NAME

RiveScript::Tutorial - A beginner's guide to creating their first RiveScript brain.

INTRODUCTION

This tutorial outlines the various capabilities of the RiveScript specification and offers some recommended pointers for creating a well-formed RiveScript brain. What you do with this knowledge is up to you; be creative!

Be sure to skim over the RiveScript manpage first, because this tutorial jumps right in to using the various RiveScript commands without always explaining what each of them do.

A Simple RiveScript Interpreter

Here is a simple Perl script for running a RiveScript interpreter. This assumes that the brain's RS files will be stored in a directory called "tutorial", local to the Perl script. You'd want to edit certain parameters in this code if you see fit.

  #!/usr/bin/perl -w

  use strict;
  use warnings;
  use RiveScript;

  # Create the RiveScript interpreter.
  my $rive = new RiveScript();

  # Load the RS tutorial brain.
  $rive->loadDirectory ("./tutorial");

  # Sort them.
  $rive->sortReplies;

  # Go into a chatting loop.
  while (1) {
    print "User> ";
    my $msg = <STDIN>;
    chomp $msg;

    # Grab a reply.
    my @reply = $rive->reply ('user',$msg);
    print " Bot> $_\n" foreach(@reply);
  }

GETTING STARTED

Structure of a Good Brain

By "A Good Brain" I mean a brain written to certain standards. Much like how an HTML file can be written any way you want, but the W3C sets up standards for how to write them.

A good RiveScript brain has at least files named config.rs and begin.rs. "config.rs" would be for configuration of the bot, setting up variables, substitutions, etc. This file's name isn't absolutely required to be "config.rs" but that's a good name as far as following standards goes.

The "begin.rs" file, on the other hand, is more of a required name. In the loadDirectory method, a file named "begin.rs" is loaded in first, before any other file. In this way, it has two purposes:

  1. To load in any libraries or packages, since they'll be then
     loaded in first and any brain-specified objects will override
     the library's ones.
  2. To hold the BEGIN Statement.

We'll get into both of these later.

The config.rs File

To begin the creation of our RiveScript brain, save the above Perl code into a file (if you haven't already done so), and then create a folder named "tutorial" in the directory you saved the Perl file.

Open a text editor (such as Notepad). RiveScript documents are just text files so any editor should do just fine (provided it doesn't assume you're creating a non-plaintext file and adds in strange characters).

First, we'll set up some global bot variables to give our bot a name and some basic information.

  ! var name     = RiveScript
  ! var lastname = Tutorial
  ! var gender   = machine
  ! var age      = 0

You can follow that pattern to make any additional variables to give your bot a "personality."

Next we'll create an array or two. Arrays can be used later in triggers to make a specified set of matchable words. We'll create two arrays, one for color names and one for the verb "to be"

  ! array colors = red blue green yellow cyan fuchsia orange
  ^ black white silver gold gray grey pink
  ! array be = am is are was were

Then we'll set up some substitutions. These are statements to substitute things in the user's message with something else. This set of code should get you started, but you can add substitutions for anything else you can think of, too.

  ! sub &quote;   = quot
  ! sub '    = apos
  ! sub &     = amp
  ! sub <      = lt
  ! sub >      = gt
  ! sub +         = plus
  ! sub -         = minus
  ! sub /         = divided
  ! sub *         = times
  ! sub i'm       = i am
  ! sub i'd       = i would
  ! sub i've      = i have
  ! sub i'll      = i will
  ! sub don't     = do not
  ! sub isn't     = is not
  ! sub you'd     = you would
  ! sub you're    = you are
  ! sub you've    = you have
  ! sub you'll    = you will
  ! sub he'd      = he would
  ! sub he's      = he is
  ! sub he'll     = he will
  ! sub she'd     = she would
  ! sub she's     = she is
  ! sub she'll    = she will
  ! sub they'd    = they would
  ! sub they're   = they are
  ! sub they've   = they have
  ! sub they'll   = they will
  ! sub we'd      = we would
  ! sub we're     = we are
  ! sub we've     = we have
  ! sub we'll     = we will
  ! sub whats     = what is
  ! sub what's    = what is
  ! sub what're   = what are
  ! sub what've   = what have
  ! sub what'll   = what will
  ! sub can't     = can not
  ! sub whos      = who is
  ! sub who's     = who is
  ! sub who'd     = who would
  ! sub who'll    = who will
  ! sub don't     = do not
  ! sub didn't    = did not
  ! sub it's      = it is
  ! sub could've  = could have
  ! sub should've = should have
  ! sub would've  = would have
  ! sub when's    = when is
  ! sub when're   = when are
  ! sub when'd    = when did
  ! sub u         = you
  ! sub ur        = your
  ! sub r         = are
  ! sub im        = i am
  ! sub wat       = what
  ! sub wats      = what is

Finally, we'll define some "person" substitutions. These are special substitutions used for the bot's reply. Anything between the tags {person} and {/person} in the bot's reply uses these substitutions. The purpose of this is to swap first- and second-person pronouns.

  ! person you = me
  ! person me  = you
  ! person am  = are
  ! person are = am

And now save this file as "config.rs" (you may need to put quotes around the filename in Windows if you haven't already associated a program with that extension!)

The begin.rs File

The second file recommended for a good brain is call "begin.rs" and here is where we'll load in libraries, and set up our "begin" statement.

So first, we'll load in a library that comes with the RiveScript distribution: English. It gives us a lot of arrays of various English verbs, nouns, and adjectives, along with their other varied forms.

  ! include English.rsl

And now for the BEGIN statement. The BEGIN statement is a block of replies that uses the > label for "begin". By default, this block takes only one preset trigger from the user: "request"

When the reply to "request" contains the {ok} tag in it, then a normal reply for the user's message is obtained and substituted back into the "request" reply, where the {ok} tag is.

  > begin
    + request
    - {ok}
  < begin

This is the simplest form for a BEGIN statement. You can use conditionals and other things here too, which we'll get into later in the tutorial. But for now, the user requests a reply, and the bot returns one.

CREATING REPLY FILES

While config.rs and begin.rs are RiveScript documents, there's not much for it being able to give an intelligent response to your message.

Add Some "Hello" Replies

A Basic Trigger

We'll start in a logical place: hello's and goodbye's. After all, hello's are the first things exchanged when you talk with somebody. Create a new RiveScript document (new text document) and add a simple reply like so:

  + hello
  - Hello there, human.

Now we have something to test. Save this as "greetings.rs" in the "tutorial" folder with your other RiveScript documents, then run the Perl code you saved earlier to test the bot. At the "User>" prompt, type "hello" and the bot should reply "Hello there, human."

Random Responses

Now we'll make this trigger return a random reply. Currently, no matter how many times you say "hello", the bot's reply will never change. This can be fixed by adding more than one response to the trigger:

  + hello
  - Hello there, human.
  - Hey, how are you?
  - Hi there.
  - Hello. :)

Save and test the bot again. Now, say "hello" to it a few times and it should randomly choose one of those four responses to give you.

Weighted Random Responses

Now, let's put some weights on certain responses. For instance, the "Hello there, human." seems quite cold and heartless--a good response for an impulse-driven machine. But for the sake of emotional security, the bot should use that reply a lot less than the other ones.

Using the {weight} tag on the responses will change the probability of them being chosen. So here, we'll add weights to all the triggers except for the one calling you a human:

  + hello
  - Hello there, human.
  - Hey, how are you?{weight=3}
  - Hi there.{weight=2}
  - Hello. :){weight=4}

Note that if a {weight} tag isn't specified, the trigger assumes a weight of 1. Now, there is a total of 10 weights in these replies: 3 + 2 + 4, and then +1 for the "Hello there, human." response.

So, "Hey, how are you?" has a 3 in 10 chance of being chosen, "Hello. :)" has a 4 in 10 chance, "Hi there." has a 2 in 10 chance, and finally "Hello there, human." has a 1 in 10 chance of being chosen.

Try it! Save and test the bot again and say "hello" quite a few times. Compared to the other responses it gives you, it should say "Hello there, human." considerably less often.

Redirected Triggers

Now that we're fairly comfortable with this response, we'll add a couple other triggers that will point back to this one. So the user can say "hi" or "hey" and still get the same effect as if he had said "hello." We could, of course, just make substitutions to change "hey" into "hello" but that would ruin the sake of learning how to code RiveScript files. ;)

  + hey
  @ hello

  + hi
  @ hello

The "@" command redirects the user to another trigger: "hello". In this case, saying "hey" or "hi" will return a reply as if the user had just typed "hello" instead. Save and test this.

You can follow this pattern to add some more triggers for goodbyes now, too.

Wildcards, Variables, and String Modifiers

Using Wildcards

Let's add a few more triggers. These ones will be open-ended, allowing the user to say something to "fill in a blank," and what they say can be matched using the <star> tags.

Create a new RiveScript document and save it as "users.rs" -- this file will contain triggers specifically geared for learning and repeating user details.

For our first example of using a wildcard, we'll allow the user to tell us their name.

  + my name is *
  - Nice to meet you, <star>!

Save and test it. Say "my name is Bob" or "my name is Steve" and see how the bot replies. Before moving on, we'll put in another trigger to test the use of multiple wildcards:

  + * would describe me as *
  - So do you always get described as <star2> by <star1>?

When you're dealing with more than one wildcard, you'll need to include a number in the <star> tag. The stars are matched from left to right. In this example, if you say "John would describe me as eccentric", <star1> would be "John" and <star2> would be "eccentric"

Saving User Variables

The <set> tag can be used to set a user variable. User variables are variables specific to the current user. For example, your bot can learn a user's name and address that user by name from then on, while keeping a different name for a different user. You can also think of these as "local variables," local to each user.

So, when the user tells us their name, we should remember it for them:

  + my name is *
  - <set name=<star>>Nice to meet you, <star>.

Here's another example for clarification:

  + i am * years old
  - <set age=<star>>A lot of people are <star> years old--you're not alone.

Numeric Variable Modifiers

There are four tags, <add>, <sub>, <mult>, and <div>, for adding, subtracting, multiplying and dividing numeric variables, respectively.

  + today is my birthday
  - <add age=1>Happy birthday, <get name>!

If the variable being modified hasn't been defined yet, it is set to 0 before being manipulated. If you manipulate a non-number variable, or add a non-numeric value to a variable, you'll get an error in place of the tag.

Retrieving User Variables

The <get> tag is for retrieving a user variable and putting it into the bot's response.

  + what is my name
  - You told me your name was <get name>.

Bot Variables

While we're on the topic of variables, the bot variables we declared back when we created the "config.rs" file can be retrieved in <bot> tags.

The <bot> tag is also the only tag that can go inside of the trigger part. So while we're on the "users.rs" file, we'll add a trigger for if the user has the same name as the bot.

  + my name is <bot name>
  - <set name=<bot name>>Is it really? Don't tell me your last\s
  ^ name is <bot lastname> too. ;)

Also note that this example used a carat to extend the reply. When you do this, you'll usually want to include a \s at the end of the previous line, to put a space between the two parts of the reply (otherwise it would say "Don't tell me your lastname is Tutorial too.", without a space between "last" and "name")

Environment Variables

RiveScript environment variables can be retrieved in the <bot> tag. The variable names all begin with "ENV_" and are all uppercase.

RiveScript's native variables are as follows:

  ENV_OS         = The operating system running the interpreter
  ENV_APPVERSION = The version of RiveScript being used.
  ENV_APPNAME    = The full string of RiveScript version (i.e. "RiveScript/1.0")

To access Perl environment variables, append "ENV_SYS_" to whatever environment variable you want to use. For instance, when used in online CGI context, the following reply would be valid:

  + what is my ip address
  - Your IP is <bot ENV_SYS_REMOTE_ADDR>.

String Modifiers

By now you may have noticed that the bot stores your name in lowercase. This is because, regardless of how you format your message while typing, RiveScript removes punctuation and lowercases what you say.

In most cases, for variable storage uses, this is alright. But for formal things such as a name, you want them to be formatted to start with capital letters.

To start this section, we'll go back to the my name is * trigger added under E<"Saving User Variables">.

To formalize the name, use the {formal} tag:

  + my name is *
  - <set name={formal}<star>{/formal}>Nice to meet you, <star>.

Now the name is stored with capitilizations where needed (if you ask the bot your name, as we did in the last step, it will say it correctly). So we have one of two options:

  + my name is *
  - <set name={formal}<star>{/formal}>Nice to meet you, {formal}<star>{/formal}.

This one will formalize <star> in both instances. It's not the greatest to look at.

A feature of RiveScript is that <set> tags are run before <get> tags are. So to condense that last example:

  + my name is *
  - <set name={formal}<star>{/formal}>Nice to meet you, <get name>.

That looks a little nicer.

While we're on the topic of string modifiers, the other ones are {uppercase} and {lowercase}, which will make the text uppercase or lowercase, and then {sentence} is to make it sentence-cased (where the first word is formal but the rest is lowercase).

Catch-All Trigger

By now in your testing, you may notice that if you say something that isn't matched by a trigger, you always get an "ERR: No Reply" back. This can be done by creating a trigger that consists of nothing but a wildcard symbol.

Internally, replies are sorted with triggers containing the most whole words first, all the way to the triggers containing no whole words. So in this way, a trigger of just * is checked last.

Create a new reply file called "star.rs" with this trigger inside:

  + *
  - I'm not sure how to reply to that one.

Save and test. Say something that the bot doesn't already know how to reply to.

You can be creative with the user of this trigger too: using random replies, weighting the replies, etc. -- treat it like a normal trigger.

Alternations, Arrays, and Optionals

Note that alternations and optionals are matched in the <star> tags. If you combine them with wildcards, make a note of this.

Alternations in the Trigger

Alternations can be used in the trigger to add a little bit of "lee-way" among the user's choice of words, but not as much leeway as you'd get with a wildcard.

Create a new file named "bot.rs" which will be used to keep triggers relating to the bot itself (users the question would ask about the bot). Now for a quick example of using alternations, add a trigger like this:

  + what (s|is) your (work|home|cell) phone number
  - You can call me at 555-5555.

Save and test it. This will match every message from "what s your work phone number" to "what is your cell phone number", but it won't be matched if the user doesn't choose any of the words provided.

Optionals in the Trigger

An optional word is one that can be matched in a trigger, but doesn't have to be. We'll edit the reply we just created in the last section to read like this:

  + what (s|is) your [work|home|cell] phone number
  - You can call me at 555-5555.

Now you can ask "what is your phone number" without specifying which number you wanted. The "work OR home OR cell" words don't have to exist in the user's message at all. What purpose does this serve? You can take care of possible useless variations of a trigger.

  + how are you [today]
  - I am great, how are you?

So that would match "how are you" OR "how are you today", and you don't need to use multiple triggers to handle it.

Arrays in the Trigger

When we created the "config.rs" file, we declared some arrays. One of these was an array named "colors" which was full of color names.

Open the "users.rs" file we created before (if necessary). We'll be adding a trigger that allows the user to tell us their favorite color, but ONLY if they chose a valid color name from our array.

  + my favorite color is (@colors)
  - <set color=<star>>I think <star> is a nice color too.

When the array is inside of parenthesis, it is matchable with a <star> tag. If you omit the parenthesis, it is not matchable. Here's a quick example to test that:

  + do you like the color @colors
  - Yeah, <star> is a nice color.

In this example, the color you say should not appear in the bot's reply.

Creating Complex Triggers

Optionals, alternations, and arrays can all be combined into triggers to condense a lot of triggers down to just one or two.

In this example, the user can ask anything from "what is the time" to "what time is it"

  + [what] [@be] [the] time [@be] [it]
  - I don't know what time it is, go find a clock.

RiveScript is capable of actually giving the time, through an object. We'll get into this later in the tutorial.

Reply Conditionals

"Equals" Condition

Conditionals can be used in responses to check for values (or lack thereof) of user OR bot variables. In this example, the bot will give a specific reply if the user is named "Bob" and a different one otherwise.

In your "users.rs" add this trigger:

  + is my name bob
  * name = Bob => Yes, that's your name.
  - No, you are <get name>, silly!

Multiple conditions can be used to one trigger as well:

  + is my name bob or tom
  * name = Bob => You're Bob.
  * name = Tom => You're Tom.
  - You're neither Bob nor Tom.

To check the value of a bot variable in a condition, put a pound sign in front of the variable name.

  + is your name rivescript
  * #name = RiveScript => Yes, that's my name.
  - No, I am <bot name>.

Numeric Comparisons

In place of the "equals" sign in the conditions above, you can use numeric operators to check values of numeric variables.

  + am i a juvenile
  * age <  18 => Yes, you're under 18.
  * age >= 18 => No, you're at least 18 years old.
  - I don't know; I don't think I know how old you are.

Conditionals are checked in order of appearance. When one returns false, the next one down the line is checked, and so-on. If all conditionals are false, a normal reply should be provided. Random replies can be used here too.

Note: this reply won't work unless you have another reply for setting a value for the variable "points"

  + do i have enough points
  * points >= 500 => Yes, you have at least 500 points.
  - No, why do you want to know?
  - No, go earn some.
  - Why do you care so much about points?

"Defined" Condition

Sometimes, rather than checking a value of a variable, you just want to know if the variable has been defined. In this case, just use a "?" as the check sign.

  + do you know my name
  * name ? => Yes, your name is <get name>.
  - No, what is your name?

Keeping it in Context

You can use the %PREVIOUS command in a RiveScript file to continue a thought about the last thing the bot said. Open "greetings.rs" and add a trigger for this:

  + how are you
  - Good, how are you?

Now to use the %PREVIOUS command, take the bot's last reply, lowercase and without punctuation, and use it like this:

  + good
  % good how are you
  - That's great.

So if the user says "how are you" and then says "good", the bot will say "That's great." but if the user says something else, the bot will answer that like normal too.

Person Substitutions

The point of person substitutions is to replace first- and second-person pronouns in the bot's reply. For instance:

  + do you think i *
  - I don't think you {person}<star>{/person}.

  "Do you think I hate you?"
  "I don't think you hate me."

In that case, it substitutes "you" for "me" so that its reply makes more sense. Otherwise it would've said "I don't think you hate you" and not make any sense.

Topics

Topics are sets of responses. While a user is in a topic, he can only match the triggers specified under that topic. He can not match triggers under a different topic.

The default topic is "random"

To set a topic, use the {topic} tag. To declare a topic, use the LABEL commands.

Here's an example of a topic set to handle abusive users:

  + i hate you
  - I don't like you either. Not apologize or I won't talk with you.{topic=sorry}

  > topic sorry
    + *
    - Say that you're sorry.
    - Apologize for being so mean to me.

    + sorry
    - Okay--I'll forgive you.{topic=random}
  < topic

RiveScript Objects

Objects are bits of Perl code which can be called within a reply. To declare an object, use the LABEL commands to name the object, and then provide some Perl code.

Create a new RiveScript document named "objects.rs" and we'll start with a simple localtime-returning object:

  > object localtime
    my ($method,$data) = @_;
    return localtime();
  < object

Now we have an object named "localtime" which returns Perl's localtime() value. Now how to use it?

Refer back to the trigger above that looks like this:

  + [what] [@be] [the] time [@be] [it]
  - I don't know what time it is, go find a clock.

Now, instead of the bot telling you it can't get the time, we'll make it so it can get the time!

  + [what] [@be] [the] time [@be] [it]
  - The time is currently &localtime()

Test it.

So what are $method and $data that the object received? You can micro-manage your objects by creating methods for them. Here's an example of a (simple?) text-encoder, which will encode things in either Base64 or MD5_Hex.

  > object encode
    my ($method,$data) = @_;

    use MIME::Base64 qw(encode_base64);
    use Digest::MD5 qw(md5_hex);

    if ($method eq 'base64') {
      return encode_base64($data);
    }
    elsif ($method eq 'md5') {
      return md5_hex($data);
    }
  < object

And now we can test that with these triggers:

  + base64 encode *
  - "<star>" encoded in Base64: &encode.base64(<star>)

  + md5 encode *
  - "<star>" encoded in MD5: &encode.md5(<star>)

So, $method is the word after the "." and $data is the info passed inside the parenthesis. Note that $data is always a scalar. So if you use commas to separate multiple arguments, your object will have to split the scalar into an array.

Reply Tag Reference

Various tags have been shown in examples here, but there are many other tags available in RiveScript responses. Here is a listing of all of them and how to use them:

<star>

Inserts the captured value of wildcards or alternations/optionals or arrays. When dealing with multiple stars, use <star1>...<starN>

<input1>-<input9>

Inserts the last 1 to 9 things the user said, or "undefined" if the user hasn't sent enough messages to match whatever input history you're trying to show.

  + what did i just say
  - You just said "<input1>"

<reply1>-<reply9>

Inserts the last 1 to 9 things the bot replied with.

  + why did you say that
  - I said "<reply1>" because you said "<input1>"

<id>

Inserts the user's ID (the one passed in with the Perl code from reply())

<bot>

Insert a bot variable or an environment variable.

  + who are you
  - I am <bot name>, a bot created by <bot owner>.

<get>, <set>

Get and set a user's variable.

<add>, <sub>, <mult>, <div>

Add, subtract, multiply, or divide numeric user variables, respectively.

  + double my points
  - <mult points=2>I have doubled your points. You now have <get points>.

{topic=...}

Set the user's topic to something else.

{nextreply}

Breaks the bot's reply into two (or more) replies at this tag.

{weight=...}

Give responses a different probability of being chosen. The weight must be greater than or equal to 1.

{@...}

An inline redirection.

  + * or something
  - Or something. {@<star>}

"Are you a bot or something?" "Or something. How did you know I am a machine?"

{!...}

An inline definition.

  + set your name to joe
  - {! var name = Joe}I have set my name to Joe.

{random}...{/random}

Insert a small bit of random text within a single reply.

  + say something random
  - This {random}reply response output{/random} has a random {random}word noun{/random}.

To have multiple-word items, separate them with pipes instead of spaces.

  + give me a fortune cookie
  - Your fortune: {random}You will be rich.|You will be famous.{/random}

{person}...{/person}

The text between these tags will have Person Substitutions run on them.

{formal}...{/formal}

Makes Your Text Formal (Every Word Begins With A Capital)

{sentence}...{/sentence}

Makes only the first word of the message formal, the rest lowercase.

{uppercase}...{/uppercase}

FOR UPPERCASE TEXT.

{lowercase}...{/lowercase}

for very lowercased text.

{ok}

Used in the BEGIN statement to know it's okay to get a reply.

\s

Inserts a white space.

\n

Inserts a new line.

RIVESCRIPT LIBRARIES AND PACKAGES

Back in the beginning of this tutorial, we told RiveScript to use the library "English.rsl" to give us lots of useful arrays we might want to use in triggers. But what is a library?

Libraries and Packages are special RiveScript documents. Their purpose is to allow multiple brains to share common objects and array definitions, without having to copy and paste them into each brain's directory. Think of them like Perl modules. Any number of Perl scripts can use the same modules.

Library and Package Search Paths

Currently, RiveScript searches for included files in the following locations:

  1. Any of the values of Perl's @INC, with "/RiveScript/RSLIB" appended.
  2. The working directory of the Perl script running the RiveScript interpreter.

Since English.rsl comes with the RiveScript distribution, it can be found in your /usr/site/lib/RiveScript/RSLIB/ directory. This is the recommended place to put all your RiveScript libraries and packages.

If you would like to specify a new path for RiveScript to look for include files, use the !addpath directive:

  ! addpath C:/MyRsLibs

RiveScript Library

A RiveScript Library file is a special RiveScript document which consists of nothing but array declarations, substitutions, etc. (things you may declare in a config.rs file too).

The "English.rsl" library that comes with RiveScript includes "English/EnglishNouns.rsl", "English/EnglishAdj.rsl", and "English/EnglishVerbs.rsl" which define tons of arrays for English words and their other varied forms.

RiveScript Package

A RiveScript Package is a special RiveScript document which consists of one (or more) object declarations. For example, you might create a RiveScript package full of objects for performing web searches, and have a different object for each of Google, Yahoo Search, MSN Search, etc.

CREATIVITY

By now you should have a grasp on the RiveScript language. What you do with it is up to your own creativity. You can combine any of the elements you learned about to get some pretty neat effects.

SEE OTHER

Look at the RiveScript manpage. It would definitely help.

AUTHOR

Cerone J. Kirsle, kirsle --at-- rainbowboi.com

cpanm RiveScript

perl -MCPAN -e shell
install RiveScript