psych0naut

My former teacher, Svava Skúladóttir, is now offering Icelandic lessons online at her site Online Icelandic Lessons. I took evening courses from her for several years at University College London, and I can personally attest to the quality of her instruction. This latest undertaking of hers is a fantastic opportunity for those who aren't lucky enough to live near a school that offers Icelandic courses (which is probably most of the world).

Dinner last night: smoked mackerel and Nordseekrabben chowder, salad with garlic–chili yogurt dressing, and raspberry–elderflower smoothie.

In January I posted about the Kaiserschmarrn we often have for breakfast on weekends. My grandfather saw the post and suggested that we try my grandmother's recipe for grízsmarni, the Hungarian version.

As you can see above, it turned out pretty good, but was a bit drier than the Bavarian recipe we had been using. The Hungarian version distinguishes itself by the use of semolina, which I think gave the dish a less fluffy texture. If anyone's interested in comparing for themselves, here's my grandmother's recipe:

Grízsmarni

Ingredients

• 125 g flour
• 125 g semolina, farina, or Cream of Wheat
• 250 mL milk
• 2 eggs
• 25 g sugar
• 3–4 tbsp cooking oil
• pinch of salt
• 50 g ground almonds or ground walnuts
• 50 g icing sugar
• jam

Directions

Mix the flour and semolina, stir in the milk, and let stand for one hour. Beat the egg yolks with the sugar and salt, and in a separate bowl beat the egg whites until firm. Stir the egg yolk mixture into the batter, then fold in the egg whites. Heat the cooking oil in a large frying pan and pour in the batter. When the bottom becomes light golden brown, turn the pancake over, cut it into pieces, and continue to cook until golden brown all over. Serve with jam, ground almonds or walnuts, and icing sugar.

Serves two.

My current research involves word sense disambiguation, an open problem in natural language processing concerned with determining which meaning of a word is used in a particular context. For example, in English the word "bank" can mean (among other things) a financial institution, or the edge of a lake or river. When we see or hear the word "bank" in a sentence, which of its meanings is intended is usually made obvious by the context. For example, if I say to you, "I'm going to the bank to get some money", it's obvious to you that I mean "bank" in the financial sense, because you know that money is often obtained from financial institutions, but not so often from bodies of water.

Being able to disambiguate word senses is important not just for understanding texts but for translating them as well. Consider that the different senses of an English word are often translated into entirely different words in a foreign language. To continue with our "bank" example, the "financial institution" and "river edge" senses in Russian are банк and берег, respectively. Therefore it's not possible to correctly translate the word "bank" from English into Russian unless you know what sense it's being used in.

The process of word sense disambiguation is largely unconscious and automatic in humans, but it's quite difficult for computers, which lack the real-world knowledge necessary to make connections between word senses. Why should we care if computers know how to disambiguate word senses? Well, consider the problem of machine translation: if you type the aforementioned sentence into Google Translate and expect it to translate it into Russian, you would hope that it would render the word "bank" as банк and not берег. Unfortunately, Google Translate, like every other state-of-the-art machine translation system, often gets word senses wrong, resulting in comical or nonsensical translations.

One word sense disambiguation algorithm I'm working with was first described by Roberto Navigli and Mirella Lapata in their 2010 paper "An Experimental Study of Graph Connectivity for Unsupervised Word Sense Disambiguation". This algorithm uses WordNet, which is like a huge computer-readable dictionary of the English language. WordNet lists all the senses for English words, and moreover lists all the semantic relations between word senses, such as when two senses are synonyms or antonyms. Navigli & Lapata's algorithm starts by building a huge graph of WordNet where the vertices represent word senses and the edges represent semantic relations between them. Then, given a sentence to disambiguate, it consults WordNet to find all the possible senses for all the words in the sentence, and then adds the corresponding vertices (but not any edges) to a new "disambiguation" graph. For each of these vertices, it then goes back to the WordNet graph and searches for a path from that sense to any of the other senses; if it finds any then the intermediate vertices and edges are copied from the WordNet graph to the disambiguation graph. Once the search is over, each vertex in the disambiguation graph received a numerical score equal to the number of edges containing it. For each word, the sense with the highest score is chosen as the correct sense.

This week I implemented this algorithm in Java using the JUNG graph library, which it turns out has some nifty visualization features. Below is a video I made of the algorithm disambiguating the sentence "Drink the milk", where the computer knows in advance only that "drink" is a verb and "milk" is a noun. At the beginning of the animation you see the unconnected disambiguation graph of all of the various senses of the verb "drink" and the noun "milk"; these vertices are coloured blue. The path searching then begins, and you can see the algorithm add new edges and vertices (coloured red) connecting the blue vertices together. Once all the paths under a certain length have been found, the algorithm calculates the score of each vertex, and then highlights in green the sense of "drink" and the sense of "milk" with the highest score. As it turns out, in this case it got the correct answer.

Þjófur has biting problems.

One of her favourite things to do is to walk up to innocent, unsuspecting people (like Nadya or myself, but also random houseguests) and sink her teeth into them. Once she's firmly attached (which necessarily requires her sharp little fangs to have fully pierced the skin), she likes to squeeze as hard as she can and shake her head back and forth, as if trying to tear out a chunk of flesh. This is very painful, not to mention very messy, what with all the blood oozing out.

She's not doing this because she's a mean or vicious ferret. She's happy to let herself be picked up and held and petted and cuddled, and loves to play with people in various nonviolent ways. But for some reason, whenever her jaws come within the range of certain body parts (particularly legs, torsos, heads, and faces), they just automatically open and clamp on. It's like she can't help herself.

Months of patient training have failed to correct this behaviour. We've tried telling her "No!" We've tried scruffing her. We've tried putting her alone in her cage for five minutes. We've tried applying bitter-tasting spray to our skin. But no matter what we do, she just comes right back for more biting.

So today I chanced upon someone on eBay selling Frettchenmaulkörbe, or ferret muzzles:

It looks like this would be pretty easy for a dedicated ferret to remove, but maybe it's possible to fit it snugly enough that it can't be slipped off. Maybe we can get her to wear this and let her get used to harmlessly poking her nose and mouth around our faces. I hope it works, because I don't know what else to try.

At the end of January Nadya and I went to Berlin to submit my passport application. While we were there we did some sightseeing. Nadya took most of the pictures, but following are the ones I took:

Museumsinsel

The Berliner Dom, Ishtar Gate, and Market Gate of Miletus

Sachsenhausen Concentration Camp

A Nazi concentration camp for political prisoners; later run by the NKVD

Tierpark

The zoo in East Berlin; unfortunately my camera battery died early on

Treptower_Park

Enormous Soviet war memorial

Miscellaneous

The New Synagogue, Hohenschönhausen prison, Hackesche Höfe, DPRK Embassy, and more

As previously mentioned I've been taking a German language course, Hören – Verstehen – Diskutieren II, at the university here. I had my Klausur (final exam) a few weeks ago and got my final grade today. (Follow the image link to see it at full size.)

That's right, I got a "1,3".

Are you as confused as me? If so, that's because German grades make absolutely no sense.

You'd think that a country which prides itself on early adoption and adherence to the metric system could come up with something just a little less confusing. Like maybe, say, making all grades integers from 0 to 100, where 0 is the worst and 100 is the best. That's how grades were assigned at my high school, and also at my undergraduate university, and also at my graduate university. (At some of the institutions, certain grade ranges were mapped to letter grades A, B, C, etc., though the raw numbers were always given as well so that you could compute an average grade.)

But no… At German universities, they have some weird, decidedly non-orthogonal system where the grades range from 5,0 (the worst) to 1,0 (the best). This seems backwards, but simple enough, right? Unfortunately no: You can't just pick any fraction between 5 and 1 and call it a grade. Between 5,0 and 4,0 there are no subdivisions, but between 4,0 and 1,0 you get steps which alternate in magnitude between three tenths and two fifths! So the complete series goes 5,0, 4,0, 3,7, 3,3, 3,0, 2,7, 2,3, 2,0, 1,7, 1,3, 1,0. So really what we have is some strange eleven-point scale which doesn't particularly lend itself to computing averages, assigning bonus marks, etc.

Amazingly, the grading system is probably the least confusing thing about the educational system here. In my job I'm regularly called upon to translate job postings from German, and if there's one thing I've learned it's that German educational qualification terminology is vast, intricate, and has no appropriate or agreed-upon mapping to English terminology. Part of the reason for this is that the educational systems in German-speaking countries are very different from those found in English-speaking countries; another reason is that governments and translators haven't sat down with each other and agreed upon a consistent translation for the various terms. Ask four translators to translate a job application from German into English and you'll get five different mutually (and often also internally) inconsistent translations.

A couple weeks ago I got a German census form in the mail which asked all sorts of detailed questions about my education. Many of them were multiple choice questions about the type of secondary and post-secondary institutions I attended and the qualifications I received there; there were literally dozens of answers to choose from. Since I was not educated here, it was extremely difficult to decide on the most appropriate answer. For instance, I had to choose between choices like Sonderschule, Hauptschule, Realschule, Oberschule, Gymnasium, Gesamtschule, Berufsfachschule, Berufsaufbauschule, Fachoberschule, berufliches Gymnasium, and so on. There was an annotation on the question which rather unhelpfully explained that people who were educated abroad should simply chose the "most appropriate" German equivalent, as if we foreigners are supposed to be familiar with the subtleties of not only the current German system but also those previously in use here and in the former German Democratic Republic.

I just received the following e-mail from one Eugen Fischer, who says that three years ago he wrote up a statistical study on the likelihood of finding a suitable romantic partner. He did this ignorant of my 1999 exposition on same (and its 2003 German translation), but as he recently learned of its existence he has supplemented his thesis with an appendix comparing his work to mine (to put it charitably). :)

Nun, vor 3 Jahren habe ich beschlossen auszurechnen, wie viele Frauen in meiner Umgebung wohnen, die für mich passend sind. Von da an habe ich an einer Ausarbeitung mit dem Namen "Das Vielmädchenproblem" gearbeitet, es beinhaltet alles was du jemals wissen wolltest. Vor einiger Zeit habe ich herausgefunden, dass schon jemand das ausgerechnet hat, und das warst du. Ich kannte deine Arbeit nicht. Ich habe mir sie angesehen und beschlossen die Fehler zu zeigen die du gemacht hast. Im Anhang dieser Mail ist unsere Arbeit und auf Seite 50–54 zeigen wir deine Fehler. Auf Seite 53 ist auch ein Vergleich zwischen deiner Arbeit und unserer, da kannst du auch sehen was wir alles berücksichtigt haben. Ich schreibe dir weil ich denke, dass es dich interessieren würde. Sie ist auch auf meiner Homepage www.super-physik.de zu finden

Fischer's thesis is entitled Das Vielmädchenproblem and is available for download on his website. Here is the abstract:

In dieser Arbeit werden Gleichungen hergeleitet, mit denen man die Anzahl der Mädchen / Jungs bestimmen kann, die zu einem passen. Es wird auch anhand eines Beispiels gezeigt, wie man die Gleichungen anwendet und wie man für sich selbst die Werte bestimmen kann – es lässt sich auch bestimmen wie das andere Geschlecht auf dich steht, also jedes wievielte Mädchen / Junge dich attraktiv findet. Es wird geklärt auf was die Mädchen wirklich stehen: Geld, Aussehen, Intelligenz oder doch was anderes und was davon wichtiger ist. Wir klären auch, wer wie viele Sexualpartner hat und wer öfter Fremdgeht, Männer oder Frauen und was das ganze mit der Intelligenz zu tun hat. Zudem wird die optimale Taktik bei der Partnerwahl erläutert und die Onlinedating-Seiten diskutiert. Es gibt noch viele andere spannende Erkenntnisse, welche man hier nicht alle aufzählen kann. In dieser Ausarbeitung wird einfache Schulmathematik verwendet und ist damit für alle verständlich. Viel Spaß.

Rakott krumpli is one of my favourite dishes, and it's so easy to make: just slice some boiled potatoes, smoked paprika sausage, and hard-boiled eggs, put them in alternating layers in an oven-safe bowl or casserole dish, top with sour cream, and bake in a 200°C oven for about 35 minutes.

Nadya makes delicious Zwiebelkuchen, a kind of German onion pie. She uses the recipe from Bavarian Kitchen but adds lots of extra cumin for an interesting Mexican note. It's normally got diced bacon, but we recently had some vegetarian folks over for dinner and found that substituting sun-dried tomatoes works well.