One of my readers has asked me why I do not want to debate with him. This is my reply:
The way to persuade people you’re right is by demonstrating that your ideas work on some task, not by arguing with them. The proof of your ideas is their performance. For example, there are benchmarks for parsing. If you have a great idea for a parser, build it and test it on the benchmarks, to show that your idea is better than the competition. If you don’t agree with the benchmarks, make your own benchmarks, and challenge the competition to try your benchmarks. Scientific theories rise and fall based on their empirical performance, not on how well their proponents argue. Falsification is empirical, not argumentative. You can argue for years and never get anywhere. But one experiment can cut the Gordian knot. To me, this is what science is all about.
In computational linguistics, experimentation is slowly taking over the field. This is a very positive trend. If you look at my blog posts, you will see that empiricism is a theme that underlies several of the posts. This is why I don’t want to engage you in an argument. In my opinion, it is a waste of time, for both of us. In my blog, I just want to share some ideas. If you can use them, great. If you can’t, then just ignore them.
If you have interesting experimental results and some moderate skill at writing, you can get your results published and cited, without following any “school”. I know this for a fact, because I have done it many times.
Argumentation is fine as long as the parties involved are learning something from the discussion. But when the discussion is no longer progressing, it’s time to do some experiments.
Filed under: Computational Linguistics, Philosophy of Science | Tagged: argumentation, debate, experimentation
I entirely agree. In fact, experimental results are more important than theory (that’s what I claim somewhere on my blog).
To many people, experimental work is “less work” than theoretical work. But I honestly think that the reverse is true. Theory without experiments to back it up is nothing.
However, “experiments” have a way to take many forms and I think mathematicians run experiments, they just do not call them experiments.
Theory and experiment take turns leading. Sometimes a theory makes a surprising prediction, which is later validated by experiment. Sometimes an experiment makes a surprising observation, which is later explained by a theory. An excellent book on this topic is Ian Hacking’s Representing and Intervening. What I’m trying to say is not that theories are inferior to experiments. Rather, when two theories disagree, we should turn to experiment as the arbitrator, not to debate.
Or, as my mother likes to say, “Why think when you can do the experiment?”