Language Generation - Latest Commercial Developments

 The US company CoGenTex offers software development for tailored document generators and authoring systems in different application areas. The systems realized so far cover various document types like for example weather forecasts, textual summaries of economic statistics, project management reports, on-line help for graphical user interfaces, etc. Development relies on customizing the proprietary text generation shell COGENT, which follows a template-based approach and provides means for bilingual generation in English and French.

AlethGen is an automatic multi-paragraph text-generation toolbox available from GSI-Erli , France. The system offers a hybrid architecture and provides a direct generator, which is more or less template-based, as well as sophisticated modules for text planning and realization. The toolbox allows flexible configurations, i.e. the modules can be integrated and used in various ways to satisfy different applications' requirements. The languages currently supported are French and English.

(New Apr. 97) Daimler-Benz Research and Technology has developed a sophisticated natural language generation component as a basis for novel products that provide natural language interfaces for various application contexts. The EFFENDI system as described in Poller et al. [1996] and Poller and Heisterkamp [1996] is a real-time incremental generation module which has been especially developed for use in speech dialog systems. The generator is capable of deriving prosody information and is able to convey this information to the speech synthesis module in order to produce more natural sounding speech output. Syntactic generation relies on a variant of the TAG (Tree Adjoining Grammar) formalism, which is specifically suited for incremental processing as noted in Kilger [1994]. EFFENDI is currently available for German and English.

Some potential application contexts for this natural language generation technology include products related to on-board driver support systems, traffic management systems, and automated call-centers.

(New Sep. 97) Logox is a novel low-cost text-to-speech system for generating German on a standard PC with Windows 95 and a normal soundcard. The product, which is available from the German company G Data , combines the advantage of minimal memory requirements with good intelligibility and high segmental and prosodic acceptability. The system relies on a new technique called microsegment synthesis which is based on the concatenation of pre-recorded speech units. These microsegments are phone-sized for consonants and sub-phone-sized for vowels. Following economic principles within the microsegment inventory as well as for prosody, a recognizable human voice can be provided without imposing special requirements on the underlying hardware platform.

(New April 98) In collaboration with leading speech technology companies - namely Apple Computer, Inc. , AT&T , Dragon Systems, Inc. , IBM Corporation , Novell, Inc. , Philips Speech Processing , and Texas Instruments Incorporated - Sun Microsystems, Inc. has developed an application programming interface specification to incorporate speech technology into Java. The Java Speech API is a standard extension to the Java platform that enables Java applications and applets to use speech input and output. It defines a cross-platform API not only to support command and control recognizers and dictation systems but also for speech synthesizers.
The so-called Java Speech Markup Language (JSML) has been developed to provide cross-platform control of speech synthesizers. JSML is used by applications to annotate text input to Java Speech API speech synthesizers in order to provide the speech synthesizer with detailed information on how to say the text. Appropriate markup of text improves the quality and naturalness of the synthesized voice. JSML includes elements that describe the structure of a document, provide pronunciations of words and phrases, and place markers in the text. Furthermore, prosodic elements are provided that control phrasing, emphasis, pitch, speaking rate, and other important characteristics. Since JSML relies on the Unicode character set, it can be used to mark up text in most languages of the world.

An alpha implementation of a core subset of the beta version of the Java Speech API called Speech for Java is available from IBM Corporation . Speech for Java is a Java programming interface for incorporating IBM's ViaVoice speech technology into user interfaces and it supports voice command recognition, dictation, and text-to-speech synthesis, based on IBM's ViaVoice technology.

(New Oct. 98) Sun Microsystems, Inc. has officially released the Java Speech API 1.0 specification for the Java platform. This application programming interface specification has been developed together with Apple Computer, Inc. , AT&T , Dragon Systems, Inc. , IBM Corporation , Novell, Inc. , Philips Speech Processing , and Texas Instruments Incorporated as a vendor-neutral and cross-platform interface that extends the Java platform to allow for speech recognition and speech synthesis in applets and applications written in the Java programming language. The Java Speech API specifies a single interface for development and deployment of a wide range of speech technology applications on the desktop, in small portable devices and for telephony servers. In the case of the Java Speech API , Sun Microsystems only delivers a specification. Implementations of Java Speech rely on third party vendors.

(New Feb. 99) Boston-based Artificial Life, Inc. is a new software company which develops, markets and supports intelligent software robots that automate and simplify certain business-related Internet functions. The so-called ALife-SmartEngine technology is the core component that gives Artificial Life products the expertise to communicate with users in natural languge text or speech. The softbots are designed so that users, especially corporations, can adapt them to their own applications' needs by adding user or company specific knowledge and by customizing the softbots.
Artificial Life 's product line comprises several intelligent software robots. ALife-WebGuide is designed to reside on a WWW site and help users to navigate. Using natural language conversation the softbot tries to match the interest of the visitor with relevant information and Web content. ALife-Knowledge-Manager is being designed to extract information embedded in a company's intranet documents and enables easy retrieval of natural language requests. ALife-Messenger acts as a natural language-based automated email reply and answering service. The ALife-Call-Center-Agent can be used by call centers and help desks as a component that integrates with pre-existing call center software. Based on natural language processing it provides an automated first response to incoming voice telephone calls or email requests, handling basic call center tasks with virtually no aid from human operators. ALife-SalesRep with its very advanced avatar based intelligent user interface is being developed for electronic commerce retailers to use for direct marketing on the Internet. ALife-Personal-Tutor is a natural language based tutoring program being developed to automatically adapt the difficulty and content of lessons to the skill level of the student which is dynamically analyzed using the natural language conversations between softbot and student user. ALife-Portfolio-Manager monitors an individual's investment portfolio according to concerns and criteria specified. When certain triggers are activated, it autonomously contacts the individual by telephone, pager, email or other means and alerts that trading actions might be warranted.