Loading ...
Sorry, an error occurred while loading the content.

Re: Machine translation and AI

Expand Messages
  • Daniel
    I like the create and test idea. If a system created say 5 possibles sentence as appropriate for a translation. How would it test each one. There would need to
    Message 1 of 9 , Apr 15, 2011
    • 0 Attachment
      I like the create and test idea.

      If a system created say 5 possibles sentence as appropriate for a translation. How would it test each one. There would need to be a scoring system of some sort. Since language can be infinitely complex and extendable the number of possible ways to say the same thing could be huge.

      E.G
      I like apples
      I am fond of Apples
      Apples are nice
      I love Apples
      Apples are fabulous
      Apples are so sweet
      I am in love with the Apple
      I have an Apple passion
      My passion is Apples

      Etc. Etc. And this is for a three word base sentence. Can you imagine the different ways to say a longer sentence?

      The list goes on and on, how do you tell which one is "best"?

      Dan

      --- In artificialintelligencegroup@yahoogroups.com, "bromer2007" <jimbromer@...> wrote:
      >
      > > So how do we make sentences in the first place? Any ideas anyone? I have a few but wondered what others felt.
      > >
      > > Dan
      >
      > Of course we use memories of events as well as memories of how to use words and sentences. However, since we have to comprehend ongoing events in the terms of past events it seems clear that the methods we use to comprehend what is going on are built from some kind of compounds of generalities. Although using language is different than general conceptualization, I believe that there are many similarities.
      >
      > So this means, for example, that I don't think iconic grounding is necessary (absolutely necessary) for higher intelligence, and so far, there is no evidence suggesting that it is.
      >
      > The problem as I see it is just one of complexity. If a computer program could examine many different possible expressions that might be used to describe a situation it might be able, after a lot of learning, to decide which one is most appropriate. But because of the problem of referential ambiguity the number of possible combinations of meanings currently increase at a rate that is nearly intractable as the amount of knowledge learned increases.
      >
      > If this were not the case, it would be easy to test different strategies.
      >
      > My basic strategy would be for the program to create a possible sentence and then examine it using a variety of analytical methods that are related to the subject (the subjects) of the sentence. Since we are talking about a situation where words, word-phrases and sentences may take on different meanings, this method of testing an expression makes a lot of sense. In other words, the program is not testing every possible interpretation of a sentence, but it does have to examine a great many of them.
      >
      > Computers work well with mathematical problems in which a narrow resultant of a sequence of computations is then used as the input of the next step in a problem that can be eventually solved with a narrow solution. (A narrow solution is a solution with a feasible number of precise correct evaluations.) General AI (or AGI) does not seem to reduce to systems of problems that all have narrowly correct values. If it did, it would be easy to test different strategies.
      >
      > Jim Bromer
      >
      >
      > --- In artificialintelligencegroup@yahoogroups.com, "Daniel" <daniel.burke@> wrote:
      > >
      > > In order to translate from one language to another obviously the person needs to know both languages well. It started me think about how language is stored in your head and how you generate sentences in the first place.
      > > Seems to me the translation process works like this.
      > >
      > > Take source material, read it through and make sure you understand the MEANING of all the sentence.
      > >
      > > Then in the target language make sentence that have the same meaning.
      > >
      > > This is why machine translation is so hard since word that individually mean the same may not mean the same when in bigger language chunks such as phrases or sayings or local terminology.
      > >
      > > How do we make sure the meaning is the same in the source and target language.
      > > And finally how do we ensure the correct word order in the target language?
      > >
      > > So how do we make sentences in the first place? Any ideas anyone? I have a few but wondered what others felt.
      > >
      > > Dan
      > >
      >
    • Pabitra Saha
      So is Sanskrit ... [Non-text portions of this message have been removed]
      Message 2 of 9 , Apr 15, 2011
      • 0 Attachment
        So is Sanskrit



        On 14 Apr 2011, at 20:07, Recai Alkan <recai_alkan@...> wrote:

        > You must first take a base language, it must be a language that doesn't have
        > irregularities. Once i heard that in machine translation between English and
        > Spanish, they had chosen Aymara as a base language.
        >
        > Turkish is also a regular language, almost a mathematical language.
        >
        > ________________________________
        > From: Daniel <daniel.burke@...>
        > To: artificialintelligencegroup@yahoogroups.com
        > Sent: Tue, April 12, 2011 7:40:22 PM
        > Subject: [Artificial Intelligence Group] Machine translation and AI
        >
        >
        > In order to translate from one language to another obviously the person needs to
        > know both languages well. It started me think about how language is stored in
        > your head and how you generate sentences in the first place.
        > Seems to me the translation process works like this.
        >
        > Take source material, read it through and make sure you understand the MEANING
        > of all the sentence.
        >
        > Then in the target language make sentence that have the same meaning.
        >
        > This is why machine translation is so hard since word that individually mean the
        > same may not mean the same when in bigger language chunks such as phrases or
        > sayings or local terminology.
        >
        > How do we make sure the meaning is the same in the source and target language.
        > And finally how do we ensure the correct word order in the target language?
        >
        > So how do we make sentences in the first place? Any ideas anyone? I have a few
        > but wondered what others felt.
        >
        > Dan
        >
        > [Non-text portions of this message have been removed]
        >
        >


        [Non-text portions of this message have been removed]
      • Jim Bromer
        Presumably, the program would have the potential to understand things. If it only used text, then a good AGI program would have to have the potential to
        Message 3 of 9 , Apr 17, 2011
        • 0 Attachment
          Presumably, the program would have the potential to understand things. If
          it only used text, then a good AGI program would have to have the potential
          to acquire a lot of knowledge about a lot of different things using words.
          The problem is that because of the ambiguity of words and the ambiguity of
          the use of words, it would have a lot of trouble interpreting the meaning
          for a sentence. The different ways you can speak of your fondness for
          apples demonstrates a little bit about the extent of this knowledge. My
          theory is that we should use this range of knowledge in determining the
          meaning of words. Since this is the potential, the problem and
          computationally feasible I think it might work. I consider this method to
          be a way to form good judgement.

          I am really thinking of a general AI program, not just a translator. But, I
          think that this method is relevant. For example, all your sentences about
          your fondness of apples reveal something about the subject and those
          sentences could subsequently used for a variety of situations that use some
          of the same words - so long as the program had some good way to start
          learning about sentences.

          By exploring the different ways people talk about things that they like, by
          making distinctions between different kinds of likable things, and by
          integrating this knowledge with other kinds of knowledge, I believe that a
          program could choose good translations for the particular situation that it
          is in.
          Jim Bromer




          On Fri, Apr 15, 2011 at 9:03 AM, Daniel <daniel.burke@...> wrote:

          >
          >
          > I like the create and test idea.
          >
          > If a system created say 5 possibles sentence as appropriate for a
          > translation. How would it test each one. There would need to be a scoring
          > system of some sort. Since language can be infinitely complex and extendable
          > the number of possible ways to say the same thing could be huge.
          >
          > E.G
          > I like apples
          > I am fond of Apples
          > Apples are nice
          > I love Apples
          > Apples are fabulous
          > Apples are so sweet
          > I am in love with the Apple
          > I have an Apple passion
          > My passion is Apples
          >
          > Etc. Etc. And this is for a three word base sentence. Can you imagine the
          > different ways to say a longer sentence?
          >
          > The list goes on and on, how do you tell which one is "best"?
          >
          > Dan
          >
          >
          > --- In artificialintelligencegroup@yahoogroups.com, "bromer2007"
          > <jimbromer@...> wrote:
          > >
          > > > So how do we make sentences in the first place? Any ideas anyone? I
          > have a few but wondered what others felt.
          > > >
          > > > Dan
          > >
          > > Of course we use memories of events as well as memories of how to use
          > words and sentences. However, since we have to comprehend ongoing events in
          > the terms of past events it seems clear that the methods we use to
          > comprehend what is going on are built from some kind of compounds of
          > generalities. Although using language is different than general
          > conceptualization, I believe that there are many similarities.
          > >
          > > So this means, for example, that I don't think iconic grounding is
          > necessary (absolutely necessary) for higher intelligence, and so far, there
          > is no evidence suggesting that it is.
          > >
          > > The problem as I see it is just one of complexity. If a computer program
          > could examine many different possible expressions that might be used to
          > describe a situation it might be able, after a lot of learning, to decide
          > which one is most appropriate. But because of the problem of referential
          > ambiguity the number of possible combinations of meanings currently increase
          > at a rate that is nearly intractable as the amount of knowledge learned
          > increases.
          > >
          > > If this were not the case, it would be easy to test different strategies.
          > >
          > > My basic strategy would be for the program to create a possible sentence
          > and then examine it using a variety of analytical methods that are related
          > to the subject (the subjects) of the sentence. Since we are talking about a
          > situation where words, word-phrases and sentences may take on different
          > meanings, this method of testing an expression makes a lot of sense. In
          > other words, the program is not testing every possible interpretation of a
          > sentence, but it does have to examine a great many of them.
          > >
          > > Computers work well with mathematical problems in which a narrow
          > resultant of a sequence of computations is then used as the input of the
          > next step in a problem that can be eventually solved with a narrow solution.
          > (A narrow solution is a solution with a feasible number of precise correct
          > evaluations.) General AI (or AGI) does not seem to reduce to systems of
          > problems that all have narrowly correct values. If it did, it would be easy
          > to test different strategies.
          > >
          > > Jim Bromer
          > >
          > >
          > > --- In artificialintelligencegroup@yahoogroups.com, "Daniel"
          > <daniel.burke@> wrote:
          > > >
          > > > In order to translate from one language to another obviously the person
          > needs to know both languages well. It started me think about how language is
          > stored in your head and how you generate sentences in the first place.
          > > > Seems to me the translation process works like this.
          > > >
          > > > Take source material, read it through and make sure you understand the
          > MEANING of all the sentence.
          > > >
          > > > Then in the target language make sentence that have the same meaning.
          > > >
          > > > This is why machine translation is so hard since word that individually
          > mean the same may not mean the same when in bigger language chunks such as
          > phrases or sayings or local terminology.
          > > >
          > > > How do we make sure the meaning is the same in the source and target
          > language.
          > > > And finally how do we ensure the correct word order in the target
          > language?
          > > >
          > > > So how do we make sentences in the first place? Any ideas anyone? I
          > have a few but wondered what others felt.
          > > >
          > > > Dan
          > > >
          > >
          >
          >
          >


          [Non-text portions of this message have been removed]
        • mikeus22
          Having an intermediary language between starting language and target language might simplify automatic translation, if the intermediary language was less
          Message 4 of 9 , Apr 19, 2011
          • 0 Attachment
            Having an intermediary language between starting language and target language might simplify automatic translation, if the intermediary language was less complex than both the others.

            Q Is Esperanto a simple language? I know its manufactured.

            Q Does Esperanto have fewer words than other languages?

            If so, the ways of translating certain words or phrases will be limited, and the quicker it should be.

            If we are after a "Universal Translator" this might be the way to go. If, for arguments sake, this way produces enough of a meaning transfer to understand the "gist" of what is being said this might be enough for most instances.

            If we want a greater meaning transfer then processing time/power will need to increase exponentially.

            Someone said that a translator needs to read/understand the whole of a sentance to transfer as much meaning as possible. Q How come Human Translators can talk the translation whilst hearing it at the same time? Is it because they are expert, or, what they translate is limited to what is under discussion?

            Mike
          • Recai Alkan
            That s why we need a base language, this is the language that we build our meaning base, in a mathematical format. You must keep it somewhere as meaning so
            Message 5 of 9 , Apr 20, 2011
            • 0 Attachment
              That's why we need a base language, this is the language that we build our
              meaning base, in a mathematical format. You must keep it somewhere as "meaning"
              so you can translate it to some other language(s)




              ________________________________
              From: Daniel <daniel.burke@...>
              To: artificialintelligencegroup@yahoogroups.com
              Sent: Fri, April 15, 2011 3:56:01 PM
              Subject: Re: [Artificial Intelligence Group] Machine translation and AI


              Why use a base language, seem like extra work to me and not how the brains works
              (as far a I know). I think you need to strip down any input sentence to it's
              basic meaning chunks and then find the equivalent meaning chunks in the target
              language. Re-assemble the sentence based on the rules for the target language.
              I think this is a more logical approach.

              --- In artificialintelligencegroup@yahoogroups.com, Recai Alkan
              <recai_alkan@...> wrote:
              >
              > You must first take a base language, it must be a language that doesn't have
              > irregularities. Once i heard that in machine translation between English and
              > Spanish, they had chosen Aymara as a base language.
              >
              > Turkish is also a regular language, almost a mathematical language.
              >
              >
              >
              >
              > ________________________________
              > From: Daniel <daniel.burke@...>
              > To: artificialintelligencegroup@yahoogroups.com
              > Sent: Tue, April 12, 2011 7:40:22 PM
              > Subject: [Artificial Intelligence Group] Machine translation and AI
              >
              > Â
              > In order to translate from one language to another obviously the person needs
              >to
              >
              > know both languages well. It started me think about how language is stored in
              > your head and how you generate sentences in the first place.
              > Seems to me the translation process works like this.
              >
              > Take source material, read it through and make sure you understand the MEANING

              > of all the sentence.
              >
              > Then in the target language make sentence that have the same meaning.
              >
              > This is why machine translation is so hard since word that individually mean
              >the
              >
              > same may not mean the same when in bigger language chunks such as phrases or
              > sayings or local terminology.
              >
              > How do we make sure the meaning is the same in the source and target language.
              > And finally how do we ensure the correct word order in the target language?
              >
              > So how do we make sentences in the first place? Any ideas anyone? I have a few

              > but wondered what others felt.
              >
              > Dan
              >
              >
              >
              >
              > [Non-text portions of this message have been removed]
              >




              [Non-text portions of this message have been removed]
            Your message has been successfully submitted and would be delivered to recipients shortly.