Hacker Newsnew | past | comments | ask | show | jobs | submit | lokl's commentslogin

Has there been any progress on undeciphered languages?

The training set is rather sparse for those.

I hope not, I'd hate for my enigmatic Linear-A tattoo to end up reading "Fish monger".

“This stele intentionally left blank”

Incidentally, Mie Prefecture ranked 37 out of 47 Japanese prefectures for overnight stays by foreign tourists in 2024, but it has a lot to offer in addition to Ise Jingu. For those seeking the "real Japan" beyond Tokyo, Osaka, and Kyoto. If you want to explore, consider Kintetsu's 5-day plus pass and fit this into your itinerary between Nagoya and Osaka, Kyoto, or Nara.


please god do not. japan is already chock full of tourists. please don't overrun one of the most sacred places in japan. kyoto’s most important shrines and temples have become basically a theme park packed with tourists and locals barely visit anymore. just leave Ise alone.


I understand your concern and agree about Kyoto, but tourists can behave well or poorly and tourism can be managed well or poorly. At Jingu, I saw foreign tourists, but no problems, and locals were welcoming. Elsewhere in Mie, I saw no foreign tourists and locals were also welcoming. I would like to believe that HN readers would be beneficial tourists and might themselves benefit from experiences that cannot as easily be found in the top destinations.


I think this mentality breeds anti-foreign sentiment, which is then further fueled by the actions of a few bad apples. Japan isn’t the only country facing overtourism in certain hotspots. Also, the non-foreign tourists visiting Ise are simply Japanese tourists. It’s about managing tourism and finding a way to accommodate all kinds of visitors.


If you are interested in seeing original cave paintings, not reproductions, I recommend Pech Merle near Cahors, France.


What about SPARK? Not enough training data?


Please compare with FineReader.


I've been using Sphinx for 20 years for full-text search with a custom stemmer. I considered switching to Manticore, but didn't see a huge need to do so, because Sphinx still works well for me and requires zero maintenance. Any big new features that might entice me to switch? (I only have a few GB of indexes, covering a few million documents.)


If your setup works fine and doesn't need any maintenance, there's really no reason to switch to something else. But once you upgrade to a newer version of Sphinx and it crashes, I personally feel more comfortable knowing I can report the issue on GitHub and expect it to be fixed eventually. Unfortunately, that's not how it works with Sphinx.

Speaking of features, both Sphinx (as a closed-source project) and Manticore (as an open-source project) have added some nice improvements during last years. But again, if you're happy with the 20-year-old version, there's probably nothing to worry about.


Im a fan of manticore and promote it when I can, but this doesn't seem like an approoriate response. It appears generous on the surface, but has various unnecessary and subtle digs in it - all while it would be easy to show how manticore is a more appropriate choice.

They're surely not using a 20-year-old version of Sphinx - it appears that the latest open source version on Github is 8 years old. They have a closed source version which appears to be maintained, though has infrequent releases.

Why not just say these things, which make manticore self-evidently a better choice?

https://github.com/sphinxsearch/sphinx https://sphinxsearch.com/downloads/current/


If you have the opportunity to visit, I recommend Nagasaki over Hiroshima and especially these two places in Nagasaki:

Shiroyama Elementary School

Nagai Takashi Memorial Museum Nagasaki

These felt much more personal than anything I saw in Hiroshima and there were zero (other) tourists to interrupt the experience (very much unlike the museum in Hiroshima).


Used cameras and lenses can offer tremendous value.


Exactly. And the "sensor size myth" is nothing a kid should (or would) care about.

There are some very good YouTube channels talking about micro four thirds cameras, which are still a good choice, especially when used as a camera to carry every day.

Disclaimer, own and use full frame, APS-C, and even Ixus and Powershot cameras which all can produce decent images if one knows how and when to use them. Oh, smartphone, of course.


This. My humble setup is a Sony a100 with some good old Minolta lenses.


My first SLR was an A100.

Fond memories, I carried that camera all over the world for 6 years.


Yep, an online acquaintance gifted me an Olympus 35 RC he hadn’t touched in decades. My 35 DC was <$200 last year. My E-M5ii was <$500 years ago.

Yeah, the first two are 35mm film, but they’re phenomenal cameras within the scope of what they are (fixed lens rangefinders - basically the 1970s version of today’s Fuji X100). The E-M5 hasn’t let me down, and the latest models from OM don’t offer enough to make me upgrade (and I have little desire to switch mounts).


Tried with a few historical handwritten German documents, accuracy was abysmal.


Semi-OT (similar language): The national archives in Sweden and Finland published a model for OCR:ing handwritten Swedish text from the 1600s to the 1800s with what to me seems like a very level of accuracy given the source material. (4% character error rate)

https://readcoop.eu/model/the-swedish-lion-i/

https://www.transkribus.org/success-story/creating-the-swedi...

https://huggingface.co/Riksarkivet

They have also published a fairly large volume of OCR:ed texts (IIRC birth/death notices from church records) using this model online. As a beginner genealogist it's been fun to follow.


HTR ( Handwritten Text Recognition ) is a completely different space than OCR. What were you expecting exactly?


It fits the "use cases" mentioned in the article

> Preserving historical and cultural heritage: Organizations and nonprofits that are custodians of heritage have been using Mistral OCR to digitize historical documents and artifacts, ensuring their preservation and making them accessible to a broader audience.


There is a difference between historical document and "my doctor prescription".

Someone coming here and saying it does not work with my old german hanwriting doesn't say much.


You're making a strawman, the parent specifically mentioned "historical handwritten documents"


For this task, general models will always perform poorly. My company trains custom gen ai models for document understanding. We recently trained a VLM for the German government to recognize documents written in old German handwriting, and it performed with exceptionally high accuracy.


Probably they are overfitting the benchmarks, since other users also complain of the low accuracy


Also working with historical handwritten German documents. So far Gemini seems to be the least wrong of the ones I've tried - any recommendations?


my recommendation is to train a custom model


Optical Character Recognition (OCR) and Handwritten Text Recognition (HTR) are different tasks


MUDs were the most enjoyable social networks I ever experienced, and the least harmful to society.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: