Semantic Compositions

As the machine laid translating

Courtesy of Arts & Letters Daily, here's a Washington Times book review of a new biography of Edith Wharton that argues for a higher place for her, as well as Willa Cather and Dawn Powell, in the literary canon. SC is regrettably less familiar with the works of Ms. Wharton than he ought to be ([you've read the complete Dune pastiche of Messrs. Herbert and Anderson, but not The Age of Innocence? 'Nuff said -- ed.]), and the article forcefully makes the point that no small part of her comparative neglect is her failure to live a debauched life of crudeness and self-promotion. Put more bluntly, "Misters Fitzgerald, Hemingway, and Faulkner were all alcoholics", and the article follows that up with examples of the hard living that guaranteed them a steady stream of headlines until death.

But the part that caught the eye of this computational linguist was the smackdown of the grossly-overrated Faulkner that went like so:

Miss Powell's New York books re-create a milieu every bit as richly imagined and unforgettable as Mr. Faulkner's Yoknapatawpha County — and a lot more, um, intelligible. "I can't read Faulkner," confesses Mr. Page. "He does absolutely nothing for me."

He's not the only one.

Some enterprising soul has posted on the Internet "Machine translation or Faulkner?" — a quiz asking you to deduce whether quotations are computer-translated text from the German or samples of Mr. Faulkner's prose.

Needless to say, SC immediately Googled "machine translation or Faulkner?" and was rewarded with this site. The machine-translated German texts are revealed to be largely appropriate literary comparisons, albeit not of Faulkner's writing (think Goethe), although some might hold that an ideal test would be a machine translation of a human translation of Faulkner's writing back into English. SC scored a meager 42% on the quiz, and was frankly shocked to see that some of the questions he got wrong revolved around quotes from Faulkner works that he's actually read. Evidently your host didn't read them attentively enough.

No hockey for you, s'il vous plais

Via ESPN's Page 2, not normally a source of stories described as poignant or controversial, comes the incredible news that an 8-year-old ice hockey player in Quebec has been banned from his league because he doesn't speak French. SC has written about Quebec's outrageous Bill 101 in the past, but this is a new low in language politics. Or is it?

About the only fact of the case that is not in dispute is that Jared Murray, the player in question, is in fact 8 years old. This matters because, as Gare Joyce summarizes the decision to ban him:

1. He's too big to play with other 8-year-olds. At 4-foot-9 and 110 pounds, he's at least a head taller than the other kids on the ice.

2. He's too good to keep games fair and competitive.

It's far from unprecedented for kids to have their ages questioned if they seem a little too big or too talented; witness the cases of Danny Almonte or the early-90s scandals that cost the Dominican and Philippine Little League World Series teams. In this case, however, nobody is disputing his age. As for the second issue, whether or not he's simply too talented and ought to be playing in a higher league, young Mr. Murray scored 51 goals in 17 games, which is admittedly a very impressive number. However, as this hockey blog points out, Wayne Gretzky scored 6x as many goals at the same level of competition without getting bounced out of the league. If you consider Gretzky too extreme an example to be relevant, consider that the allegedly too-dominant Jared Murray only managed to bring his team to a third-place record in a 16-team league.

Let's stipulate for the sake of argument, though, that these issues at least ought to have been debated by the adults in charge of the hockey league. Canadian novice hockey has two categories of play, with A and B-level teams so that boys of roughly equal talent level can face off against each other. Mario Lemery, the president of Hockey Outaouais (the governing body for the league in question), claims that the team was asked to move to the A level:

"We asked if the complete team would go up to the A," Mr. Lemery said. "They didn't want to. It's no fun if they cream everyone else, 7-1, 8-1. They should play in A. No one wants to play against those teams."

He said Fort Coulonge agreed to move up. After the Shawville team wouldn't move up, he offered to let the rest of the team play in the B division as long as their star player didn't play. That's when the whole team decided to boycott.

"At the beginning of the year, they said we should play in the B because we're not strong. They were not honest at the beginning of the season," Mr. Lemery said. "I asked them to go A and they said no.

"It's no fun squashing every team. No one wants to play the team."

In other forums, though, the team's coach claims that this request was never made (see this CTV.ca story, and also this comment at the HockeyAnalysis.com blog by an individual claiming to be the coach).

Since nobody from either side is making an admission against their interest (i.e., the hockey team acknowledging getting a request, or the league acknowledging it wasn't communicated in English), it's hard to be sure of precisely if or when Jared Murray's eligibility was first questioned. Is it possible that the request was made through a letter sent only in French? It's not an unreasonable hypothesis. A look at the official Hockey Outaouais website indicates not the slightest concession to English-speaking members of the league; not one page is available in anything other than French. This makes it easy to credit the claim of Gare Joyce's original ESPN.com article that "[N]ot one of Shawville's volunteer officials went to league meetings where only French was spoken (emphasis added), which basically shut out the town where few residents are bilingual."

Ordinarily, when a language majority appears to be running roughshod over language minorities, the debate revolves around whether there should be an outreach effort by the majority (i.e., creating bilingual documents, providing translators) or the minority (i.e., working to become bilingual in the majority language). This scenario breaks the mold because while the majority/minority roles are obvious in context, the French speakers are also in the minority role in the larger context of Canada. Revisiting the HockeyAnalysis.com discussion, a number of comments from disgruntled English-speaking Shawville residents suggest that they feel Hockey Outaouais has a pattern of timing decisions like this to make it difficult for Shawville players to compete. It's possible that some, maybe even all, of the decisions made by the league's officials, are correct on the merits. But it's also possible to imagine sitting down over drinks with Hockey Outaouais officials, and hearing them say (in French, of course), "Payback's a bitch". SC certainly won't claim to be able to solve the historical grievances that underlie Canadian language politics. Would it be too much to ask for them not to be inflicted on 8-year-olds, though?

A taxing argument

Something feels a little off today. It's the deadline for filing tax returns in the United States, but it's not April 15th. Not that taxes have been due on April 15th fo eternity -- for the first half of the 20th century (well, that part of said first half after the income tax was first imposed), it was March 15, only becoming April 15th after Congress changed it in 1955. According to the IRS, whenever April 15th falls on a weekend or Federal holiday, you have until midnight of the next business day, a fact which pushed taxes to April 17th just last year. No, it's the April 17th-is-a-Tuesday part which makes it feel off, due to the fact that Washington D.C. now has a holiday on April 16, and again according to the IRS, if D.C. wants to nap, we're all going to take a snooze. Wait, that's not exactly how they put it (and in all seriousness, it's actually a good reason for a holiday).

Tax day has a tendency to bring out SC's inner libertarian, especially since the mortgage deduction didn't apply to him this year (oh, how bitterly he's missing Chez SC right about now). And that brings us to a phrase with a surprisingly contested meaning, "Tax Freedom Day".

The basic concept is simple enough -- if you consider the part of your earnings that you get to keep every year after taxes to be what you've earned after you've made enough to pay your taxes, then simply multiplying your effective tax rate (what you actually pay, not your marginal rate, which only applies to a portion of your income) by 365 tells you how many days you had to work to pay your taxes. One could quibble with certain issues there -- salaried employees don't generally get paid for weekends, there are a variable number of holidays depending on where you live -- but the calculation wouldn't change overly much even if you did correct for these things.

There are good reasons why you wouldn't want to publish your personal tax freedom day given the above methodology, though, which is that it's a handy way for people to figure out roughly what you earn. With the exception of people whose salaries are required by law to be published, this is generally Not Anyone Else's Business (except for the IRS). So the Tax Foundation, which holds the copyright on this idea, publishes an aggregate figure for the U.S. as a whole, as well as state-by-state. And every year in rebuttal, an organization called the Center for Budget and Policy Priorities publishes an argument that the figure is misleading ((the first two pages of the Google link will suffice to get you the last decade's worth of press releases). The argument seems to revolve around the meaning of the word "average". On one side, the Tax Foundation computes total tax receipts divided by total income; on the other side, the CBPP argues that the average person falls in the middle quintile of taxpayers, and since those people don't pay anywhere close to the effective tax rates implied by the Tax Foundation figures, the notion of tax freedom is highly misleading. You can read last year's version of the argument here (the Tax Foundation side) and here (the CBPP side).

Arguments about the right choice of statistic to mean "average" have been going on forever. The Tax Foundation clearly would prefer that you think of the arithmetic mean; the CBPP would prefer that you use the median. There are enormous policy consequences that follow from your choice in this regard, as the former is useful for arguing that taxes are too high, and the latter for arguing that they're either too low, or at any rate, not onerous (the question of what the money is being spent on is certainly related, but separable for our purposes). SC is sure of one thing, though, which he'll bet both sides can agree on: when the mean and median taxpayers get in line at the post office today, they'll both be grumbling.

A bit of Passover lexicography

Your host has written himself into some awkward situations in the past when discussing elements of the Passover holiday, but this year, before it's over, he wants to wade into something Passover-related anyway.

The subject this time mercifully takes place completely in English, so no translation issues are involved. In particular, it revolves around one of the very few Passover foods SC is willing to eat year-round without any hesitation. No, not Joyva Ring Jells -- per the links above, SC does not consider them Passover food anymore (although he plans to eat some after the holiday). In this case, SC speaks of the least prototypical food ever to wear the name of cookie, the tricolor:

(Picture borrowed from Gothamist.com; if they object, SC will take it down.)

A truly prototypical cookie is round, features bits of the filling (if any) interspersed throughout the dough, and is frosted one side at most. The tricolor breaks all of these rules, and is assembled like a layer cake -- if "snack cake" wasn't a coldly corporate phrase suitable only for printing on Twinkie wrappers, it would probably be the right descriptor for tricolors.

Now here's the linguistic hook. Of all the bakeries he's ever tried, SC feels that the best tricolors are made by the Beverlywood Bakery (sorry, no website -- they just started taking credit cards last year), in the Pico-Robertson neighborhood of Los Angeles. SC held this view long before he actually moved to Los Angeles, and it isn't just something that's developed of convenience. However, the Beverlywood Bakery does something that just about no other bakery does in making tricolors -- they add a fourth layer, tinted with brown food coloring.

The hyper-logical Mrs. SC has been all over this one, saying that there are but two reasonable linguistic behaviors for your host to pursue. One is to acknowledge the impossibility of calling the Beverlywood cookies by the tricolor name, and to refer to them as tetracolors. The other would be to adopt the name actually used by the Beverlywood employees, and call them rainbow cookies.

While SC has had to adopt the Beverlywood nomenclature for purposes of orrdering, both of these proposals strike him as unacceptable. Tetracolor is fish food (and yields no hits for the phrase "tetracolor cookies" in Google; ditto for "quadricolor", in case that crossed your mind). As for "rainbow cookies", well...this guy gets it just about right. Rainbow cookies just seem so lacking compared to the mystique of the tricolor (not quite his point, but read the post).

Having said that, we're all about hard linguistic science at Semantic Compositions, and the usage evidence is more than a little in Mrs. SC's favor, as often happens precisely at those moments when your host's neck is most stiff. "Rainbow cookies" stomps all over "tricolor cookies" by a 130:1 margin, dropping to 20:1 if you add the hyphenated version of tricolor. This result holds up even more graphically (pardon the pun) when you use Google's image search; you get just four genuine hits for "tricolor cookies" there, but valid results can still be found for "rainbow cookies" into the fifth page. Against that, your host can only hold up the argument that "rainbow cookie" also pretty clearly refers to a bunch of cookies which bear no relationship at all to the type under discussion (see the aforementioned image search results), and the comment of the author of The Pop View, who indicates that he knows the cookies in an Italian guise as tricolore. OK, there's also an appeal to etymology; one of the commenters in the Gothamist post linked above writes:

i knew the "rainbow" cookie debate would eventually rear its head. my position is that these cookies should only be called "tri-colors" because they are based on the Italian flag- a flag that is often referred as a the italian tri-color.  calling them rainbow cookies is like calling a hamentashen a "triangle" cookie- it doesn't do justice to the history.

But SC can't really hang an argument about what to call a food produced for a specifically Jewish context on what the same thing is called in an Italian context. Can he?

The ultimate Natural Language Processing-based scheduler

SC has long been plagued by the boring tedium of the scheduling tools built into Microsoft Outlook. Every day, he'll get a half-dozen identically written meeting requests, all automatically generated from a couple of safe, inoffensive templates. These things all tend to run together after a while, but they work because businesspeople have discovered a powerful method for manipulating other people -- if you place a little automatically-generated reminder in other people's calendars (assuming they let their lives be dictated by a single piece of unimpressive, buggy proprietary software), they will robotically show up for meetings, call into teleconferences, or otherwise do what you want. The downside is that in order to make it effective, you need to set the alarms on the reminders to go off just minutes before the intended event, and the illocutionary force of "Please call into 500-555-1234, passcode 5678" is rather limited.

Nobody knows better than linguists that speaking the right language to people can improve the success of your communicative goals, and so your host decided to apply his knowledge of both linguistics and Java to create the world's first scheduling program fully backed by the power of natural language processing. Behold -- the iNag!

iNag starts with helpful user-nameable contact names that will never be reproduced in your e-mail, so you can group people into the mental categories you actually file them under. iNag will integrate these names with names in your Outlook address book so that "The Ones You Invite To Social Events Or Else" never know that you feel that way about the Vice President of Operations and the annoyingly chatty secretary down the hall.

But iNag is so much more than a tool for commerce -- the real power behind it comes from the REiterating Mail INtegration Device (REMIND), a sophisticated natural-language generator that not only can adopt your persona, but customizes reminder messages as important events grow closer. To demonstrate, we'll walk through an example useful for business and personal life -- sending a birthday card.

You begin by selecting from a library of personas appropriate to business or personal life:

Then you set the event type that you're going for, a date, and reminder notices, just like in any other scheduling application. iNag will then produce a customized set of notifications and escalating reminders. If the message recipient doesn't get their task done, iNag will automatically send out escalating reminders as the deadline gets closer. Here's how iNag would remind your recipient to send a birthday card, in a couple of styles.

First, the Jewish Mother:

The default manager persona:

And former coach of the Minnesota Vikings and Arizona Cardinals, Dennis Green:

iNag with REMIND technology -- coming soon in the Home and Office Productivity aisles of fine retailers like Staples, Office Depot and Best Buy near you (just as soon as Semantic Compositions Software can get some venture funding).