4
$\begingroup$
  1. Are there lexical/semantic knowledge bases for physics that can be used for automated reasoning and AI (like Princeton's Wordnet and MIT's Conceptnet for common sense English usage)?

  2. If not in existence, are there physics-specific issues to keep in mind while developing such a knowledge base? For example, can it be developed by using existing semantic relationship clauses in Conceptnet? Pointers to review papers, books etc will be good too.

I understand that such a knowledge base will not be complete without the mathematics related to the lexicon. But, when browsing through Q&As here, a lot of them do not seem to involve any mathematics at all (just pure English words). In any case, it should not be impossible to add the mathematics once there is a non-mathematical lexicon/semantic network (and there seem to be a few initiatives to include math in semantic web, like OpenMath, Content-MathML, OMDoc etc)

PS: This is not some crackpot rambling. I am a phd student in high energy theoretical physics. I arrived at this question while thinking about how much of current physics can be automated through AI.

edit 1: Something related: http://www.cs.utexas.edu/users/novak/physics.html

edit 2: Created a chat thread (update 2a: Apparently, the chat thread has been closed)

edit 3: Included few resources for included mathematics in semantic graph data

edit 4: Ed Shaya's astro-physics ontology, which also includes quite a bit of other areas of physics: http://www.astro.umd.edu/~eshaya/astro-onto/

$\endgroup$

Before answering, please see our policy on resource recommendation questions. Please write substantial answers that detail the style, content, and prerequisites of the book, paper or other resource. Explain the nature of the resource so that readers can decide which one is best suited for them rather than relying on the opinions of others. Answers containing only a reference to a book or paper will be removed!

  • 2
    $\begingroup$ This seems to be a question about philosophy of science or about artificial intelligence, rather than about physics. $\endgroup$ – rob Aug 15 '14 at 1:19
  • 1
    $\begingroup$ Science of knowledge representation is certainly not philosophical! My question could belong to AI as much as it could belong to physics. By the way, AI.SE is closed. $\endgroup$ – GuSuku Aug 15 '14 at 1:22
  • $\begingroup$ I'm afraid I don't understand exactly what you are seeking: Are you searching for something like the basis for an automated proof checker as they exist for certain maths, but for physics? Or do you want a compilation of the terms physicists use, where they differ from common English usage? Or a formal account (in EBNF or whatever) of the "physical language"? $\endgroup$ – ACuriousMind Aug 15 '14 at 2:28
  • 1
    $\begingroup$ I doubt such a thing exits or can even exist (in the sense of being successful). Physics is very much an interplay of human ideas coupled to experimental data. The ideas themselves are completely meaningless without a deep understanding of how they relate to the data. So unless a knowledge graph of physics knows how to build an accelerator or what a good design for a Stern-Gerlach experiment is... $\endgroup$ – CuriousOne Aug 17 '14 at 0:57
  • 1
    $\begingroup$ I don't have time to write this up as a proper answer, so a comment. Benjamin Kuipers was long been active in the field of qualitative reasoning. His works span from the mid 1980s to the early 2000s. His seminal paper Kuipers, B. (1984). Commonsense reasoning about causality: deriving behavior from structure. Artificial Intelligence, 24(1), 169-203 is a good place to start. A number of other researchers have since taken up the cause. $\endgroup$ – David Hammen Sep 9 '14 at 16:18
3
$\begingroup$

I am the author of the astronomy and physics ontology mentioned in the original question. The original purpose of that ontology was to improve search for data and articles in astronomy. The idea was to have data sets tables and individual columns in tables marked up with relevant keywords.

The rows of data in astronomy are usually different astronomical objects of different types. A search could then be done for a range of values on a property for some type of object and the return would be all relevant data in all astronomical archives.

But as we progressed, we thought of many more ways in which such an ontology could be used. A newbie could quickly learn on his/her own from the ontology all the different astronomical species and subspecies, their properties and their brightest or closest examples. One could ask for the latest papers dealing specifically on a particular type of observation of a particular type of object within a range of distances or direction on the sky.

I think most of this can carry over into physics as well. One could ask for specific experiments or papers on a topic and then depending on the results ask for results on either broader or narrower terms. Newbies can learn which terms mean nearly the same thing and how they differ, if they do differ.

One thing we looked at is whether a complex long paper can be boiled down to a few simple ontological statements. It helps that a reasoning machine can tell you which statements are repeats of already known things and which are new. Then, with training one could read the results of an entire Physics Review journal in a few minutes. The list goes on and on.

However, the funding required to do this is large and right now the only groups that I see doing these kinds of things are Microsoft, Google and Apple, and all of that is behind closed doors.

$\endgroup$
  • $\begingroup$ Thank you for answering my request and sharing your thoughts here! Could you also perhaps share your thoughts on (1) Crowd-sourcing such an initiative among the physics community (like Freebase or OMCS/ConceptNet did for general knowledge ontology) (2) Incorporating semantics of mathematical knowledge in the ontology (cf my Sep 9 comment to Ben, starting "Also, take a look at Content-MathML and OpenMath options..."). $\endgroup$ – GuSuku Sep 15 '14 at 19:30
  • $\begingroup$ Also, do you happen to know other projects that have built (or building) ontology in physics? As for funding, we could explore putting together a credible and subjectwise well-represented consortium of members from the physics community, and seek crowd-funding from public (through KickStarter, Experiment/Microryza etc). The funding might be enough to cover hardware costs and, if media picks it up, it can catch the attention of potential volunteers (students, postdocs etc) from the community. $\endgroup$ – GuSuku Sep 15 '14 at 19:31
  • $\begingroup$ I don't think crowd-sourcing would work well because only physicists have the subject knowledge required and there are just are not enough of them. However, I think automated readers could be used. There are programs out there that read newspapers and gather intelligence info for the military. A specialized one for physics could perhaps be taught to read physics articles and quickly create an immense database of physics knowledge. Perhaps crowd-funding would work for this if we aligned with say a small branch of physics (like cosmology) and the appropriate semantics engineers. $\endgroup$ – eshaya Sep 16 '14 at 18:53
  • $\begingroup$ Auto-mining is a great idea. Coming to think of it, even Yago and v5 of ConceptNet use automated Ontology learners like ReVerb to populate their general knowledge base. We can perhaps do a small proof-of-concept demo (for a narrow field, like you suggested) and use it to gather a bigger team and seek initial funding! $\endgroup$ – GuSuku Sep 17 '14 at 17:09
  • 1
    $\begingroup$ nltk discourse processing looks interesting as well. $\endgroup$ – eshaya Sep 18 '14 at 18:56
3
$\begingroup$

I'm the developer of a project called the Physics Derivation Graph, see it in GitHub, too.

My intention is to develop a set of derivations into a graph which would capture the current state of knowledge in Physics. Although I consider automated reasoning outside the scope of my project, you are welcome to look at the databases and think about what you can use.

I am intentionally avoiding dependence on English to construct the graph. The graph should be able to be analyzed by a computer algebra system. This implies it could be accessible to your interests in automated reasoning if you are approaching this mathematically.

PS: I too think I'm not a crackpot since I have a PhD in computational Physics


Edit 20150708: Link to the site and to the source code in GitHub.

$\endgroup$
  • $\begingroup$ Since your original chat room seems to be gone, please let me know if you're interested in making a new room. I would like to discuss. $\endgroup$ – Ben Sep 1 '14 at 12:34
  • $\begingroup$ That is an interesting project! It would certainly be useful in automated reasoning. Do you have a discussion forum that is native to your project site? That way, you can keep together conversations between collaborators. $\endgroup$ – GuSuku Sep 8 '14 at 23:48
  • $\begingroup$ Also, take a look at Content-MathML and OpenMath options. LaTeX is well developed for presentation, but it is not as well developed for authoring semantic contents (sTeX is one of those rare extensions, but it is not under active development). NIST wrote a LaTeX-to-MathML tool to allow semantic web processing (for eg. mathwebsearch) of its digital library of mathematical functions. $\endgroup$ – GuSuku Sep 9 '14 at 17:27
  • $\begingroup$ I did consider both Content MathML and OpenMath. I agree with you that Latex is also not intended for semantic content, but it does render well. I am aware of the Latex-to-MathML projects; my current focus is on rendering a visual graph. $\endgroup$ – Ben Jul 10 '15 at 1:49
  • $\begingroup$ I wrote my observations in this report: github.com/allofphysicsgraph/proofofconcept/blob/gh-pages/doc/… $\endgroup$ – Ben Jul 10 '15 at 1:50
1
$\begingroup$

You might find this paper interesting:

Predicting Research Trends with Semantic and Neural Networks with an application in Quantum Physics. Mario Krenn and Anton Zeilinger. arXiv:1906.06843 (2019).

In their own words,

Here we demonstrate a method to build a semantic network from published scientific literature, which we call SemNet

where they use Wikipedia as a source of concepts forming nodes in the knowledge network, and the published literature as a source for edges linking those nodes together with the strength of those edges.

I don't know how useful this will end up being, but it's worth a look if you're interested in that genre.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.