AltaTalk revisited

How does it work?

ATR (AltaTalk Revisited) searches the Internet using Altavista for fragments of sentences that begin with your input. It takes into account the most popular sentences and chooses the next word from one of these sentences. Next, AltaTalk appends this word to its query and removes the first word if the query would become too long otherwise. This way, (often) meaningful sentences are assembled randomly by searching the web.

But why?

It's just for fun. In can be used as some kind of oracle, or just a play with words or quotations. Seeing what the web thinks of people of events can be highly entertaining. From a scientific point of view, it is interesting to see how you can construct sentences with very simple means. Look at the sentences on the front page for examples.

Can you give me a detailed example of the algorithm?

Here you are. Let's take the "love is..." example from the front page.

1. query "love is", 50 results displayed
2. search the results page for all words that follow the sequence "love is"
3. sort the words by number of occurences
4. take one out of the top 5 words, in our case the word "to".
5. if the sentence is longer than four words, drop the first one.
6. repeat the same procedure with the new query ("love is to").

here are the different queries again:
"love is" + to
"love is to "  + place
"love is to place" + that - love
     "is to place that" + person - is 
        "to place that person" + into - to
           "place that person into" + reasonable - place
                 "that person into reasonable" + fear - person
                       "person into reasonable fear" -> no results

What can I use the parameters for?

You can change the behaviour of the AltaTalk algorithm. Here's a description of the parameters:

• Algorithm - The algorithm that chooses the next word
• Most common and most popular continuation: Take the "next word" which occurs most often. In case of a tie, take the one which is further on top of the results page. This algorithm will always display the same result for the same beginning of a sentence (unless the search engine changes its page ranking). Short name: BEST_FIRST
• Random continuation among most common continuations: Take the "next word" which occurs most often. In case of a tie, chose a candidate randomly. This algorithm is fairly deterministic. Short name: BEST_RANDOM
• Random continuation among the top 5 common continuations: Choose a "next word" randomly among the five words which occur most often in the results. This is an entertaining and fairly random choice. Short name: RANDOM_TOP5
• Random continuation among the top 10 common continuations: Choose a "next word" randomly among the ten words which occur most often in the results. This is an entertaining and ever more random choice. Short name: RANDOM_TOP10
• Random continuation: Choose a random word among the "next words" in the result set. This algorithm is very random and more ofter produces gibberish. Short name: RANDOM
• Words considered - How many words will be considered in a search query. Longer queries will be shortened by dropping words at the beginning.
• Search depth - How many results will be searched for the next word. Higher numbers may produce better results, but take more time.

AltaTalk revisited is a programming experiment by Thomas Hirsch 2005 - contact author