practical tips for confused erics

Some of this is stuff I only figured out this year.

Yes, it is highly prescriptive and rigid. That's the whole point. Bordelique people like me need a system with hard (but ridiculously easy to follow) rules. It may not work for you. People like me need to think about these things deliberately. The goal is to have something that's reproducible and that you can use without further thought. The moment you have to think about where to put something, your system is broken.

Space

• Five things to have within reach
  1. inbox - to be emptied daily
  2. meetings tray - very few things: meetings book, folder for upcoming meeting(s)
  3. thematic folders - scientific GTD project support, e.g. SCFGs, SMT, SDRT, content representation, etc
     
  4. scrap paper - hoard this stuff, I go through reams of scrap paper (learned this trick from Carlos and Claire)
  5. big stack of empty folders - for creating thematic folders
     
• When no longer likely to need an article, file in cabinet: alphabetical order by first author's surname
• Date each piece of scrap paper on the top-right corner as you are using it; a little label may be useful too (the point is to simplify the questions "wtf is this? can I chuck it?")
  
• Have a log book dedicated to meetings - date each meeting
  
• If you want to store something long-term, consider putting it on a computer (e.g. re-do back-on-envelope sketches in inkscape) and ditching the scrap paper
  

Time

• Use brain-dead time to do stupid-but-useful things
• Only do stupid things (e.g. re-arranging office, writing blog entries) when brain-dead (rule I am currently violating... then again, when am I not brain-dead?)

Stuff

• folders
  1. long term storage - simple manila if possible, hanging if you really have to
  2. thematic folders - simple manila
  3. meetings folder - pochettes, basically fancy folders
  4. projects folder - pochettes
  5. super-important (immigration papers) - plastic pochettes
• music - keep a Speechless playlist for things you can listen to whilst reading

dead eric day / dissertation title

Today was a day of erician zombiehood, a combination of (a) neighbours across the street and their stupid 4am parties [thump thump boom chicka boom chicka boom thump thump] and (b) overindulgence with the rosé last night (ran into some people hadn't seen in ages, socialised for the first time in a long while). Spent early morning sort of puttering around at home... actually, no, just plunked down in front of my computer and not wanting to move.

thesis title

Spent late morning figuring out what the title of my dissertation was going to be. My first cheesy idea was something like "The threefold path: seeking determinism in surface realisation". It had to be a French title, so then I tried "L'ambiguïté et le déterminisme dans la réalisation de surface"... then me and C sat down to think about it (she didn't like my French title)... decided we couldn't really do better, although maybe we could do away with the determiners and the preposition... and that's it, eureka! The title of my thesis is going to be

Réalisation de surface : ambiguïté et déterminisme

The English one will probably be the same

Surface realisation: ambiguity and determinism

although for some reason it sounds nicer to me in French. But I guess I'm sticking with it

new filing system!

Another semi-useful thing I did to avoid doing real work was to clean up my desk (yay! it makes a huge difference when you can just open a book on your desk, take notes and stuff) and re-file all the papers I had printed out, half of which were in the drawer and half of which were haphazardly stacked on the desk. Was getting out of hand, couldn't find anything anymore. I used to have thematic filing, like a folder for "Generation" and a folder for "Parsing" and one for "TAG", etc... but that didn't work at all because papers will span more than one category, so you can never remember where you put something. My new filing system is simple and stupid: alphabetical by first author, one folder per letter. What I figure is that

1. it's easier to retrieve things, because I usually remember the author (cite key) of the paper I want, or if not, can always look it up on the web
2. it's easier to file things because I don't have to think about what the right place to put things is (GTD technique - reduce mental friction)

It's a bit late in the game to be working out your filing techniques, but at least it got me motivated to clear all the shit off my desk. And clean desk is really really great.

chart generation

Spent some more time reading kay1997cg and neumann1994ucm, mostly trying to fit together a coherent story on the chart generation stuff. I'm kinda confused about neumann1994ucm, like (i) how come SEL(x/Ea) is already instantiated (ii) are SF and SEL the same thing? Random list of generation issues:

• lfe, can't find a canonical logical form, BUT at least the trivial stuff shouldn't have to get in the way - kay1997cg (this is my umpeenth re-reading, but each time I understand something new
• indexing - shieber1988uap basically has no indexing / kay1997cg uses flat sem indices, distinguished index... but you don't really NEED a flat sem for indexing / neumann1994ucm just goes and uses the whole semantics (or string) whydoncha
• intersective modifiers - what's interesting is that we can think of the kay1997cg internal index mechanism as a disambiguation strategy (smart pruning) / carroll1994ecg solves it a whole other way, by picking a nice tree traversal strategy (delayed adjunction). You can attack the same problem from wholly different angles
  

Also bothered by the flat sem stuff... want to introduce that in its own chapter after the main issues one, but I guess I also need the flat sem stuff to talk about the chart gen stuff, so maybe a quick gloss-over intro... don't know. Probably also need to do the kay1997cg -> carroll1999ecg link.

traversal / disambiguation / control

I think this way of breaking it down could really work

1. traversal - avoiding non-determinism
   • horizontal = left-first / head-first
   • vertical = top down / bottom up / x-corner
2. disambiguation - making the right choices
   • search ordering - depth first / breadth first / best first
   • commitment - commit to choices / backtracking
   • pruning - nipping bad choices in the bud
3. control - recomputation vs charts

It's a gazillion-dimensional problem, making a parsing algorithm...