I can read the text in the snippet without issue, but I clicked through to some of the full images in the web viewer and couldn't read the text at all. I tried downloading a few PDFs and still couldn't read them.
Not sure if I'm just not getting the full quality somehow or if the image quality just isn't there for OCR to ever work?
I can read them as long as I zoom in the exact right amount. Too far and it becomes a blocky mess, too little and its too small, but theres a goldilocks point where it's readable.
It also seems likely that the jpeg2000 versions are the original, so they may be slightly better, but haven't been able to open one of those yet.
I would start by either improving the OCR, or piggyback on GCV/Amzn, who have better tools. For instance, what if I drag-and-drop their sample image to this demo page https://cloud.google.com/vision ?
All text is recognized, and as far as I can tell, there are no errors.
There are a lot of difficulties with older text, but I would start with low-hanging fruit such as trying to use better tools. On top of that, you have the problem of making sense of the layout, fixing common typos, etc.
I tried the GCV demo as well and got one mistake "prevent:" instead of "prevents" as it is in the text. But that's the only I could find, which puts it several categories ahead of the txt file.
I'm not sure though whether the IA can get Google to OCR it for them for a budget they can afford. Likely they'd want OCR solutions that have a one-time cost, so volume based SAAS offerings won't work.
Commercially available OCR is amazingly bad. The errors it makes are crazy. For example, if the top of a 'd' is missing it might read it as an 'a', while a human can see that it's identical to all the other 'd's on the page except that a bit is missing, while it looks nothing at all like an 'a' in any of the fonts used in that book. Or the OCR sees a random blotch and guesses it might be a comma, although it's in completely the wrong place and has the wrong size to be part of the text.
I think perhaps the developers took a wrong turn when they started trying to improve OCR with language models rather than font models. Humans can accurately transcribe a printed text without knowing the language. They do that with a mental model of font metrics and so on. In fact, a human transcribes more accurately (though more slowly) when they don't know the language because they don't erroneously "autocorrect".
I wonder if Google has in-house OCR that works much better than commercially available OCR. Google has OCR-ed at least 25 million books. You can't download the complete texts, but you can see snippets. Perhaps someone would like to publish a paper assessing the quality of Google's OCR and comparing it with commercial software. (Probably someone has done that already; I'm just bad at finding papers.)
Have you evaluated the latest version of Tesseract (developed by Google)? They added an LSTM-based OCR engine starting with version 4 and the improvements over version 3.x are startling.
Also, ocrmypdf [0] is a great Tesseract-based tool which makes it easy to add ocr layers to raster pdfs. Plus, it takes care of optimizing the resulting file (compression, etc). I used it on several old academic papers and have been pleased with the results.
FWIW, Tesseract does pretty well on the second image (though I don't know how well it would do on the uncropped version):
THE NORTH STAR.
ee ee
ROCHESTER, DECEMBER 3, 1817.
— — ——--
‘THE COLORED CONVENTION,
We give Mr. Nell’s report of the doings of
this Convention, as the best we have scen. |
The crowded state of our columns prevent:
our publishing in the present number, any of
the able and interesting reports which en-.
gaged the attention of that body. We shall
attead to them in our next.
For the confidence reposed in me,
Your remark about font models makes me wonder whether it might be feasible to extract recurring shapes in a scan to identify the glyphs used and reverse-engineer a font from that.
Some years ago I ran across someone's research based on some Google book scanning technology reporting on the fantastic number of instances on f()ck in books prior to the 20th century. This is simply not true. The problem was the scanning technology which interpreted the old elongated S type face in old books as F.
For about a year now i am working on an iOS Application that requires good OCR. With the release of iOS 13 Apple has added an API for this in their Vision Framework and i have to say i am shocked how good it works. It works completely on device and, according to Apple, uses Machine Learning.
You can feed it an image that contains text, with the only constrained being that the text should not be rotated. Depending how much text there is it can digitize a 12MP image in about .5 to 2 seconds on an iPhone Xs.
I compared it to ABBYYs OCR SDK and i have to say Apple's OCR outperforms it nearly every time both in speed as well as in recognition quality. I am still fascinated by it and would really like to know how this is working under the hood.
Yes! I hope that one day the vast majority of human written text will be first available online and second OCRed for automatic treatement (search, machine translation, etc).
Don't let the perfect be the enemy of the good. XIX english is pretty close to XXI english, XIV not so much[1]. I make frequent use of https://gallica.bnf.fr/ but the older the document the more unfamiliar the dialect. In german the older scripts (let alone XX forms such as Sütterlin!) are likely to be even less OCR/search friendly.
Just like middle age Portuguese isn't like modern Portuguese, and we also have stuff like Linear B and whatever else exists since there are written documents.
Related to the overall culture tech, most of the scientific revolution occurred in NeoLatin which isn't taught, mostly remains untranslated and untranscribed. For instance: Descartes first book, which dealt with music theory and human emotion.
Sounds like a great project. There are many remarkable documents from that century ... explorers in many realms, science pioneers making great strides.
Re their statement: "What we do not have is a good way to integrate work on these projects with the Internet Archive’s processing flow. So we need help and ideas there as well."
... maybe there are still people from the Gutenberg project around, they used to be handling human-transcribed stuff at quite a volume. (Personally I'd rather type stuff in than 'repairing' bad OCR, that constant back-and-forth is just aggravating. I'd be glad to dig in on something personally interesting, like say a little-remembered expedition or research in an area of interest ... they could arrange stuff by subject that way and then put up wish-lists for volunteers. History is full of forgotten amazings.)
I’ve used it quite a bit and they have some of the best OCR results I’ve ever encountered when it comes to scanned newspapers. But it’s just the tip if the iceberg. Historical newspapers seems to be one of the largest corpuses of material yet to be digitized.
Thank you so much for the chroniclingamerica link. That is seriously cool. I heard my personal "Dangerous Rabbit Hole" alert when I clicked on it, but what the heck, it's Saturday. I'm diving in and definitely won't be out before dinner.
Why not just do a SETI@HOME style thing where you get a slice of a newspaper article, a human being types it up, each is compared to one or two other people typing the same thing up, when they all match that piece is done and you move on.
I guess that would take a tremendous amount of time, but it would be really cool to get an old newspaper article to look at, and I bet a lot of people would find interesting things to talk about.
You could have a "comment thread" for people who added the piece, kind of like adding a digital layer to history.
I love this idea. What I hate about captchas is not the challenge, but the purpose. However, part of me liked the having to find out what was written on the image. So I would gladly do that on my spare time for a communal project like this.
Besides, as you said, reading the old paper would be enough of a prize in itself.
Captchas original purpose was transcribing books, no. And then Google used their version for doing house numbers. (I've a feeling there was some other common type I saw, but can't recall). Now, they're used to train vehicle AIs and map road features in USA.
Less happy about the latest one but the book transcription and map correction were good as I benefited from both those projects (used the tools).
We recently analysed a dataset from an 1859 geomagnetic observatory in Rome [1].
We found the document (the observatory yearbook written in 1859) via Google Books. As I don't speak Italian, I initially typed out passages into Google Translate in order to find the information I needed. Google Books has a view text option, but the format of the page and font often made it garbled (when pushed through translate in any case). Decent OCR likely would have made my life a lot easier.
There is a recent trend in space weather research to study extreme geomagnetic storms that happened in the 19th and early 20th century, and is aided partly by all of the scanned documents from that era available on the likes of Google Books, The Internet Archive and HaithiTrust. Better OCR would be a great help.
Although even having access to all of the documents is already incredible!
Can someone help me understand the greater vision of this project?
I do like the idea and efforts from a technical point of view. Tinkering with OCR on unusual (or old) languages.
But that's not the goal of this project as far as I'm concerned (it's a byproduct?) Archiving every single news entry for the sake of completion sounds more like obsession than purpose. We're creating so much information that it will be even harder to separate garbage from valuable information (you have to spend time reading the useless stuff before you can justify whether or not it's valuable to you).
Information overload IS a problem and by adding more information to an already saturated ecosystem I don't see the vision here but would like to understand :)
It almost seems like a hording problem but for the digital natives. People accumulate a lot of stuff but rarely can they actually appreciate what they possess as time & perception is a very limiting factor.
An article that would shed some light would be highly appreciated.
They're not generating new information though, they're aiming to make old information available. I'm sure that it's quite helpful for historians to be able to search & read old newspapers, it gives more details about what people read about and often establishes a more specific time line.
We're probably not that good at recognizing which bits will be of interest to future generations, so archiving everything (to a point ... but I believe that newspapers are well within reasonable) sounds like a good idea. Plus you never know what you discover when you make things available.
Not to mention that from 1800 to 1930 the world population doubled (1 billion to 2 billion, passing the 1.5 mark just around 1900).
While this pales in comparison with the 6 billion humans added since, that's a significant change - particularly for most likely available recorded sources - at a time of monstrous evolutions to major world powers/empires, expansion into vast new areas of the world (namely North America) of essentially the British empire (while at the same time the East India Company ceased to exist by the end of the century for contrast), some abolition of slavery becoming a reality in places (1833 for the British), and what arguably kickstarted much of the mental frameworks for our entire lives: the first two industrial revolutions (for example: democratization of once-monastic school system while adopting the year-of-production type of mental model for its promotions).
1804 is the first locomotive. 1859 is The Origin of Species by Darwin. 1861 is Maxwell equations. 1869 is Mendeleev's period table. And so on and so forth[0]. Measurement devices also improve in reliability and efficiency, leading to many of the early recordings we can now look back at when it comes to the consequences of the explosion of human activity with regards to the environment.
It's quite a fantastic century to keep a trace of, frankly.
Think about it this way - to view the particular page, article etc you would have to be able to travel to the library in person to view it. If digitized multiple people across the entire globe can access the content. Furthermore printed material degrades over time and at some point becomes unreadable. By digitizing it it remains accessible to future generations. Who knows what may or may not be of interest to future generations. Storage is cheap - historic information is priceless.
39 comments
[ 5.0 ms ] story [ 84.3 ms ] threadNot sure if I'm just not getting the full quality somehow or if the image quality just isn't there for OCR to ever work?
It also seems likely that the jpeg2000 versions are the original, so they may be slightly better, but haven't been able to open one of those yet.
All text is recognized, and as far as I can tell, there are no errors.
There are a lot of difficulties with older text, but I would start with low-hanging fruit such as trying to use better tools. On top of that, you have the problem of making sense of the layout, fixing common typos, etc.
I'm not sure though whether the IA can get Google to OCR it for them for a budget they can afford. Likely they'd want OCR solutions that have a one-time cost, so volume based SAAS offerings won't work.
I think perhaps the developers took a wrong turn when they started trying to improve OCR with language models rather than font models. Humans can accurately transcribe a printed text without knowing the language. They do that with a mental model of font metrics and so on. In fact, a human transcribes more accurately (though more slowly) when they don't know the language because they don't erroneously "autocorrect".
I wonder if Google has in-house OCR that works much better than commercially available OCR. Google has OCR-ed at least 25 million books. You can't download the complete texts, but you can see snippets. Perhaps someone would like to publish a paper assessing the quality of Google's OCR and comparing it with commercial software. (Probably someone has done that already; I'm just bad at finding papers.)
[0]: https://github.com/jbarlow83/OCRmyPDF
https://www.natecraun.net/projects/smoothscan.html
You can feed it an image that contains text, with the only constrained being that the text should not be rotated. Depending how much text there is it can digitize a 12MP image in about .5 to 2 seconds on an iPhone Xs.
I compared it to ABBYYs OCR SDK and i have to say Apple's OCR outperforms it nearly every time both in speed as well as in recognition quality. I am still fascinated by it and would really like to know how this is working under the hood.
[1] and good luck with Beowulf: http://www.bl.uk/manuscripts/Viewer.aspx?ref=cotton_ms_vitel...
Related to the overall culture tech, most of the scientific revolution occurred in NeoLatin which isn't taught, mostly remains untranslated and untranscribed. For instance: Descartes first book, which dealt with music theory and human emotion.
Re their statement: "What we do not have is a good way to integrate work on these projects with the Internet Archive’s processing flow. So we need help and ideas there as well."
... maybe there are still people from the Gutenberg project around, they used to be handling human-transcribed stuff at quite a volume. (Personally I'd rather type stuff in than 'repairing' bad OCR, that constant back-and-forth is just aggravating. I'd be glad to dig in on something personally interesting, like say a little-remembered expedition or research in an area of interest ... they could arrange stuff by subject that way and then put up wish-lists for volunteers. History is full of forgotten amazings.)
https://arxiv.org/abs/2005.01583
https://news-navigator.labs.loc.gov/
What they have done already is great:
https://chroniclingamerica.loc.gov/newspapers/
I’ve used it quite a bit and they have some of the best OCR results I’ve ever encountered when it comes to scanned newspapers. But it’s just the tip if the iceberg. Historical newspapers seems to be one of the largest corpuses of material yet to be digitized.
I guess that would take a tremendous amount of time, but it would be really cool to get an old newspaper article to look at, and I bet a lot of people would find interesting things to talk about.
You could have a "comment thread" for people who added the piece, kind of like adding a digital layer to history.
Besides, as you said, reading the old paper would be enough of a prize in itself.
Less happy about the latest one but the book transcription and map correction were good as I benefited from both those projects (used the tools).
https://www.pgdp.net/c/
We found the document (the observatory yearbook written in 1859) via Google Books. As I don't speak Italian, I initially typed out passages into Google Translate in order to find the information I needed. Google Books has a view text option, but the format of the page and font often made it garbled (when pushed through translate in any case). Decent OCR likely would have made my life a lot easier.
There is a recent trend in space weather research to study extreme geomagnetic storms that happened in the 19th and early 20th century, and is aided partly by all of the scanned documents from that era available on the likes of Google Books, The Internet Archive and HaithiTrust. Better OCR would be a great help.
Although even having access to all of the documents is already incredible!
[1] https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/201...
I do like the idea and efforts from a technical point of view. Tinkering with OCR on unusual (or old) languages. But that's not the goal of this project as far as I'm concerned (it's a byproduct?) Archiving every single news entry for the sake of completion sounds more like obsession than purpose. We're creating so much information that it will be even harder to separate garbage from valuable information (you have to spend time reading the useless stuff before you can justify whether or not it's valuable to you).
Information overload IS a problem and by adding more information to an already saturated ecosystem I don't see the vision here but would like to understand :)
It almost seems like a hording problem but for the digital natives. People accumulate a lot of stuff but rarely can they actually appreciate what they possess as time & perception is a very limiting factor.
An article that would shed some light would be highly appreciated.
We're probably not that good at recognizing which bits will be of interest to future generations, so archiving everything (to a point ... but I believe that newspapers are well within reasonable) sounds like a good idea. Plus you never know what you discover when you make things available.
While this pales in comparison with the 6 billion humans added since, that's a significant change - particularly for most likely available recorded sources - at a time of monstrous evolutions to major world powers/empires, expansion into vast new areas of the world (namely North America) of essentially the British empire (while at the same time the East India Company ceased to exist by the end of the century for contrast), some abolition of slavery becoming a reality in places (1833 for the British), and what arguably kickstarted much of the mental frameworks for our entire lives: the first two industrial revolutions (for example: democratization of once-monastic school system while adopting the year-of-production type of mental model for its promotions).
1804 is the first locomotive. 1859 is The Origin of Species by Darwin. 1861 is Maxwell equations. 1869 is Mendeleev's period table. And so on and so forth[0]. Measurement devices also improve in reliability and efficiency, leading to many of the early recordings we can now look back at when it comes to the consequences of the explosion of human activity with regards to the environment.
It's quite a fantastic century to keep a trace of, frankly.
[0] https://en.wikipedia.org/wiki/19th_century#Science_and_techn...