Regular Progress

I like to keep the news relatively fresh, even if nothing major is happening. But even though this is a somewhat slow time for Numen development, there are still a constant stream of small updates. I'd like to keep everybody apprised of the situation.

One of the benefits of having students who actively use this dictionary is their feedback. One of the things they noticed was that sometimes an error message pops up saying, "AJAX has timed out", and it would happen relatively often -- especially on the flashcard practice tool. So I dug into the code and found the issue: I designed the site so that -- if any request took more than a few hundred milliseconds -- it would give an error. Such timeouts always involve a balance between briefness and lengthiness of waiting, and I quickly realized that I had not struck that balance. So I bumped up the timeout duration to something reasonably middle-of-the-road and. Voila! Problem solved! Gratias vobis ago, discipuli.

More: My big project with Vergil and Livy is still paying off. I continue to correct dozens of tiny mistakes and errors in the data every week, and I was able to run some statistics. Excluding proper names and place names, the parsing engine can analyze and pin down around 98.5% of all the words in these two Augustan authors. So accuracy is definitely improving daily! I can't yet account for all false positives, but they seem to be less than a fraction of a percent (anecdotally).

Speaking of accuracy, there is still room for improvement in three key areas:

1. syncopated forms (laudaverunt => laudarunt), which Livy loves by the way!
   
2. irregular forms (bobus, filiabus)
3. proper and place names (which do not, for the most part, exist as a regular part of the Lewis Elementary dictionary).
   

I have been cogitating over solutions for all three issues, but it might take a while to implement them -- probably not until December or January (winter holidays, yay!).

In other news -- I guess I have more than I had first assumed -- I've almost got a word list feature finished. This is for people who prefer to work with formatted word lists as opposed to flashcard decks (which I understand are sometimes referred to as index cards). The only major problem I have with this process is that the Lewis Elementary dictionary does not provide a "core" definition for most words, so the word list would have extremely lengthy definitions. I think one of my best options is to import the data from Whittaker's Words, which have more simple, more core-like definitions. But this could be problematic as a 1:1 mapping between Lewis' forms and Whittaker's forms would be difficult to achieve. Alas, I shall continue to think on this one.

Okay, so that's enough for now! Keep using it, and please keep reporting problems and errors! It may take a few days or weeks, but I eventually do fix all the errors!

IE8 Flashcard Bug Fixed

Salvete omnes!

I fixed a small bug that affected flashcard decks in Internet Explorer 8 (and presumably earlier versions). If you couldn't create a flashcard deck in that browser, it should be fixed now!

As a side note, I've been working on a big project with Livy and Vergil. I've essentially been editing all the mistakes and unfound words in those authors. This is especially useful in Livy because we have a corpus of about 1 million words! So the accuracy of this dictionary is creeping up to the highest possible levels! With the exception of proper names and place names, I'll ballbark its accuracy with common classical authors at about 95%.

Also, thanks to the people who have been reporting errors and bugs! It's really helpful to have your feedback!

J's and U's Updated / Speed Increases

I mentioned a few weeks ago that I planned on making I's/J's and U's/V's look the same on the back-end, while preserving their traditional orthographies on the front-end. I've just completed this task!

My main motivation for making this update is because certain passages stored in The Latin Library reflect the older conventions of using J's for consonantal I's or U's for both consonantal and vocalic V's. Numen's parsing engine was having trouble recognizing forms like jecit (iecit) and uuius (vivus). So now as a result -- after a bit of work -- the engine is updated and now recognizes more possibilities than ever. Incidentally, internally J's are stored as I's and U's are stored as V's.

Another project I completed at the same time is an order-of-magnitude speed improvement for parsing. I was trying to figure out ways to make the engine faster and I discovered a shortcut that boosts speed tremendously. When parsing a word, the engine used to spend between 250ms and 500ms parsing each word! That was always disappointing to me, but I had gotten around the problem by caching the results. Now, however, word parsing takes about 25ms!

Why bother improving the speed? Because soon I will be implementing word lists and frequency lists! A word list, of course, is just a "mini-lexicon" that defines only the words in your chosen passage, and a frequency list is a list of words in order of how often they appear in a passage. The word list will be helpful to quickly work on vocabulary for a passage, and a frequency list will help Latin students study more effectively by giving them the most frequent words first. I'm very excited about this feature, but I don't anticipate it will be done before January 10th (giving me the winter holiday to work on it).

That's all for now!

I's and J's and U's and V's

So one problem with Numen is that it doesn't recognize the different possibilities when dealing with I's and J's and U's and V's. As you know, the J and the U were not Classical Latin letters. There has been a lot of back-and-forth over the past 200 years -- some editors prefer the originals and some prefer the modern versions.

But how should Numen deal with this issue? Internally, the computer is more precise and less forgiving than a human, and so in order to provide highly sensitive and accurate searches, the data needs to be "normalized". For example, I recently normalized verbs for consistency by changing all deponent verbs into their active forms and simply marking them as deponent with a data flag. Now, when you search for a deponent verb, the flashcard still shows something like sequor but internally it's stored as sequo. The reasoning here is simple: deponent verbs, regardless of their dictionary form and traditional morphology, still have active participles and their imperfect/pluperfect subjunctives are still formed from active infinitives.

But what about the I's and J's? Those are easy. Convert all the J's to I's, and most Latin readers won't have a problem -- this has been the convention for quite some time now. But then what about the V's and U's? Should I convert all the U's to V's? The opposite is true here: most Latinists would be mildly irritated by this form: uiuus (vivus).

The solution, which would be similar to the one for the deponent problem, would be to mark internally everything with I's and V's but then show the contemporary I's and U's and V's to the end users. That way, the computer can do accurate searches, but users get the information they are used to.

So, in the coming weeks, Numen will undergo this under-the-hood transformation. For the most part, users will never even notice -- except in one area. Searching for uiuus will be the same as searching for vivus!

New Server

I apologize for taking down Numen for a few hours! But it was for a good reason. A couple of weeks ago (as I posted) the server died, and I had to jury-rig a (slow) home computer to be the server.

Well, just yesterday a bunch of new parts arrived via UPS and I quickly built, tested and verified the new server. I took down the old one about 11am, copied over all the data, and fired up this new one just a few minutes ago.

This new box is quite a bit faster than the home computer which was the temporary surrogate, but you'll be happy to hear that it's also zoom-zoom-zoom fast compared to the original server! It's somewhere in the range of 2 to 3 times faster overall.

The computer power users might be wondering what the new server consists of. It's a very simple configuration. It's a dual core Athlon 245 (?) Athlon II 2.9GHz processor, an nVidia-based AM3 motherboard, 4GB of DDR-2 1066, an Antec EarthWatts 380W power supply and a WD Green 1TB hard drive. I was specifically trying to build a power-sipper here. My best guess -- because I forgot to bring a Kill-A-Watt to test it, is that this server runs somewhere around 85W and around 110W at peak. I wanted to get an Intel SSD (160GB Generation 2), but ... I felt that the performance gain from that wouldn't outweigh its cost (around $420). Now why does that matter? Because this entire computer, including shipping and a $20 mail-in-rebate was about $292! Nice!

Okay, enough geeking-out for now. I hope you enjoy using Numen, and as always, feedback is welcome!

Server Outage / Restored

You might have noticed that this site was down for about a day. That's because the web server crashed! Well, that's a simple way of saying something more complicated. Don't worry, absolutely no data was lost. I did, however, have to load everything up on a smaller, older, slower server -- so what that means is that the site is going to be a bit slower until the replacement server arrives. It's under warranty so it won't be a problem to replace it, but it might take a few weeks.

If anything is broken, feel free to email me to let me know!

DNS Lookups Cause Site Slowness

As per the usual, I've been tinkering with the back-end from time to time. Things slowly and silently improve -- well, hopefully!

Right now I'm at the University of Kentucky participating in an intensive conversational Latin seminar. Intense is the appropriate word. Today is day 2 of the seminar itself, although I've been here since the 4th of July. We'll see how my Latin improves after 10 days of immersion.

Since I've been here at UKY I found a bug -- well, a performance issue. Some Universities (like this one) don't put individual host names in DNS for their various thousands of connections. UNM happens to be one that does. Normally, my web server is set to perform a DNS lookup on every single connection (inefficient, I know). For a normal internet connection (say, a UNM student or a home Comcast user), that lookup takes microseconds. In the case of UKY, that lookup took 6 seconds ... to timeout. That timeout happened every single time the browser connected. Needless to say, this site must seem incredibly slow at a University like this. I can verify that it did for the short time it was affecting me.

So, to all the UKY students out there ... I apologize for the slowness. I also apologize to all the users of this site who experienced the ... shall we call it a mismatched configuration? On a pleasant note, this was a truly easy situation to resolve. One # (hash) sign in the config file and all is well. C'est la vie? Oui. Or in Latin, perhaps Re vera est? Ita.

Incidentally, I clean up words on the back-end on a regular basis. Since I've been intensely studying here, I've been catching a few more errors than usual!

I also had a suggestion to incorporate the full Lewis and Short dictionary into this site. Great idea! I'm looking into it now.