Related press releases
Related research
Beyond Google
:: 17 August, 2007
Power set, a reestablishment made of San Francisco, wants to bring an innovative search machine on the market, which can be served with natural speech in September. The software concepts which are under it are based on linguistic research, which was made by decades at the renowned IT-research center Palo Alto Research center (PARC). Power set is to be able to do more than only the acceptance of search words in form of natural questions. The best search results are to be determined also by the fact that the meaning and the context of an inquiry are included - including Websites, which stand to it in relationship. “Power set extracts the more deeply lying concepts and relations from Web texts and user inquiries and brings these efficiently together”, says power set boss Barney Pell.
Although such ideas are already for a long time pursued, one wants to have developed a system, which solves the fundamental technical problems with such search models “finally” with power set now. The product come out thereby should be nevertheless economical IT-technically fastidiously.
Pell wants no specific technical break-through to call, which led to the development of power set, but seizes one on the 30 work of the PARC researchers back (an appropriate license acquired the company in February). Not an individual piece of technology solved the problem, but the combination of many theories and fragments. “After 30 years the research concerned here finally on the point, at which it can be brought into the world”, means it.
A core component of the search machine is a system for the processing of natural speech, which extracts the relations between words. It developed for a software platform, which was developed at the PARC from “Xerox the Linguistic Environment” (XLE). This platform is based again in the model of the lexical functional grammar in such a way specified, with which different grammar Engines can be provided, which can help a search machine to understand text. According to Pell these algorithms can deal for example better than other beginnings with ambiguities, in order to understand the actual meaning of a sentence on a web page. All these innovations are to make the system more flexible.
Power set technology boss Ron Kaplan was since the seventies of technical managers XLE team in the PARC and is an author of large parts of the technology, which became licensed to the start UP now. He worked for the first time together with Pell before two years on the idea to get the technology in Internet. Current search machines set rather on key terms and covered contents only superficially. “There there is area for improvements”, means Kaplan. Particularly relations between contents parts was hardly understood: “The best, which our competitors can do here, relations is to be derived on the basis words, those near other words lies.” It is necessary to put on a substantially deeper analysis yardstick.
Earlier attempts combined here the acceptance of retrieval queries in natural speech with standard search models on the basis key terms. This can be seen for example with Google, if the search machine makes a new suggestion for the user, because the current inquiry was not understood. Also Yahoo uses the recognition of natural speech partially. One completely on this technology which is based search machine did not exist for final customers so far however yet. According to Pell one of the principal reasons for the fact lay in the fact that the technology was simply so far not yet available.
Also power set competitors like iPhrase and EasyAsk, which likewise place their understanding of natural speech into the foreground, are to be able to process text contents less well like power set. “Only data bases are scanned also here for an answer to a question.” Still more strongly on the recognition of natural speech based beginnings such as Hakia and Cognition search would possess however only a smaller meaning understanding, as Pell means.
A demo version of power set is in September on special “power labs” - Website to be published. With the user feedback won there one wants to then place the final product in the next year completely. “The main challenge is to advance the system so far that the users understand, like it it to use can and it them despite existing small error increase in value supplies.” Its enterprise would stand briefly before this point, means Pell.
Also with IBM one works on similar projects. A new semantic search machine named Avatar is at present in the beta test within the IT-company. The project contacts however particularly enterprise customers. Project manager Shivakumar Vaithyanathan sees the most difficult problem in drawing important semantic information from large documents without precision and speed suffer from it.
The IBM search machine is to help particularly when scanning internal documents such as enamels and Intranet correspondences. It is for cases optimized, with which certain partial information is looked for, to find otherwise only with difficulty leaves itself - for instance a telephone number or package pursuit URL, which are enamel, which has a person on her computer in one of thousands.
Avatar sets thereby on the creation so mentioned “interpretations” of entered search words, which describe the actual search intention as model. If the user enters for example to “telephone number”, the search machine scans thousands of enamels of a user for numbers, which remind of telephone numbers. The search machine supplies then the information, which it - and not simply only enamel, which contains the search word, looks for.
In order to draw as fast as possible all meaningful information both from the scanned texts and from the retrieval query, much computing power is necessary. IBM wants to find now a way to find accurate meanings faster and with fewer servers. “If we information better extract, can we the questions, which place the users, also better answer”, mean Vaithyanathan.
Tags: future search engine , social search engine , google , web search , software , Power set ,
