footing in the future market place.

Dr. Klaus Netter
Klaus Netter, CEO XtraMind

?Our physical proximity to Saarbr?cken-University enables the continuous exchange of ideas with its scientists. None of our 34 members of staff wishes to forego the high-level creative environment,? says CEO Dr. Klaus Netter, a ex-DFKI researcher himself. He also values the world wide network of technology and research partners, such as its recent association with Materna GmbH. XtraMind is quick to capitalise on these benefits through an ongoing expansion of its product range.

The XM-MailMinder addresses and solves this email problem. It either forwards incoming mails automatically to the right expert (and supports his answer by offering alternative solutions), or it returns an reply message itself ? depending on the selected mode. ?The precision level of our classifier is greater than 90%.? Dr. Netter is proud of this achievement, which scores better than competitors' products in all benchmark tests.

The learning procedure the XM-MailMinder uses is anything but simple. Small amounts of around 50 emails already produce excellent results and a stable learning model, which is capable of being scaled up semi-automatically. It is possible, threfore, to achieve significantly greater classification differentiation, allowing greater precision in responding to customers' inquiries. This also has the effect of shortening the response time to changed inquiries, as might be the case after a special advertising campaign.

Deployment of XtraMind-products in call centres

C-2-B infogrqphic

Call centres with a volume of up to 1,000 emails per day will value the XM-MailMinder as a very welcome support tool. Installation effort of the system is not too heavy. ?With the clustered and tailored customer data we can compile a model within two or three days. This will subsequently be tested by the customer for a period of one to two weeks. Then starts the operative deployment?, says Dr. Netter on the process of implementation. This is one reason why many of Germany's largest call centres can be found on on XtraMind's user evaluation list.

The clustering of data is essential in the preparation leading up to the installation of XM-MailMinders. XM-Gravity is a XtraMind system component for the automatic grouping of documents. Without any human intervention, it identifies thematic coherences as well as similarities and is able to form corresponding categories. Thus a system emerges which is based on relevant terms and formulations in texts.

New markets might develop

XM-Finder can search associatively for unstructured information within a variety of document-, information- or knowledge management systems. ?Large companies often store between 10 and 20 percent of redundant documents.? says XtraMind's CEO, who sees a potential market for cost saving, with data storage costs amounting to an average of 0.80 EUR per megabit per month. With XM-Finder, identical documents can be found and deleted quickly and efficiently.

In particular the XM-Finder has good potential in new, smaller niche markets. ?We have developed a DNA search engine with our finder which is proving to be very promising in tests,? says Dr. Netter. ?This product can be applied in all sectors with comparable data structures?. XtraMind, for now at least, intends to concentrate on its core market - the development of language technologies and products aimed at call centres.

Thea Payome

is an independent journalist based in Munich, Germany. Her main focus is on technology and economy matters within IT and the Internet Economy.

The editors of HLTCentral would welcome any feedback on the article.
Please send your comments to the .

Disclaimer: Any opinions expressed in this article are strictly those of the author and do not necessarily reflect the viewpoint of EUROMAP site or its editors. Copyright ? 2002 HLTCentral. All rights reserved.

go back  Version for printing Last updated: 1/15/03 9:54 AM

Please report problems to