Skip to content
The methodOpen data
Back to the collection

Teaching Machines to Read and Comprehend

Karl Moritz Hermann, Tomáš Kočiský, Edward Grefenstette

Show all 7 authors

Lasse Espeholt, Will Kay, Mustafa Suleyman, Phil Blunsom

2015 · arXiv

Machine learning
Read the paper
HN score
0
Mentions
4
Readers
3
Months
4
What’s in a score?

From the abstract

Teaching machines to read natural language documents remains an elusive challenge. Machine reading systems can be tested on their ability to answer questions posed on the contents of documents that they have seen, but until now large scale training and test datasets have been missing for this type of evaluation. In this work we define a new methodology that resolves this bottleneck and provides large scale supervised reading comprehension data. This allows us to develop a class of attention based deep neural networks that learn to read real documents and answer complex questions with minimal prior knowledge of language structure.

Metadata: arXiv. Abstracts retain their original authors’ rights.

In the conversation

Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels

4 selected excerpts · 4 total mentions

nl
Neutral mention · rule
It's true that we (as in everyone working on these problems - not just IBM) don't have this working. Two points, though: 1) This is probably the most active area of research in machine learning/AI at the moment. The same type of progress that was made in image understanding 2 years ago using deep learning techniques is now being seen in NLP. See for example[1]. 2) It isn't at all clear that "making connections" of the type he talks about is needed to generate the types on insights that he claims. For example, the work StitchFix is doing combining analogy reasoning on words along with other non-textual data[2] indicates to me that the current state-of-the-art absolutely…
Read the full comment
could "understand" that Bob Dylan wrote about themes that were associated with anti-war sentiment. [1] "Teaching Machines to Read and Comprehend" http://arxiv.org/abs/1506.03340 [2] http://multithreaded.stitchfix.com/blog/2015/03/11/word-is-w..., https://github.com/stitchfix/context2vec,
Original comment IBM is not doing "cognitive computing" with Watson
deepnet
Neutral mention · rule
non pdf links to arxiv abstracts: Teaching Machines to Read and Comprehend by Hermann, Kočiský, Grefenstette, Espeholt, Kay, Suleyman, & Blunsom http://arxiv.org/abs/1506.03340 and (same group) Reasoning about Entailment with Neural Attention http://arxiv.org/abs/1509.06664 Learning to Transduce with Unbounded Memory http://arxiv.org/abs/1506.02516
Original comment Teaching Machines to Read and Comprehend [pdf]
paulsutter
Neutral mention · rule
The original Deepmind paper [1] is based on a really smart idea. Algorithmic development relies on measuring the performance of any proposed algorithm. For reading comprehension, performance is evaluated using Q&A about the corpus. Its difficult to find a large corpus with a comprehensive set of questions about the content. Deepmind is cleverly converting the Daily Mail article summaries into questions by removing a proper noun. For example: Question: Producer X will not press charges against Jeremy Clarkson, his lawyer says. Answer: Oisin Tymon They are using the Daily Mail corpus to develop their algorithm, and that's smart. They aren't relying on it as an important source of information. Maybe all you guys with the dismissive comments have a better idea? [1] http://arxiv.org/pdf/1506.03340.pdf EDIT: Thanks Otik, reworded the opening sentance
Original comment DeepMind uses the Daily Mail as a huge training corpus for text comprehension
nl
Neutral mention · rule
No doubt there is plenty of hype. From where I sit though, a lot of it is justified (Not the general intelligence stuff of course). Choosing problems for their theatrical effect, rather than utility. Writing articles and research papers as if they're marketing pamphlets. Claiming that incremental improvements are paradigm shifts. Treating arbitrary achievements as if they were commonly agreed upon milestones all along. I'm not sure what to say to this. There are no "commonly agreed upon milestones". The closest things are the academic benchmarks/shared tasks that you seem to be critical of. I guess the closest thing you'll find to a "commonly agreed upon milestone" is…
Read the full comment
something like the Winograd schema[1]? Based on progress like "Teaching Machines to Read and Comprehend" I wouldn't be betting against deep learning on that. If you look for practical applications that give tangible benefits to people outside of academia, the achievement of applied deep learning so far aren't nearly as impressive as you make them out to be. Could you explain what you expecting? Deep learning techniques aren't exactly wide spread yet, and outside Google and a few other companies it takes time for things to migrate into products and have tangible benefits. Nevertheless, Google Search, Pinterest, Facebook image tagging, Android Voice Search etc, etc.. these all are used by billions of people daily. I think it's hard to argue there isn't at least some practical applications. [1] https://en.wikipedia.org/wiki/Winograd_Schema_Challenge [2] http://arxiv.org/abs/1506.03340
Original comment Go, Marvin Minsky, and the Chasm That AI Hasn’t yet Crossed

Keep following the thread.

More in Machine learning

More from the same topics.