Google's Top AI Folks in an Interesting Panel

Google’s Top AI Folks in an Interesting Panel

Reading Time: 1 minute

Google’s Top AI Folks in an Interesting Panel

This is a great panel discussion of what Google is up to on the AI front.

Be sure to listen to Feifei Li, talking about why machine vision is the killer app for AI. It’s pretty compelling, and I think she’s absolutely right. That’s at minute 25:00

Also of interest is Fernanda Viégas’s segment on the discovery of a kind of virtual language, or “interlingua”, in the semantic space made possible from mapping sentences from multiple languages into a data visualization space. That’s at minute 31:00.

https://events.google.com/io/schedule/?section=may-19&sid=3c8cf4b6-e2c5-4ef0-b3e4-ba9fec6294fd&track=machinelearning%26ai

15 comments

  1. Part (a) from the figure above shows an overall geometry of these translations. The points in this view are colored by the meaning; a sentence translated from English to Korean with the same meaning as a sentence translated from Japanese to English share the same color. From this view we can see distinct groupings of points, each with their own color. Part (b) zooms in to one of the groups, and part (c) colors by the source language. Within a single group, we see a sentence with the same meaning but from three different languages. This means the network must be encoding something about the semantics of the sentence rather than simply memorizing phrase-to-phrase translations. We interpret this as a sign of existence of an interlingua in the network.

    research.googleblog.com – Zero-Shot Translation with Google’s Multilingual Neural Machine Translation System

  2. “Meaning”… As a simultaneous interpreter, I am often asked, “What does he mean” I have to answer, “I know only what he said.” Note, the question rarely arises if both the source language and the destination language people know one another. The word meaning must be used with caution!

  3. Paolo Pignatelli Wait, am I understanding your note right? If two people, speaking different languages, know each other, they might understand the meaning of the translation better than the translator does?

  4. Definitely yes! They have more of a common internal representation of the subject. Language is a fantastic compression scheme – with small specialized compressions and error correcting going in both directions.

  5. The whole portion of the talk about the interaction of vision and language is interesting – I have been advocating it here and on Google’s space and elsewhere for a long time. Vision is but one of the senses, but certainly one of the most interesting along with hearing. All the senses contribute to an internal “map”. To the degree that language is algorithmic, it encodes that internal map. How it encodes it is the interlingua that happens before any Chomskyan transformations.

    When Ms. Fernanda Viegas refers to, “A model here for English, a model for spanish.. or the semantics and the meaning of words. ** and not necessarily the language it comes from or the language it’s translating to..” that is what I am referring to. The clusters are nearby neighborhoods in the internal map. I use a Voronoi map to visualize “Meaning” neighborhoods. What is the function that maps internal representations to the Voronoi maps?

    The end of Viegas’ presentation (at around min 30) about bring in other disciplines other than engineering is also interesting at a meta-level. One can’t help but notice that the panel is all women. As an interpreter, I too notice that the majority of good interpreters are women. This is not a coincidence. Why is the interesting question that should be investigated by NLP.

    If anyone is interested in what I am saying? I will gladly continue, and would be delighted to contribute.

  6. One more note -Wittgenstein’s first model of language in his Principia is the Picture theory of Language. Later he changed to “word play”. If in any way there are parallels to the two philosophical views of language and the A.I investigations today, will it play out in any way similarly? It may seem to be in reverse, but is it?

  7. Natural Language is human communication, but not all human communication is language. Music, for me is the great example (for me, Beethoven and Haydn in particular). So, in CS, what do we risk by slighting these other ways we have for communication?

  8. Just one more observation on the link between vision and natural language. For a simultaneous interpreter, it is much harder to interpret if we do not see the person speaking. Visual clues are like error correcting codes for us. A.I. should take notice.

  9. So, given the above, one possible experiment would be to have an interpreter (interpreters) be shown a speaker (maybe even follow his eye focus), and see the interpreted output. Then the same speaker, but no video. The difference, if any, is what vision adds. Study the difference… . even more “experimenty” , would be a PET study of the differences.

  10. La vérité [lave-hérité] conformité aux faits d’origine qui le cas échéant compense le vice d’un oui-dire

    [l’avère-idée] conformité aux faits ultimes qui le cas échéant compense le vice d’une spéculation

    langage [lents gages] conventions qui ne sont pas l’objet immédiat d’une mise en oeuvre qui les illustre, et les communique de ce fait

    I am no musician but yet intrigued (just right now) by how the above definition of language may include music or fail to, and why.

  11. I see human language as an entropy machine. The language machine’s entropy

    space is a semantic network of some kind… . Successful communication

    minimizes the difference between the networks. …

    Music (I am referring to classical music, because that is all I know about)

    elicits some part of the cognitive network (at various levels, many

    interesting experiments could be done), that is what composers do. So,

    notes are written, then are read by performers, where they activate certain

    nodes of the network (feeling!?). The sound is generated, and an analogous

    (to performer’s) process is repeated.

    This is the way I see music as Language.

    I would be glad to amplify if anyone is interested.

    Paolo

  12. Testimony. I was a cleaner in a company and my wife left me for another man and things was very tight with me financially until i met the spiritual healing spell. who turn my life around for good. he make me a manager of the company where i was working as a cleaner and bring back my ex wife within 24 hours and now I’m wealthy living with my lovely wife. contact him through. drpaulspellcaster@yahoo.com cell line +2347013291233 reach him if you have any problem and may God bless you

  13. What comes to mind as the prototype for “music” is the jam session.

    Conventions illustrated and thus conveyed in uses whose purpose they are not

    — is the rough translation of the definition I pulled by the effort to justify the wild etymology “lents gages” (slow gages) for the French “langage” (language).

    I understand to be at the focus of jam sessions the negotiation of conventions introduced by original use, so whether it’s language according to the above def… is disputable.

  14. Why i cant open thant appss 😰

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sign up here for the latest articles. You can opt out at any time.


Subscribe by email:



Or subscribe by RSS:

Subscribe-by-RSS---Orange-Background
%d bloggers like this: