Lately, the media has directed a whole lot of attention to the upcoming Wolfram|Alpha, a “computational knowledge engine” that seeks to answer any factual question posed by users in the form of natural language that some say could end up competing with Google on at least a portion of user queries.

Yet, projects of this kind are not at all new to the market: Douglas Lenat’s Cyc and True Knowledge, LLC are just two of the most well-renowned “answer engines” in the market that Alpha will be competing with.

Wolfram|Alpha

Stephen Wolfram, a reputed child prodigy and the creator of the excellent Mathematica software for symbol manipulation, managed to keep this years-long project away from the eyes of the media up until March, when he finally announced in a blog post that his company would be ready to release Alpha later this month.

The Wolfram|Alpha homepage

What’s interesting about this project is that, in the words of its founder, it attempts to “create” knowledge by combining the various pieces of information in its databases, rather than returning closest matches from indexed documents from the Web like Google and all other traditional search engines do.

To achieve this, Alpha needed to develop fairly sophisticated language parsing algorithms in order to understand the meaning of user queries. Wolfram said that the set of parsing algorithms used was chosen from “the space of all parsing algorithms” and that, following that road, the team thinks it has made breakthrough advances in the field of natural language parsing, which is possibly one of the hardest subfields of artificial intelligence.

After parsing the queries, Alpha returns users information in an appealing form — often a graph — that allows for an easy comparison and analysis of the information contained in its databases across a number of domains.

Wolfram said that chemistry is probably the field that Alpha “knows more” about, while other bits of knowledge will be input by users themselves as they use the tool, similarly to what they would do on Wikipedia, except a) they will have to provide reliable sources and b) changes will likely not be reflected immediately in the knowledge base, but rather checked first by Alpha’s staff of domain experts.

However, the concept of “knowledge” in Wolfram|Alpha seems to be severely limited by one factor: by Wolfram’s admission, the project’s knowledge base doesn’t currently organize the objects into an ontology, which makes it impossible to make a very interesting kind of inference across its KB.

Can a can can-can?

For instance, lacking an onthology means not being able to answer questions like “is a pea a vegetable?”, or even one of my favorites: “can a can can-can?”, irregardless of the parsing capabilities that Alpha might have. In other words, all that Alpha really knows about the items in its KB is given through independent, domain-specific scripts, but it is unable to classify and organize the knowledge it collects in a meaningful way.

True Knowledge, LLC

True Knowledge is another interesting project that is both very similar and  very different from Wolfram|Alpha.

It is similar to Alpha in that it also aims to answer factual questions based on knowledge contained in its databases; but it is profoundly different in the way it gets there.

Unlike Alpha, True Knowledge does have an onthology, which means that it can answer a much wider range of questions even with a smaller knowledge base (see picture above). The user interface may not be as easy and the knowledge base not as well-curated as Alpha’s, but chances are that, without an onthology, Alpha will never be as powerful as True Knowledge.

Computing distances in natural language

This service is still in beta, but you can become a user (and contributor) by signing up on their website. Once you’ve been accepted, you can start testing the platform and even add new bits of knowledge in real time, in full accordance with the wiki spirit: other users can then corroborate or deny any previously assessed fact and lead the platform to believe a statement to be true or false with a certain degree of confidence based on all the evidence at its disposal and the inference it can make based upon it.

Finally, I find their approach to parsing quite interesting since, at least according to this Wikipedia article, TrueKnowledge “attempts comprehend posed questions by disambiguating from all possible meanings of the words in the question to find the most likely meaning of the question being asked.”

A final word

Using my status of web journalist, and being search engines one of my main focuses, I have just asked the staff of Wolfram|Alpha for a media preview of their platform. I’ve also asked them whether they plan to add an onthology to their framework anytime soon. I will post an update ASAP — stay tuned.