Skip to end of metadata
Go to start of metadata

Please join us this Thursday, 17 November, to hear about Linguistics Prof. George Lakoff's The Berkeley Center for Brain, Mind, and Society.

Thursday, Nov 17, 2016 from 12 - 1pm
Where: 200C Warren Hall, 2195 Hearst St (see building access instructions on parent page).
What: George Lakoff: The Berkeley Center for Brain, Mind, and Society
Presenting: George Lakoff, Ellen Dodge, Luca Gilardi
Facilitating: John Lowe (Linguistics and Research IT)

To prepare for the meeting and discussion, please read/review the following overview of the MetaNet system beforehand:

For those who wish to do some more in-depth reading, the following articles forthcoming in the Constructions and Frames Journal may be of interest:



Presenting: George Lakoff, Ellen Dodge, Luca Gilardi (ICSI, Center for the Neural Mind and Society, Linguistics)
Facilitating: John Lowe, Research IT


Aaron Culich, Research IT
Aron Roberts, Research IT
Chris Hoffman, Research IT
Maurice Manning, Research IT
Richard Katz, SAIT
Ron Sprouse, Linguistics
Scott Peterson, Doe Library
Steve Masover, Research IT

[see slide deck, PDF]

MetaNet project: origin in govt interest in how understanding of metaphors (or not) helps or hinders policy success in a multicultural world.

Metaphors provide a way to reason about one domain of experience in terms of some other domain. They affect inferences we make about a given domain. E.g., poverty as disease that requires intervention/cure vs. location where people get stuck, needing help getting unstuck.

20% of word meanings explained by metaphor; MetaNet study found 1/3 of discussion of poverty is metaphoric.

MetaNet system: database of annotated examples of metaphoric language.

Working with Aaron and Nick Adams (BIDS), we're getting access to serious computing infrastructure. Research IT and BIDS, with mandate to serve whole campus, easier to do this for hard scientists, less so for social scientists -- but Research IT and BIDS are helping social science and humanities researchers.

Big motivation to make MetaNet accessible to other researchers; but, equally important and as a precursor, system improvement and "more linguistics" (analyze more metaphors) is necessary.

MetaNet is currently implemented in Python. Contemplating a rewrite to a big data pipeline (Spark or Storm). "Cloudification" ... Docker for personal use ... MetaNet as a service.

"People think that data out there is meaningful -- that isn't true" - GL



Maurice wonders whether NLP software built by a student or postdoc (reasoners, etc.), and oriented toward a single publication, have evolved to a more robust, enterprise quality ... and is that a required step for serious scaling.

GL points out that Srini Narayanan - built a big part of MetaNet, also Semantic Web.

ED points out that using frames and analyzing relation of frames or groups of frames allows varying levels of granularity in analyzing a corpora.

GL: differing world views (which exist) cause words and frames to carry different meanings, meaning is dependent on world view -- not a static, authoritative set of meanings. Paper just submitted to The Atlantic explores how this was manipulated in the recent U.S. election.

GL: People tend to have multiple world views, applied to different issues. Main WV apply to lots of issues (for the person who holds it), lesser apply to fewer issues, where multiple world views apply they cancel each other out. This was a lesson learned 20 years ago, "The Leadership Institute" trained 160K people, spread them out, and got them to talk in ways that strengthened conservative world views. Democrats have copped onto this >~12 years ago, but have not organized as effectively to propagate a liberal world view, in part because of a reluctance to "be like the Republicans."

GL: Studying Bretbart and Daily Kos using MetaNet will, when we get to it, provide some very interesting insights.

ED: And looking for other ways of speaking about issues may prove quite fruitful. Exploring possible use of different metaphors.

Richard Katz mentioned Tensor Flow; GL is interested in knowing more about it; notes that Neural Net did not work for language. GL requested some materials to read; Aaron will send a collection of links.



  • No labels