Feat: Numbers to strings that generate coherent speech

[Wessel-Boers]

I wanted to use this gem to change numbers that represent years to strings that can be used by AI's to generate speech. However, at least for Dutch:

platform(dev):016> I18n.with_locale(:nl) { 1910.to_words }
=> "duizend negenhonderdtien"

Is not how we pronounce years. Instead it would be: "negentienhonderd tien". I wrote a quick Module extension that could be added for just Dutch or also other languages for which this is a similar problem. I'm not sure how to add this as a PR, since I'm not familiar with the conventions and how this would become accessible as an option, but perhaps this is useful for others as well.

class NumberConverter
  def self.to_words(number, locale: I18n.locale)
    new(number, locale).to_words
  end

  def initialize(number, locale)
    @number = number
    @locale = locale
  end

  def to_words
    return @number.to_words unless year_like?

    I18n.with_locale(@locale) do
      century = @number / 100
      remainder = @number % 100

      if remainder.zero?
        "#{century.to_words}honderd"
      else
        "#{century.to_words}honderd #{remainder.to_words}"
      end
    end
  end

  private

  def year_like?
    @number >= 1000 && @number < 3000
  end
end

[dblock]

Is not how we pronounce years.

So is the translation of 1910.to_words incorrect or is it correct, but in a different context is different? Sorry, not a Dutch speaker.

Appreciate a PR, start by adding failing tests.

[jlduran]

I agree that a PR with failing tests is a good first step.

If I understand correctly, you’re looking for something like:

I18n.with_locale(:en) { 1984.to_words, year: true }
=> "nineteen eighty-four"

I18n.with_locale(:en) { 2003.to_words, year: true }
=> "twenty oh three"

I18n.with_locale(:en) { 2026.to_words, year: true }
=> "twenty twenty-six"

[Wessel-Boers]

I apologize, my remarks could have been more clear.

@jlduran 's examples are indeed partly what I'm looking for. I want to translate written numbers (1984) to the way a human speaks these numbers out loud.

Current behaviour:

I18n.with_locale(:en) { 1984.to_words }
=> "one thousand nine hundred eighty-four" ❌

I18n.with_locale(:en) { 1099.to_words }
=> "one thousand ninety-nine" ✅

platform(dev):059> I18n.with_locale(:en) { 1100.to_words }
=> "one thousand one hundred" ❌

Certain translations are already the same as how the numbers would be spoken. Others aren't. It seems like the years 1100 and up till 1999, 2100 till 2999, etc. are candidates for a different translation.

Desired behaviour:

I18n.with_locale(:en) { 1984.to_words, spoken: true }
=> "nineteen eighty-four"

platform(dev):059> I18n.with_locale(:en) { 1100.to_words, spoken: true }
=> "eleven hundred" 

You're examples of 2003 => "twenty oh three" and 2026 => "twenty twenty-six" are indeed even more natural, but not strictly necessary as we regularly say "two thousand three" and "two thousand six". Although these could be included if we want to differentiate between the options, for example spoken vs. spoken_casual.

Hope this is more clear.

[dblock]

@Wessel-Boers So just to clarify, current behavior is (a) correct but not used in spoken language, or (b) current behavior is just incorrect?

If it's sometimes correct (a), we want spoken or spoken_casual options. If not (b) we want to treat it as a bug. Either way write some tests that represent what developers would want and that solves your use-case and we can try to help implement if you can't figure it out.

[Wessel-Boers]

@dblock Current behaviour is correct for certain purposes. For the purpose of translating digits to spoken text, it's only correct sometimes. See the 1099 example, this is how text would be spoken. The other two current behaviour examples are not correct.

Getting "spoken text" is something that is only required in niche/specific situations, therefore I think it should be an option (e.g., spoken option).

I will create some tests.

[Wessel-Boers]

I've created a PR for this issue. Turns out there is already a pronounced option! It's only implemented for english however, see #214.