Hacker News Reader

Hacker News

Comment by trogdoro | Hacker News Reader

Parent Full thread

trogdoro

> letter frequency is not taken into account.

False.

> no capital letter/lowercase distinction

False.

> No circumflexes, accents, gravures, hats, dots, umlauts...

True, but not impossible to overcome though if you'd use your imagination.

View on HN

> letter frequency is not taken into account. False.

Considering the progression of pattern through the A-Z alphabet, it's extraordinarily unlikely this has anything to do with letter frequency. It's instead quite clear that if the alphabet happened to be recited in a different order, these dots would have the same progression but assigned to different letters.

Have a look at the research behind the "fitaly" stylus keyboard for what goes into "taking letter frequency into account".

With a stylus, the predominant goal is likely to keep movement to a minimum. It may seem to you that a similar dynamic would apply to letters, but in practice it's not the only driving factor. Other factors like the ambiguity caused by sparseness must also be weighed.

> quite clear that if the alphabet happened to be recited in a different order, these dots would have the same progression but assigned to different letters

That is very far from clear. If you'd think for a moment, you'd probably be able to come up with about 10 patterns that could have been used, that appear about as deliberate. As mentioned below, mappings with and without patterns (many more without) were considered.

If you think you've identified a better pattern, post it here. I'd be interested in seeing it.

I have an idea for an improved pattern based on providing visual cues and accounting for letter frequency:

http://esploded.s3.amazonaws.com/anon_data/2012/eyS/-dotsie2...

Vowels touch the top and bottom of the line, visually outlining the word and providing a sense of center (vowels also use the most dots, where consonants are sparse)

More common consonants use smaller (sparse) dot patterns getting denser as you get into less common (frequent) letters

More common consonants frame the center dot (to contrast the vowels which touch the edges/outside dots) using the less centered patterns as frequency decreases

All the glyphs fit into the 1x5 grid (the letter "Z" was 2 columns wide in the original)

Also worth noting: where possible I tried to make the glyphs memorable ("i" is the best example, followed by "o" and "z")

In retrospect, it may be a good idea to swap the glyphs of my current "B" for the "Y" since y is a semi-vowel it would touch both edges, where "B" has no reason to.

EDIT: here is a second attempt where ONLY vowels touch both edges, and Y is swapped (and touches both edges as a semi-vowel)

http://esploded.s3.amazonaws.com/anon_data/2012/e5YO-dotsie3...

Thoughts?

> Thoughts?

Not bad!

> More common consonants use smaller (sparse) dot patterns getting denser as you get into less common (frequent) letters

A few mappings emphasizing fewer dots for frequent letters (making for "sparser" words) were tried out and moved away from. The more speckled-looking words tend less to form concrete shapes and create "blocks" that stand out to the eye. If words are too dense you get similar issues, only in the negative. The optimum seems to be on the sparse side, but not overly sparse.

> Also worth noting: where possible I tried to make the glyphs memorable ("i" is the best example, followed by "o" and "z")

The original dotsies O and I are relatively memorable.

> More common consonants frame the center dot (to contrast the vowels which touch the edges/outside dots)

I could see how that could be interesting. How that would balance against the trade-off of ambiguity between similar letters (like s and t) would probably be hard to tell before one tried it.

I wonder if it would be worth coding up a page that lets people make their own mapping and preview them on some text, and posting it to HN.

With a few minutes of thought and with reasons behind your choices, you've produced a more legible system.

Regardless of OP's protestations, the diagonal one dots, two dots, etc., have nothing to do with legibility and everything to do with alphabetic order mirrored in dot order.

Your system has both reason and rhyme.

You're continuing to presume there's only one pattern, and it's the one that was picked. To drive home the point that this isn't the case, I'll list some out:

00001 00010 00100 01000 10000 ... 10000 01000 11000 00100 10100 ... 00001 00010 00011 00100 00101 ... 10000 11000 11100 11110 11111 ... 01111 10111 11011 11101 11110 ... 10000 11000 01000 01100 00100 ... etc.

Now, why is it implausible that one or two of the many possible mappings with a pattern has tradeoffs about as good as the best of the many possible mappings without a pattern? Especially when the former have the pretty significant advantage out of the gate that the pattern makes them easier to remember.

Note that a valid critique of the many possibly mappings without patterns compared to the many possible mappings with patterns is that they are harder to learn.

If you put a lot of weight on the dots visually correlating to the numbers, see A, C, F, G, H, I, K, L, O, P, R, S, V, and X - all have a decent correlation.

> Note that a valid critique of the many possibly mappings without patterns compared to the many possible mappings with patterns is that they are harder to learn.

BS. You don't read in alphabetical order. Just because you can reproduce the dot pattern doesn't mean you've "learned" it for its purpose. That the letter e comes between d and f is not relevant to trying to read.

Let's say you're looking at a letter that's the lowest dot and you've forgotten what it is, but you remember that d is the 2nd lowest and f is the next in the pattern (the 2 highest). For you to be right, it would have to not occur to the average person that they can use their Sherlock Holmesian skills of deduction to figure out they're looking at an e.

Or, of course, they could just go through the pattern in their mind until they get to that letter, then they'd have it.

I certainly don't presume there's only one pattern. I assume there are 2.63 × 10^35 patterns.

there are 2^5 patterns... aren't there?

Patterns for the sequence of letters in the alphabet.

2^5 for a, times 2^5 - 1 for b, etc.

I wrote a little code to analyze the letter density etc:

https://gist.github.com/1855582

tl;dr:

DOTSIES

    Average density:     2.02
    Vowel density:       1.52
    Consonant density:   2.33

DOTSIES 3

    Average density:     2.78
    Vowel density:       4.12
    Consonant density:   1.96

What about swapping "a" and "e"? The current "a" symbol somewhat resembles "E", and more English words start with "a" than "e", so the full line might be more useful as a hard break at the start of words than in the middle of them. (Sources: A http://www.wolframalpha.com/input/?i=letter+a, E http://www.wolframalpha.com/input/?i=letter+e)

> Other factors like the ambiguity caused by sparseness must also be weighed.

The current pattern doesn't suggest weighing of anything. You marched the dots down diagonally, first sets of one, then two, and so on. You mirrored alphabetical order with numerical dot order. You're now trying to justify that after the fact. It's not credible.

It's too bad you're choosing to argue instead of seriously considering the surprisingly thoughtful feedback several others have given you here.

For example, my point about FITALY was obviously not about the stylus movement. The similar dynamic isn't movement distance, the similar dynamic would be a statistical use of the legibility of adjacent dot columns. My point was the research that went into Fitaly:

"These figures are obtained using a corpus of digraph probabilities similar to that described by Soukoreff and MacKenzie (1995)... We have measured the frequency of letter-to-letter transitions for a representative corpus of the English language with several millions of characters. For example, this produces the number of times the letter o is followed by the letter a."

Given a dot matrix, there are a limited number of dot patterns. It's straightforward to measure the legibility of those dot patterns to humans of normal visual acuity (using mechanical turk for example), then to mathematically derive a letter assignment that maximizes legibility across, say, the Brown Corpus.

That would be something interesting to see.

My imagination doesn't seem to be good enough. How should I write "Doña Ana, New Mexico is one of the few US place names with the letter 'ñ' in its official name."?

I can't just drop the tilde in both cases because then it doesn't make sense. The best I can come up with is (transliterated) "Dona Ana New Mexico where the first n has a tilde over it is one of the few US place names with a tilde in its official name."

That's cumbersome.

Without punctuation, how do you express "eats shoots and leaves" as being different from "eats, shoots and leaves"; that being the punchline to the joke ending "Panda. Large black-and-white bear-like mammal, native to China. Eats, shoots and leaves." (See http://en.wikipedia.org/wiki/Eats,_Shoots_%26_Leaves )

Where does "without punctuation" come in to it? They say "Numbers and punctuation aren't altered." Sure enough, the bookmarklet doesn't remove punctuation.

Also, the bookmarklet seems to skip over paragraphs that have a character they can't represent. I tried it on your post and the paragraph with the 'ñ's wasn't altered.

Correct, punctuation is left alone since altering it would gain very little in the way of space efficiency.

> the bookmarklet seems to skip over paragraphs that have a character they can't represent

It's not that sophisticated. That's just due to the site's css specifying a font for some paragraphs and not others. Not sure the best way to get it to over-ride everything. I suppose it could be updated to crawl through the dom and append style attributes to all tags.

Re accents etc., currently they just show as normal letters (including the accents), which really isn't that bad. Ways to improve the situation could include squeezing some of the accents in by making their slant more vertical. Re ñ and ü, they could maybe be rotated 90 degrees, or maybe put beside the letters rather than on top of them. If there's a demand for this, it wouldn't take me long to throw a couple straw-man versions of the font out there.

Indeed, I missed that. The bookmarklet doesn't work for Safari and there's no easy way to test things out. ... A-ha! I can edit the DOM directly and see the updates.

It's implemented as a special font. That means I can put an ñ in the text and see the combination of dotsies and 'normal' text.

I needed to increase the font so I could distinguish the ":" and make out the doties. Even with the font enlargement, it was smaller than the original text. OTOH, I could reduce the normal text by a few font sizes, so the right comparison would require getting practice with both styles and figuring out what font size feels natural, and then make the comparison.

It doesn't seem worthwhile to do that.

Care to elaborate? I was thinking the same thing, all the vowels have a logical symbol in the pattern, as opposed to something unique or more striking.

If you look at dotsies.org you'll see capital letters have smaller dots above them to distinguish them from lower case.

Using a pattern was done to make it easier to learn, but many options (with patters and without) were tried out and rejected. Notice that the vowels all have dots at the top row or bottom, which is desirable as it tends to give words hollow shapes with negative space in them.

Optimizing for space efficiency has many dimensions, too dense is not good, and too sparse is not good. There were a couple versions where the frequent letters had fewer dots and were concentrated toward the bottom, but that increased possible ambiguity between words one might mistake for being shifted up or down one dot.

The problem with using the dots as a denotation for vowels is that that kills of any way to deal with diacritics, let alone any of the latin-1 supplement characters necessary for languages other than english.

That all not considering the lack of baseline etc. Also, why is "z" the only letter with a 2-dot-width?

My computer screen gets dirty. At first I thought the small dot over the capital was a speck on my display. I tried to rub it off.

I came up with this line last year: "I took a polish to the Polish readings in Reading by the nice guy from Nice." I still haven't found a word which describes two words which are spelled the same except for capitalization and which sound different.

Reply on news.ycombinator.com