Author Topic: I miss the good ol days  (Read 2781 times)

Offline Nick R

  • New Linguist
  • *
  • Posts: 2
I miss the good ol days
« on: July 16, 2021, 03:43:02 PM »
Hey everyone. I'd first like to give a shout out specifically to Malfet. Thank you again Malfet for helping me out as much as you have. I don't think that I would be where I am at today, doing what I am doing today, if it weren't for our original discussions on the previous forum that unfortunately is no longer there.

Anyways, I am posting in Computational Linguistics here for a reason. I recently had the opportunity to apply and interview with Google but they didn't think that I gave them a good enough example of how I would go about making the Google Assistant better. The reason that they don't think that I can do that is because they are doing it wrong. Plain and simple. I've been studying NLP (Natural Language Processing) and the methods used to help computers understand language and although they are working I still can't help but think that they aren't doing it right. When it comes to data science and NLP it's just not the right methods to use and the reason for that is that when we take language data and format it so that it is understandable to a computer, it exponentially makes the size of the data larger. There's a much simpler method and I think that Chomsky is on the right path about this. We enter phases based on context. This method is the best that exists because it doesn't need to change language data to numbers so that our ML (Machine Learning) algorithms can calculate it and find the patterns. The much simpler method is to just analyze the patterns in the strings (pattern recognition). So we have phases based on what we know and we enter new phases and bring previous patterns into the phase should they not exist in that phase already. As for integrating this with other phenomena (such as performing an action), the computer needs to be able to not only look at language (which is actually pretty easy) but also acquire the actions associated with the language being used (within the context of course). The problem that we are running into right now is that we don't have, or even want, to spend the time teaching our computers how to do these things. There's things that we need to do to set up the ability to learn from everything from the very first instant. The program needs to be right and it needs to observe and act. We're not there yet.

Offline panini

  • Linguist
  • ***
  • Posts: 215
Re: I miss the good ol days
« Reply #1 on: July 18, 2021, 08:13:42 AM »
I have the impression that GA has gotten worse in the past few years. I use their speech-to-text interfaces a lot, and within the past 5 years it has gotten recognizably worse at dealing with ordinary speech (not even a challenging dialect), to the point of emitting ungrammatical strings. But I wonder whether this is the result of their input: kids these days are redefining the database of "English", and maybe this is an inevitable consequence of statistical approaches.