Medie, meet Hal

So, today I tried out Medie. I’d said I’d write about it as a third party Medline tool. I can’t. At least not as a hospital librarian. I was going to try to give it a pat on its back and insincerely flash it half a smile so it wouldn’t feel too badly about itself.

Instead, I’ll tell you what I really think of it. The first problem I noticed was that it doesn’t use MeSH (and therefore you can’t focus or use subheadings), so you’re losing massive precision right away. The other biggie I couldn’t get past was that it lists by PMID number, not title and author.

Sounds like I hate it, right? Not at all. Once I learned more about it and understood what it does, I was blown away.

As per David’s original post, it is a semantic search engine. The first time I learned of semantic search and natural language, I thought “interesting!” in the way that means “when’s lunch?”

This has changed my mind completely.

Let me take a step back for a moment. One of the first things I mention when I teach Medline is that you are using a computer. You can’t talk to it like you are asking me a question. You need to represent your concepts with words and terms and phrases and connect them appropriately. Then I launch into Boolean logic and MeSH.

Medie is a project of Tsujii Laboratory at the University of Tokyo that works on Natural Language Processing and Computational Linguistics. Basically, this means that it can root your word and process algorithms so that you search the database with natural language. If it doesn’t already, it will be “thinking” that when you type heart you may also like results with cardiac. That is too basic of an example, but hopefully you get the idea.

So, when you search Medie you enter search terms into basic parts of speech (subject, verb, object) and out pop your results, with the line or two of text it picked up color-coded by part of speech. It does have some additional search options that I can see as being very useful and helpful. But as I can’t recommend it for searching now, I’ll let you discover those on your own. I should note that I didn’t find any information on what text is searched or omitted, or how results are ranked.

But in the future I can see Hal saying: “I’m sorry, Dave I’m afraid I can do that.”

2 thoughts on “Medie, meet Hal

  1. I really like the idea behind MEDIE, but I have to admit that I don’t seem to be able to construct useful natural language searches (which sounds *completely* counterintuitive). Perhaps I’ve been so conditioned into combining different nouns as keywords in my searches that I can’t come up with workable verbs. Or my patrons are already conditioned to think of keyword nouns, so that’s what they ask for in the search, and they’re not suggesting good verbs. I tried running the last five searches I did in MEDLINE, and didn’t come up with much, and almost always had to go to the Advanced Search option to add keywords, which seemed like cheating. Any thoughts on how to add verbs to tease a keyword search backwards into being a natural language search to take advantage of MEDIE?

  2. Yeah, I know it was difficult to search (!), which is why I focused on how this step towards natural language is cool, if not quite there yet.

    I did notice that the subject or object/ verb construction used in the examples was odd (“induces” becomes ‘induce’, probably to stem?), which I think illustrates how complex our natural speech is (when is the last time you thought about subjects and objects? I’m pretty sure it was around third grade for me, I had to scratch my head for a minute).

    I think it was really useful to play with in terms of learning about natural language, but unless you’re helping someone who wants to see computational linguistics, I wouldn’t recommend using or recommending this as a way to search Medline. Hopefully that will change some day soon.