Accessing the Web of Data through Embodied Virtual Characters
Review 1 by Kunal Verma:
The authors present a very interesting approach for presenting SW data to users - by having an agent intermediate between an user and the SW. The challenges presented by the authors are to the point and well articulated. However, given that this problem is analogous to automated question-answering and is will be hard to tackle, I have a couple of suggestions for the authors:
1. Can they add a motivating scenario(s) will illustrative examples to articulate the kinds of interactions that may occur between an user and an agent?
2.Can they outline what kinds of advances in the field are needed for different kinds of interactions?
Review 2 by Tom Heath:
I really enjoyed reading this paper - it's refreshing to see researchers approaching the issue of interacting with Linked Data and the Semantic Web from novel directions. I do have some outstanding questions about the research direction and it's tractability, but that's the point of a position paper, to promote discussion.
My main issue relates to the question of whether this is a tractable approach. Reading the paper it reminds me of the anecdote about Marvin Minsky's instruction to students to solve the problem of computer vision over the summer vacation. Are there just too many outstanding issues here to make conversational access feasible? That doesn't mean the authors shouldn't try, but it would be useful to gauge the authors view on how long-term/significant this challenge is. What are the risk? And the opportunities if successful? I guess I want to see a bit more 'vision'.
One thing that would help ground the position is a fuller discussion of the fundamental literature that drives the intuition (is it more than intuition?) that this is a worthwhile approach. Is the focus on this modality arbitrary or does it have a sound empirical base? If so perhaps this could be more fully exposed. If not, perhaps the authors could surface more of the case study in reference ? A motivating scenario would also be useful, e.g. information seeking while driving, via a mobile device with handsfree (dangerous perhaps, but potentially useful).
On a related note, how would this mode of interaction avoid the hatred people had for the "Clippy" in Microsoft Office? What are the precedents in the literature? What mistakes were made in that adjacent domain that would need to be avoided here?
I had a few specific comments, working through the paper:
1. Is there a reference for the claim about "several orders of magnitude" bigger in Section 1?
2. The Marbles browser by Becker and Bizer (I think it's described in the paper "Exploring the Geospatial Semantic Web with DBpedia Mobile" from JWS in 2009) may provide a good reference point for the questions raised in the bullets in Section 1 about trust, provenance, etc. The Marbles approach doesn't solve it, far from it, but may provide a good anchor point for comparison.
3. In "Conversation as interaction paradigm", is the way people interact with one another a sound basis for designing human-computer interfaces?
4. I really like the idea of hedging gestures and facial expressions to convey uncertainty. No idea if it would work, but an intriguing idea.
5. The paper "A survey of trust in computer science and the semantic web" by Artz and Gil in 2007 may provide another good reference for the topic of trust in the Semantic Web area. Also, the point about conveying trust beyond just a numerical value is a really important one.
6. Re other references, the Bizer, Heath and Berners-Lee paper "Linked Data - the Story So Far" provides a better account of the LOD project than linkeddata.org, for those not familiar with the history. At the risk of recommending more of my own papers as references, I think my paper from this issue would be very relevant as backdrop, as would my 2008 IEEE paper "How will we interact with the Web of Data?"