So far, none of the larger search engines have been able to capitalize on the promises of semantic search. Most of the innovations in the space so far have come from small start-ups and even those never made any real inroads in terms of market share when compared to the keyword driven search engines of Google, Ask, Yahoo, and Microsoft.

Powerset's technology might just give Microsoft the ability to differentiate its Live Search product from the competition.

It is encouraging to see that Microsoft has been able to integrate Powerset's technology into its own products this quickly. Live Search, which is far behind Google in terms of market share, needs exactly these kinds of features to make its search more relevant.

After the acquisition was announced, we wondered if a combination of Microsoft and Powerset could indeed beat Google. Judging from these first results of the Powerset integration, we can at least conclude that Microsoft will make a strong effort to beat Google in terms of search relevance. Whether this is enough to challenge Google's dominance remains to be seen.

Venturebeat reports that Microsoft might be close to acquiring the San Francisco based semantic search engine Powerset for about $100 Million. No announcement has been made yet by either party. We contacted Microsoft, but did not get an answer beyond "Microsoft does not comment on rumors or speculation." We will update this post once we receive more information.

Rumors about Microsoft's interest in Powerset had been swirling around the Valley since last month, when Dan Farber first brought up the possibility in a post on CNet.

Having backing from Microsoft could help the small company to expand beyond Wikipedia and start indexing more of the Internet. Powerset's technology is still unproven to work well for anything but Wikipedia, but if Powerset does manage to scale beyond this, then it would allow users to by-pass Google's keyword driven search in favor of just getting a direct answer to a large number of their questions.

Mircosoft's search products have struggled to gain any ground back from Google's search. Currently, Google has almost a 70% share of the search market, while MSN/Live Search has about 9.5%.

Powerset's capabilities have generally received very positive reviews and in his original piece on this, Dan Farber already argued that Powerset's ability to create connections between concepts, relationships, and meanings could give it a heads-up over Google's keyword and PageRank driven search.

We first reviewed Powerset vs. Google in May and at the time, Josh Catone's impression wasn't quite as positive and he concluded that "Powerset doesn't do a markedly better job of finding answers than Google for most queries."

Powerset was funded in a $12.5 Million Series A round by Foundation Capital, Founders Fund and various angel investors.

For a more in-depth look at the state of semantic search in general, see also Alex Iskold's article on the myth and reality of semantic search.]]>Discuss]]> http://www.readwriteweb.com/archives/rumor_microsoft_powerset.php http://www.readwriteweb.com/archives/rumor_microsoft_powerset.php News Thu, 26 Jun 2008 15:35:33 -0800 Frederic Lardinois Lately there's been a swirl of buzz about Powerset, a stealth natural language processing search engine. Last week they released their first "Query of the week". Today we discovered that Powerset is launching Powerlabs (screenshot below), plus we got our hands on a second query. Here are the screenshots, neither of which has been seen before elsewhere:

A screenshot of the Powerlabs interface [Ed: does that dashboard really have the P-word in its menu?!]

The Powerlabs program will take the Web community inside Powerset's development as an ongoing feedback and information portal, for groups of people who Co-founder Steve Newcomb terms "Powerlabbers".

Natural Search

Powerset is one of the most anticipated startups of 2007, but so little has really been revealed from behind their doors, and news coverage of the company essentially came to a standstill back in March. Powerset is a Silicon Valley company that received $12.5 million in series A funding back in November of 2006. The company founders - CEO Barney Pell, COO Steve Newcomb and Product Architect Lorenzo Thione - envision utilizing breakthrough technologies to provide more intuitive searches via natural language. The goal and end point are not dissimilar from those of Hakia and some others, but the technology and "middleware" will be vastly different, as we shall see. Hopefully our coverage of the Powerlabs program will enable us to differentiate Powerset from other search endeavors.

Labs

Powerlabs has been established to inform the Web 2.0 community and to gain feedback during the intermediate and subsequent stages of the engine's development. Powerset is following the track of other extraordinary startups in that they are working incrementally, meticulously and with feedback generated by the Internet community. A small group of people will test drive elements as they come online and the numbers of Powerlabbers will increase as scalability permits.

Now

The big news today is that "it is on it's way"; and we will bring you a ringside seat to developments as they happen. Mark Johnson, Powerlabs Product Manager mailed us today with Steve's latest and there will be weekly updates and goodies for Powerset fans. So for now, we wanted to provide you with the very first inside look at Powerset in development. We are looking forward to the design competitions, demo tests and a whole series of surprises that these "rocket scientists" have in store. You can sign up for Powerlabs here and also see a cool video sneak peek here.

The View

Be advised the image above is of the Powerlabs interface, but the Powerset engine will be much more typical of a search engine like Hakia (unless of course the community can come up with a better suggestion, and I already have one).

It is doubtful that Microsoft bought the company just to enhance Live Search. Possibly the plan is to replicate the Wikipedia solution, then incorporate Powerset into Internet Explorer. In this post we look at what the thinking behind the acquisition might be.

What Powerset is today

Given a set of unstructured information, Powerset applies Natural Language Processing techniques to extract concepts and the key semantic concepts out of the text. It then builds a semantic index (similar to Google's) as well as a conceptual graph of relationships between entities. This graph is typically expressed in RDF triples.

One of the Powerset innovations is surfacing of semantics to the user interface. The contextual gadget is overlaid to help navigate the unstructured information.

Many thought Powerset to be a generic semantic search engine, but its first product is limited to Wikipedia. It is not trivial to scale the technology to the entire web.

Why Powerset is Powerful

When semantic technologies emerged a few years ago, people started talking about how semantic web and/or semantic search might be a Google killer. The talk was supported by logic that semantic search can deliver more relevant results because it "knows" the content.

Industry realizes that isn't the case. Semantic search has no huge advantage over the statistical approach used by Google. We discussed this in the post Semantic Search - Myth and Reality.

What is powerful about Powerset? Precisely that it doesn't try to search the web as a whole. Right now, the solution works on Wikipedia, but the infrastructure is generic, so any other site could also be enhanced. The contextual outline developed can be used to navigate any content.

Instead of dealing with the whole web, the idea may be firstly to build solutions for specific sites.

Head-on with Google?

Powerset as it is today is no Google killer. At this point only something with huge traction and momentum would stand a chance.

In the search market, Google has a strong hold - potentially stronger if the Yahoo deal goes through. People are conditioned to Google: it's simple and, yes, imperfect, but it's good enough and the results are still better than Live Search.

If Microsoft bought Powerset with the goal to incorporate it into Live Search, then it's likely to be another acquisition to make little impact on the bottom line. In fact, the announcement on the Live Search blog states just that. The number one reason is acquiring talent; the second is the belief that NLP and semantic algorithms will be able to patch holes in today's search.

Today Powerset brings only interesting technology; it doesn't bring traction. So what were they thinking up in Redmond? There may be more subtle play, leveraging the fact Powerset works well on knowledge sets like Wikipedia.

Possibly Microsoft plans to deploy Powerset across its own sites, then perhaps incorporate Powerset into Internet Explorer.

Imagine going to Wikipedia and having a semantic overlay on each page. Now imagine scaling this experience across major information sources around the web.

Providing contextual, semantic experience allows Microsoft to retain eyes longer, shaving off the time people spend searching Google.

This is an important point because Google doesn't make money on search - it makes money on advertising.

Can Microsoft ever beat Google in Advertising?

The real problem Microsoft is seeking to solve is advertising. Until now the web has figured out two fundamentals for advertising - portals and search.

Portals show ads on each page; the more people browse the content, the more ads are shown and the more money is made. The search model emerged as an alternative, now more successful, path to advertising dollars.

With Powerset and other semantic technologies, there's another model: contextual information exploration overlaid on existing content.

If Microsoft can figure how to keep eyes off Google's home page, the game will shift dramatically. The browser is one of Microsoft's most powerful tools - and the default box is Live Search.

If Microsoft wants to win over advertisers, it might just do more with the browser. Incorporating aspects of Powerset's semantic navigator into the browser by default could be a game changer. This is not a straightforward play. A large company with bureaucracy and execution problems is unlikely to be able to merge semantics into the browser quickly and elegantly.

Conclusion

The Powerset acquisition is an interesting move by Microsoft. This hot semantic startup was on everyone's radar.

What can the plan be? It is doubtful that Microsoft bought the company just to enhance Live Search. Possibly the plan is to replicate the Wikipedia solution, then incorporate Powerset into Internet Explorer.

That is a bold play requiring exact execution - not the kind Redmond has shown lately.

What do you think Microsoft is going to do with Powerset? What are the other applications of this technology that you can think of?

Evri, a Paul Allen backed semantic search engine, is launching into a limited beta tonight. Evri was first shown publicly at the D6 conference. Evri's CEO Neil Roseman likes to talk about Evri in terms of organizing content instead of calling it a search engine. At its core, however, Evri definitely is a search engine, though it adds a very sophisticated semantic layer on top of its results that emphasizes the relationships between different search terms.

Evri especially prides itself for having developed a system that can distinguish between grammatical objects such subjects, verbs, and objects to create these connections. In his demo at D6, Roseman described the system as being similar to "an army of 7th grade grammar students graphing the Web."

Evri is entering in direct competition with a number of recent entries to the semantic search market, especially Powerset and Hakia. Powerset, however, only indexes Wikipedia articles, while Hakia tries to index all of the web, but focuses less on the relationships between objects and more on providing highly organized results for a given term.

You can sign up for invites to Evri on their homepage. The first wave of users should be receiving invites tonight.

For a more in-depth look at the state of semantic search, see also Alex Iskold's article on the myth and reality of semantic search.

Powerset does some interesting things with general queries, such as displaying "Factz," which is an ontology showing various concepts related to your query and how they relate to one another, or "Dossiers," which are a summary of key information about your query. Sometimes it yields some odd results (such as this query for "ants" for which the key finding is that ants are "a fictional race from the video game Crash Twinsanity.") However, the real promise of NLP search engines, in our opinion, is that users will be able to make search queries using natural language -- or in other words, by asking a question. So we chose a few questions at random -- things we knew Wikipedia would have answers for -- and threw them at both Powerset and Google.

Query: Who invented dental floss?

Powerset's answer for this query was curious. The number one result comes from the Wikipedia entry for dental floss and highlights this line: "It was around this time, however, that Dr. Charles C. Bass developed nylon floss." Charles Bass, however, is not the correct answer. Earlier in the same article is this line, "Levi Spear Parmly, a dentist from New Orleans, is credited with inventing the first form of dental floss." Why didn't Powerset find it? It's second results, which comes from a Wikipedia entry on scientific achievements from the year 1815, correctly highlights Parmly as the inventor.

Google performed poorly for this query. The same 1815 article is identified in the sixth spot on the results, with the sentence mentioning Levi Spear Parmly highlighted, but the first few results aren't even close. Even though that's not as impressive as Powerset's results, both would require a user to click through to the article to verify the answer (because Powerset returned two different answers), and is scrolling to the 6th spot really that taxing? Taxing enough to make you switch to a new search engine? Interestingly, this query set loose on all of Google does quite well, returning the correct answer in a link to a trivia site in the first result.

Query: What is the capital of France?

Not surprisingly, both Google and Powerset nail this one. Both point to the Wikipedia entry on Paris, France in the number one spot with the sentence, "Paris is the capital of France" highlighted.

Query: Where is Paris?

This is a fundamentally more challenging query, because there are a large number of cities and towns called "Paris" in the world. And not surprisingly, neither search engine gives what we would call a "perfect" result.

Both return the article on Paris, France first. On Google, that's followed but a handful of other articles about the city and one about Paris, Tennessee. On Powerset, the second article is about Paris Hilton -- um? -- followed by one about Paris, Texas, and in fourth place the most helpful article it could have returned, the disambiguation page on Wikipedia for Paris. (Oddly, with the question mark, the query returned "Paris, Missouri" from Freebase, and without the question mark it returned "Paris, Texas.")

On Google at large, the results focus almost exclusively on Paris, France.

It would seem that both search engines generally understand that "where is Paris" means that Paris is a place (though upon reflection, perhaps we could have been searching for the location of Paris Hilton...), but neither recognize very well that it could mean any number of different places.

Query: Who is Joey Tribbiani?

Both Powerset and Google correctly call up the article about this fictional character in their first spot, but Google actually does a better job of highlighting who he is. Compare:

• Google: After the 2003/2004 final season of Friends, Joey Tribbiani became the main character of Joey, a spin-off TV series, where he moved to L.A. to polish his ...
• Powerset: In the end of the series, Joey was the only Friend that ended up without a lover or a spouse, even though he is the one that dated the most women. ... Joey becomes good friends with an attractive female attorney named Alex, who, along with her husband, a travelling [sic] musician named Eric, is Joey's landlord.

Google has the name of both shows in which the character appears in their excerpt, while Powerset's excerpt is made up of information about the series' that only someone who already knew the character would understand (without clicking through to read the full article) -- and it doesn't differentiate between the two -- before the ellipses the excerpt is talking about "Friends" and after it is talk about "Joey."

Google at large also finds the Wikipedia article first with the same excerpt -- it also finds clips of the show on YouTube, and the actor's (Matt LeBlanc) IMDB entry, as well the official site for the spin-off "Joey."

Conclusion

This was really just a very quick and informal test, and we barely put Powerset through its paces. But our first snap impressions are that Powerset doesn't do a markedly better job of finding answers than Google for most queries. Some might argue that we didn't play to Powerset's strengths and frame our queries properly, or search for things obscure enough to notice any differentiation. But the promise of natural language search is that people don't have to learn how to search -- they can just ask questions as they normally would. We also can't expect that everything they're going to look for will be obscure and hard to find via traditional search engines -- more often than not, they probably won't be.

Powerset will have an immense uphill battle to make any sort of dent in the search market. Google controls 67% of searches in the US, and the top 4 search engines make up about 98% of searches. If Google remains "good enough," Powerset will have a hard time convincing people to switch. It will be easier to make a judgment about the company's future as a real Google competitor once it is crawling more than two sites, however.

What do you think about Powerset? Impressed? Not impressed? Let us know in the comments below.

Freebase

Daniel Hillis starts with a demo of Freebase, which aims to "open up the silos of data and the connections between them". Freebase is a database that has all kinds of data in it and an API. He shows a wagon wheel like UI of VCs, centered around John Doerr. He says it is basically objects and relationships between them. Because it's an open database, anyone can enter new data in Freebase. An example page in the Freebase db looks pretty similar to a Wikipedia page (or a Twine page). When you enter new data, the app can make suggestions about content. The topics in Freebase are organized by type, and you can connect pages with links, semantic tagging. So in summary, Freebase is all about shared data and what you can do with it.

Barney Pell is up next. Powerset (see our coverage here and here) is a natural language search engine. He says the system relies on semantic technologies that have only become available in the last few years. He says that Powerset has imported Freebase, to improve the database. He says the system can make "semantic connections", which helps make the semantic database. He uses the example of Hulk Hogan and the list of wrestlers he's defeated (Ric Flair, Randy Savage, et al). He says these connections comes from "the way the language is expressed". He says that meaning and knowledge gets extracted automatically from Powerset.

Twine

Nova Spivack is up next, regarding Twine. Our review yesterday covers this. Nova Spivack notes today that Twine automatically learns about you and your interests as you populate it with content - the "Semantic Graph". When you put in new data, Twine picks out and tags certain content with semantic tags - e.g. the name of a person. He says that an important point is that Twine creates new semantic and rich data. But it's not all user-generated. They've also done machine learning against Wikipedia to 'learn' about new concepts. And they will eventually tie into services like Freebase. Finally he compares Twine to Google, saying it is a "bottom-up, user generated crawl of the Web".

Panel Talk

Tim O'Reilly starts by asking whether all of these Semantic Web apps are available now? Hillis firstly notes that some of this technology is not necessarily Semantic Web. But to Tim's question, Hillis says Freebase is "solid alpha". Powerset has about 16,000 people already signed up, and you can sign up now. Twine "is usable today", but it's still in learning and testing phase. Spivack says it's now an "invite beta".

Tim then says that what ties these apps together is "semantics" (not necessarily Semantic Web, as Hillis noted). O'Reilly brings up Google, Flickr interestingness and how users can influence results (collective intelligence etc), but that it's usually passive and hidden behind silos. But these new semantic apps are more open and they're platform players. Hillis says that O'Reilly is on track, and eventually "there will be one Web of data". He says Web 1.0 was a "web of documents", and that it will be the same with semantic apps - it doesn't make sense to have silos of data.

Nova said this is the value of open standards and the WC3. He says the Semantic Web is a certain set of standards, and that it where you get the (open) network effect. Barney says the real value is "making explicit what was once implicit" (in terms of data).

Tim asks: where is the interoperability within the 3 platforms (3 apps above). Hillis said Freebase is a platform because it's specifically designed to be used by other apps.

In summary, Spivack notes that data portability and connectibility is the key to these new semantic apps - the Web is the platform and they're just different services within the platform. Hillis though disagrees - he said Freebase is a platform! So the web is the platform of platforms (!).

On that overly semantic note, the panel ended.

Related: Web 3.0: When Web Sites Become Web Services, by Alex Iskold, which is a great overview of this new era of Semantic apps.

We also discovered that Cognition, which currently licenses its technology to other organizations, is planning to build a general consumer search engine - which will compete with Google and others.

A Semantic Map is kind of like a dictionary, in that it's a representation of Cognition's ability to define things. Cognition claims that its Semantic Map has over 10 million semantic connections; over 4 million semantic contexts (word meanings that create contexts for specific meanings of other related words); over 536,000 word senses (word and phrase meanings); 75,000 concept classes (or synonym classes of word meanings); 7,500 nodes in the technology's ontology or classification scheme; and 506,000 word stems (roots of words) for the English language.

The company says that its Semantic Map "is more than double the size of any other computational linguistic dictionary for English".

Cognition Technologies has been working on its technology for 24 years, with a lot of input from lexicographers and linguists over that time. Because they've used a mix of algorithms and human input, Cognition has been able to discern relevancy, meaning, synonymy. Scott Janus told us that one of Cognition's strengths is that it can disambiguate words and phrases, which Janus says differentiates them from the keyword and pattern matching algorithms of Google, Yahoo and others.

For example Janus told us that Cognition's technology can find results even if direct words are not used - which he says Google can't do.

Cognition Plans General Search Engine

The comparisons to Google led us to ask the obvious question: does Cognition's semantic technology have a more general application? In other words, does Cogition plan to take on Google by creating a search engine for consumers? CEO Scott Janus replied that yes they do plan to "one day offer search on the general web". However he said that they need more capital funding to index the entire Web, put infrastructure in place, etc.

As of now Cognition will continue to license its semantic technology to verticals like law and health. Janus told us that Cognition is "good for complex content where lot of synonyms are used", so right now data-intensive industries are where it is aiming.

Cognition's current applications include legal (e.g. LexisNexis Concordance's case management), health (e.g. MEDLINE), and a semantically charged version of Wikipedia.

Image from Cognition

Cognition vs Powerset and Hakia

Two other Semantic search engines we've been tracking closely on ReadWriteWeb are Powerset and Hakia. We asked CEO Scott Janus what makes Cognition different from those two products?

In a nutshell, Janus says that its Semantic Map is bigger and better.

Specifically, he said that Powerset is actually "not so similar" to Cognition. According to Janus, Powerset does "parsing" - which it licensed from Xerox Parc. That is 20-25% of the solution, said Janus, but Powerset "doesn't have a good semantic map". Cognition went so far as to write a white paper (pdf) explaining why it thinks Powerset "misses the point".

As for Hakia, Janus said that as far as he can see Hakia is focused on "ontological classifications" - classifying words and concepts together. But he says Hakis doesn't have as full a semantic map as Cognition, so he thinks Cognition has "a better understanding" compared to Hakia.

In summary, Janus told us that semantic search companies "must include a comprehensive semantic map" to be successful. We're sure that Powerset and Hakia will have different opinions on what makes a successful semantic search company, but it does make for a good differentiator for Cognition.

Open Question

Tell us in the comments what you think of Cognition and whether you think it can compete with Google in the long run?

Semantics is expressed meaning in language, code or "other" representations of information. My discussions with Barney and Melek revealed the fundamental differences in architecture and philosophy between hakia and Powerset. The index systems of the two companies are fundamentally different, as is their philosophy - but their goals and visions are remarkably similar. They are also different in the way they apply what I term horsepower to natural language search. Like the symbolism of Shelby vs. Ferrari,– it is possible for different approaches to achieve a desired result - given enough horsepower.

Hakia has built their search in-house, refining and sculpting the QDex indexing system (like an Enzo Ferrari). Their view is that processing power should be maximized with super efficiency, via fuzzy logic and advanced semantics. Powerset, on the other hand, utilizes basically the same inverted indexing system as Google - but backed by natural language and immensely powerful processing that essentially “overpowers” the long tail query (like the GT 500). This is a vast oversimplification, but the elements involved reveal the larger story.

Technology (horsepower), communication (language) and people make up the semantic Web. The Web has not been lacking "language", but the adequate application of processing power. As Barney said: "Even five years ago we did not have the processing capability to even attempt this, but five years from now these answers will seem elementary." Google's system below, currently consumes massive horsepower with comparatively limited results - at least according to hakia and Powerset!

Diagram of Google's inverted index and search (courtesy -changturtle)

Unbending Humans

Barney described the relationship between people and computers as people being "bent" around or adapted to technology in order to utilize it. With the advent of services like Facebook, programs and applications are beginning to “understand” each other. Everyone reading this has been “forced” by technology to conform to varied “bending events”, in order to use it. Barney explained this idea by calling Facebook and the iPhone true innovations approaching total “community engagement.” Barney also said that “Facebook will become one of the primary communications platforms of the future.” Given this new perspective, I could not agree more because Facebook is one heck of a representation of information for a social network. Essentially, hakia, Powerset, Facebook and others are bending the machines to engage humans. And in a way, Facebook is the semantic Web in a microcosm - but in it's infancy.

Semantics and Search?

Search is a critical part of our daily lives, but the interface has changed very little over the years. We define search as the act of typing in a query on Google and getting results. This is a type of search, but how many other kinds of “searches” do we perform? In an earlier article, Josh Catone wrote about Yahoo!’s contention that search will not determine the future of the Web. Josh rightly asked if Facebook and MySpace might be better positioned if “personalization” was to be the future of the Web.

Conclusion

I should make it clear that neither Barney nor Melek really consider themselves as "Google Killers". Powerset and hakia are not in a race either against each other or to overtake Google, but they are on a quest for better Web communication and engagement. Both efforts emphasize the necessity for “the system” to be able to universally understand and handle data without ambiguity. Viewing Facebook and others as functional repositories of semantic data is essential in seeing the long view. Whether we are talking about object oriented data, textual semantics or complex algorithms, the semantic Web is about making people “bend” less for technology.

The panel began with a discussion of the mainstream and the Semantic Web. Everyone agrees that there is a big need to simplify messaging to the masses. Marketing the Semantic Web to consumers does not make sense because they don't care about technical specifics. People do not need to know about RDF and Microformats, what they are looking for are simple and tangible benefits that make their lives on the web easier.

Carla points out that we need to invent new terminology to describe the transformation and the new technologies that are being developed. She suggests to use the term "Smarter Web." This is a new web where people can find things that they are looking for faster, and experience more intelligent, contextual interactions with the vast amount of information out there.

What is the Killer App?

The discussion shifts from what is wrong with Semantic Web marketing to what is the "killer app" for the Semantic Web. Tom Tague says there is no killer app, and there really can't be. He thinks of semantic technologies as the spice, the infrastructure that will give rise to a family of semantic web applications. Tom argues that what we are seeing is pieces being brought slowly to market, which make search incrementally better, make browsing incrementally better, etc.

Naturally, the discussion shifts to the impact on Google. For better or worse, people expect the killer app for the Semantic Web to be a Google killer as well. As an example, after the recent launch of Powerset it has been constantly compared to Google. [Including on this blog -- Ed.] Mark Johnson points out that this is not reasonable, because Powerset is simply not yet there -- right now it can only search Wikipedia.

So coming full circle to Tom's point, if we're only going to get incremental improvements, the question is then how much effort is warranted. This is not a minor point given that the bar is set high because companies themselves and the media are creating lots of hype around the Semantic Web.

PR and Hype Around the Semantic Web

The goal of Powerset, for example, is to change the way that people interact with computers. Carla points out that marketing that claims large goals, like to change human-computer interactions, is quite ambitious. Maybe this messaging needs to be incremental just like the progress that is being achieved. As an example of adjustment in messaging, Carla suggests that it could simply be: Powerset today provides a better way of searching information in Wikipedia. As more tools are get rolled out, the messaging can adjust, and then once all the bits are in place, switch to the messaging about changing the way that people interact with computers.

An interesting twist to the whole conversation is the need for larger Semantic Web players to justify longer terms plans. Powerset, for example, is a 60 person company with > 20 employees that have PhDs in computational linguistics. Venture capitalists don't want to hear that all the company will do is to change the way that people search Wikipedia. So Powerset has been forced to justify a longer run and bigger staff by putting broader goals on the map.

Further, the messaging tool is the media, which is becoming increasingly more hype-based. Companies are being provoked to describe their broader vision, to appear big and ambitious in order to get coverage. So there is a conflict between market messaging, investor messaging and a realistic, long term plan to deliver semantic technologies to the market.

Ease of Use

Beyond marketing, everyone agrees that if semantic technologies can deliver tangible consumer benefits, then marketing problems will not be as acute. It is the tension between loud marketing and the modest forward progress being made that is leaving everyone underwhelmed.

And so, the panel agrees that Semantic Web companies need to pay more attention to the user experience, and to ease of use in particular. Tom Tague says that people don't care what is underneath. Infrastructure does not excite people, but UI does.

Conclusion

Panelists agreed that despite the challenges, companies are making definite progress. They also agreed that it would be good to tone down marketing, remove the grandiose Semantic Web slogans, and focus on specific consumer utility.

Is this true? We sought a response from both Hakia and Microsoft-owned Powerset on this semantically charged question.

Hakia CEO Riza C. Berkan responded in the comments to the original article that "hakia is deploying Ontological Semantics (OntoSem)", which he described as "a network of concepts reflecting ontology." He went on to say that hakia covers "over [a] million words in English".

However Berkan noted that the size of a Semantic Map does not necessarily matter: "the sheer size of the collection of words or concepts does not represent, by any means, the capability of the system." Hakia's position is that "there is no silver bullet for a semantic solution that will succeed", as long as the system developed is scalable and imposes "minimum reliance on 'words'".

Semantopoly: Advance token to nearest Semantic Context

At this point we were still confused. Cognition uses the term "semantic map" and said it was necessary to have. One of the commenters on the original post agreed with that assumption. Yet Hakia's Riza Berkan didn't use the term "semantic map". So we asked Hakia in a follow-up email, does it or does it not have a semantic map? Dr. Christian Hempelmann, Hakia's Chief Scientific Officer, responded:

"The term sometimes comes up in the context of data integration, but "Semantic map" is not a term used in linguistics. I can only speculate that it is what is commonly called an ontology. To the degree that they let us on about it in the documentation on their website, Cognition operates with only 2 main relations, much like WordNet: hyperonymy/hyponymy (e.g. cat is-a feline is-a mammal; their "taxonomy") and synonymy (e.g., "buy" means almost the same as "purchase"; their "thesaurus"). Furthermore, this map is not independent of English, cannot grow into other languages. hakia, on the other hand, has an ontology with many more relations, effectively raising our "semantic map" to the size of a higher power, and can and is already growing into other languages."

We also tried to get a comment from Powerset, but as of writing we haven't received it.

So, are we all clearer now on what is a Semantic Map, is it needed, and does size matter? Er, it depends. If you think you know the answers, tell us in the comments please!

Suffice it to say, while I got by, I missed Google and Yahoo!. Lexxe says it is powered by natural language processing and urges searchers to ask questions in plain English. It then attempts to give you an answer based on what it determines are the top results. Unfortunately, more often than not, the answer is gave just wasn't right, and the results it gave weren't much stronger than Google's.

So the results, when I studied them a bit, weren't terrible, but were they really enough to make me switch from Google? When I tried my first query in Google this morning, the first result I got was Wikipedia's Six-Day War entry, which would lead me to Eshkol. And the fourth result highlighted text on the search results page mentioning Eshkol as prime minister. In general, though, I wouldn't ask Google a direct question, I'd likely start with something like: israel's prime minister 1967 war, which interestingly provides basically the same results as the question.

After being reminded by my Google query that the 1967 was also known as the Six-Day War, I tried to give Lexxe one last test, repeating my first query, but replacing "1967 war" with "Six-Day War." The answer was unintelligible, but the very first result it gave was titled "Levi Eshkol" and displayed a snippet talking about him being PM during the war. Google's results for the same query were not so good, with the first mention of Eshkol in slot 5, and a snippet that wasn't very clear as to who he was. Finally, a win for Lexxe!

So what does that mean? To me, it shows that meaning-based and natural language search isn't really ready for prime time. The answers Lexxe suggested were almost never right, and though the results could eventually lead to the answer, they rarely did so any quicker than Google -- and often required tinkering with my query, the same as other search engines, so didn't save me much time. (I also tried the above queries at Hakia, but with an even less positive result: it suggested Golda Meir and David Ben-Gurion, the prime ministers before and after Eshkol, for my original question).

It would seem that Google understands natural language queries just as well as these other search engines, right now. In some cases, Google actually understands natural language better (for example, I can ask "What is 3 + 5?" and get 8, Lexxe doesn't understand what I mean). One of the most hyped natural language search engines is the yet-to-be released Powerset. On Monday, TechCrunch posted the first public screenshot of a Powerset results page. The screenshot showed the query "politicians who died in office" (without quotes) and Michael Arrington was impressed:

But for "politicians who died in office" the results on Google won’t be as good. Context is required: Google has only six results for the query in quotes, and without quotes it loses its meaning and the results aren’t useful (notice the Powerset blog is the fourth result). The Powerset results are relevant and useful.

But as many TechCrunch commenters pointed out, the Powerset results aren't really that helpful, just providing a seemingly random list of sites about various politicians who died in office. With or without quotes, however, Google points users to a site called The Political Graveyard that lists every US politician that died in office with the first result. So it would seem that Google's results are still better.

Conclusion

So my day without Google demonstrated to me just how far alternative search engines (at least those of the natural language set -- which is one of the most hyped variety as potential "Google-killers") have to go to catch up with Google and the other big 5 search engines. How was your day without Google? Share your experiences in the comments below.

Powerset is a much-hyped semantic search engine that uses natural language processing to "understand" concepts in web content and match pages to queries. Right now it only searches Wikipedia (and Freebase). We put it through some early paces last week.

Wikiwix

Wikiwix calls itself the "ultimate Wikipedia articles search engine." It searches all of the Wikipedia sites at once (i.e., Wikiquote, Wikiionary, Wikinews, etc.) and has a very handy Wikipedia image search.

AskMeNow

AskMeNow is a mobile-targeted Wikipedia search engine that does some natural language processing similar to Powerset and then attempts to cull your answer directly from Wikipedia. Like any NLP search, it's not perfect, but often enough it is right on the nose.

Similpedia

Similpedia lets you find related content on Wikipedia. Paste a URL or a paragraph of text and it will dig up articles on Wikipedia that are in some way related.

Gollum

Gollum is a Wikipedia browser that supposedly "[reduces] the complexity of information" and makes it easier to browse the online enclyclopedia. To be honest, though, we can't really see any benefit over just browsing Wikipedia in Firefox.

Qwika

Qwika doesn't just search Wikipedia -- it searches wikis. 1,158 of them. Wikipedia is included in those it searches, however, and the site makes it easier to search across multiple languages.

WikiMindMap

WikiMindMap is one of the coolest Wikipedia search mashups out there. Enter a search term, and the site will generate a mindmap based on related Wikipedia entries allowing you to easily explore a topic and its related articles in full.

Wikiwax

Wikiwax gives Wikipedia search the AJAX suggestion treatment. Get search suggestions while you type and find that Wikipedia article a fraction of a second faster.

Lexisum

Lexisum takes Wikipedia articles and summarizes them to a smaller, more digestible format that are better set up for printing. You can choose from a number of standard print sizes to display your article summary (A4, A6, etc.).

Ask.com & SearchMash

Ask.com and SearchMash (a test sandbox for Google) each augment their search results with information from Wikipedia. Not a pure Wikipedia search, but interesting stuff from a couple of major search players.

Bonus site: Wikirage

Wikirage is something like Google Trends for Wikipedia. The site shows trends on Wikipedia based on edits. Hot this week for example, the Sichuan earthquake and American Idol. We gave the site a full review last August.

The mistake is that semantic search engines present us with Google-like search box and allow us to enter free form queries. So we type the things that we are used to asking - primitive queries. It never occurs to us to type in What actor starred in both Pulp Fiction and Saturday Night Fever? or What two US Senators received donations from a foreign entity? We type simple questions, but this is not where the power of semantic search lies. Lets look at the spectrum of semantic technologies from Google, to SearchMonkey, to Powerset, and Freebase to understand what is going on.

What Problem Are We Trying to Solve?

The first confusion in the space comes from the fact that semantic search is being positioned as the answer to all possible problems - from modern search, currently dominated by Google, to problems that are computationally impossible. The situation is made more difficult by the fact that right now there is only a thin range of problems where semantic search can clearly do better. This range is complex queries involving inferencing and reasoning over a complex data set.

As shown in the diagram above basic queries are easily handled by Google. Sadly, natural language processing gives little advantage when it comes to this category of problems. Google correctly answers the question about Leonardo Da Vinci's birthday leaving no opportunities to improve the search by understanding the nouns and the verbs that user typed in.

Before looking at the problems that are perfect for semantic search, lets look at the hardest problems. These are computationally challenging problems that really have nothing to do with understanding semantics. The misconception has been perpetuated since early days of the Semantic Web that somehow, because we will annotate the web, we will be able to solve these super complex problems. This is simply not true. There are fundamental limits to what we can compute, and a class of problems that have an exponential number of possible solutions is not going to be magically solved because we represent data as RDF.

The good news is that there is a set of problems that are great for semantic search. These are the problems we have been solving so wonderfully with relational database. Way too often we forget that semantic technologies are here to help us represent relational data spread over the entire web - so it should be no surprise to us that it is relational queries that semantic search engines would excel at.

The Spectrum of Semantic Search Players

But semantic search is not just about the questions that we are asking. Because the web is just a bunch of unstructured HTML pages, semantic search is also about the underlying data. At its most structured extreme we find Freebase - the semantic database of everything. Freebase is accessible via free text search, but more importantly via MQL (Metaweb Query Language). MQL is essentially JSON with wildcards. Using it you can construct any query against Freebase and the result will be the same query with answers filled in.

Powerset, in a way, is just a relational database. It operates against certain, structured information. On the other end of the spectrum is Google, which is all about statistical frequencies and very little semantics. The recently launched SearchMonkey from Yahoo! is an interesting twist. It does not add anything to the result set, but instead uses semantic annotations to present a richer, more interactive and useful user interface.

Companies like Hakia and Powerset are probably working the hardest. These companies are trying to simultaneously build Freebase-like structures on the fly and then do natural language queries on top of them. The difference is that Hakia is using (likely similar) technology to query over the entire web, while Powerset has (probably shrewdly) chosen to restrict the search to Wikipedia.

Are Hakia, Powerset and Freebase All That Different?

This analysis brings up a question - which of these technologies are different and which are essentially the same? Lets get the easy one down first. Yahoo!'s SearchMonkey is no different from Google or any other search, as far as the core search technology is concerned. The difference is simply in the presentation layer. SearchMonkey is smart about creating a better user experience by letting publishers present the search results to the users in the best possible way.

But when it comes to Hakia, Powerset and Freebase the situation is much more complicated. On the surface all these products are different - Hakia lets you search the whole web, Powerset is restricted to Wikipedia (and Freebase!) and Freebase itself has two search interfaces - the search box and query language. Here is the problem - the natural language interface has nothing to do with the underlying data representation.

The fact is that all of these semantic search technologies allow people to type in arbitrarily complex questions and then interpret these queries and execute them against their databases. Fundamentally, Hakia, Powerset, and Freebase are databases. Fundamentally, all of them have some kind of Natural Language Processing that translates the question into a canonical query over the database.

To gain insight into all of this, think about Freebase and its query language MQL. Unlike natural language, which allows all sorts of constructs, MQL is non-ambiguous. This JSON-like language allows users to construct precise statements against Freebase. The fact that Powerset allows natural language queries does not mean that inside Powerset there is no database. For sure, though, there is a similar kind of database as there is beneath the Freebase search box. What is really different about Freebase and Powerset is the data gathering approach and user experience.

Back to the Future: It's All About UI

Probably the most striking revelation about the semantic search space is User Interface. First, to go on the tangent, Powerset got it right by realizing that semantics needs to be surfaced in the UI. After a user searches Powerset, a contextual gadget, aware of the semantics of the results, helps the user complete the search experience.

Yet the biggest mistake that I think Powerset is making is also in the UI. The search box that everyone is familiar with via traditional web search engines needs to go. Having a simplistic search interface hurts Powerset and Hakia, and to a lesser extent Freebase, which is not positioning itself as generic search.

Think about the recent launch of Powerset. The company released a vastly better way to interact with one of the most important sources of information on the web - Wikipedia. But what did the critics say? Lets see if this is a Google killer. And the answer to that is "no."

But what if Powerset restricted what can be searched? What if instead of a search box there was another interface or what if they told users not to look up things that they can find easily on Google? Why is it that new companies are expected to improve on the algorithm that has ruled the web for over a decade? Instead, the expectation should really be to solve the problems that can not be solved by Google today.

Conclusion

Semantic search is an upcoming technology that has set the expectations way too high. We have all been misled into thinking that these technologies are here to dethrone Google by delivering better search results. Neither of those things are true. What is true, however is that semantic search is going to be big and it is going to help us answer questions that we simply cannot answer today - complex, inferencing queries asked over the entire web as if it was a database.

In order for these semantic search technologies to make a dent in the market, they need to clean up their messaging and most importantly, their user interface. Presenting a search box is both misleading and detrimental, as people associate it with the simplistic questions that Google solves without any problems. To really showcase semantic search, these companies need to come up with innovative UIs that will help users to understand the power that is being put at their fingers.

As always, please tell us what you think. What should semantic search companies do to gain their place in the marketplace?