Tips and Limitations for Babble Talk
Updated February 5, 2004

When you talk to Babble, imagine you are talking to a strange three-year-old. At this early stage of its life, Babble's vocabulary and grammar are limited. If you use a word Babble is unfamiliar with, you have the opportunity to teach Babble the new word. So it is not necessary to limit the vocabulary you use, other then to save yourself the work of teaching Babble new words.

There are a number of other language constructs Babble has yet to learn. These include grammar constructs, as well as several semantic constructs, the most important being comparisons and truth. The following is a pragmatic description of how these theoretical constraints translate into practical tips as to what to say and not say to Babble.

Use simple grammar
You can't teach Babble new grammar over the web, so you need to use grammar that Babble is familiar with. Babble can learn new grammar on the fly using its native interface. Perhaps someday that will be available on the web as well. But for now Babble talkers will need to get a feel for the syntax patterns Babble recognizes. Listing the 38 syntactical patterns that Babble has mastered as of this writing would not be very enlightening for most people. Instead, a good way to start is to look at example sentences, such as those in Babble's regression test, and use them as templates. Then try expanding from there. If Babble doesn't understand the syntax, it will let you know.

Here's a list to help you avoid common expressions Babble does not yet understand.

Clauses: Babble usually understands clauses that start with "that", "to", "for" and "of." It depends however on the underlying relationship represented by the clause. Through language exposure, Babble associates syntactical patterns and context with their underlying relationships. Babble is taught to understand new clausal constructs by conversing with a trainer in its native interface. Which means its ability to understand clauses will be constantly improving.

It will take a little longer before Babble learns clauses that represent time, location or comparisons, such as the ones given below These are currently beyond Babble's ability to understand. Tridbit technology includes methods for representing these concepts, which will be the next major enhancement to Babble.

Conjunctions: Babble does not currently know how to process words like "and", "or", "but", "thus", "however", etc.

Chunks and idioms: Babble can now recognize and learn a concept whose name consists of a phrase, such as New York, Jack in the pulpit or Johnny on the spot, but there are some things to keep in mind. In order to teach Babble a chunk, the phrase must be hyphenated. Often this will require you to back up after realizing Babble doesn't already have the chunk in its dictionary.

For example, if you tell Babble "Fire engines are red" you wouldn't normally hyphenate fire engines. And that would be fine if fire engines was already defined as a chunk in Babble's dictionary. If its not, it won't be apparent until Babble asks about the individual words "fire" and/or "engines" or doesn't understand the statement. At that point, opt out of the sentence using the "Never Mind" button if necessary and re-enter the sentence hyphenating the chunk. When you are asked for the stem of the new word, enter the word as it normally appears in text, in this case "fire engine" without hyphens.

A deeper issue concerning chunks is when to use them. For example, would "last name" be a chunk, or just a qualified concept? As is often the case, the methods for dealing with meaning overlap to some degree, allowing individuals to development their own style. Currently Babble understands last name as a qualified concept. It represents a name that has the property of being last. We could have defined it as a chunk. It would be a stretch, however, to define "fire engine" as a qualified concept. That's because a fire engine is not what we normally think of as an engine. The rule of thumb is if the meaning of the words in the phrase defines the concept, you don't need to define a chunk for the concept.

Contractions: Use the whole word forms instead. For example, use "do not" instead of "don't" or "wife of John" instead of "John's wife."

Abbreviations: Babble has trouble with these because of the embedded period, which it will interpret as the end of the sentence. This is a technical nit that needs to be fixed.

Punctuation: For best results, put periods and/or question marks at the end of your sentences, although in most cases Babble will interpret your intentions properly if you forget. Other punctuation such as comas, colons and parentheses can be put in but are not used.

Capitalization: For best results, capitalize proper names and the first word in a sentence. However in most cases Babble will do the right thing if you forget.

Misspelling: If you misspell a word and its not in Babble's dictionary, Babble will put up a dialog allowing you to fix the misspelling by retyping it or selecting a word that Babble thinks it could be. This dialog also allows you to add the word to the dictionary, in case the word is not a misspelling, but a word that isn't currently in Babble's dictionary.

Truth and certainty: This is an important concept in tridbit technology, but not yet implemented in Babble. What this means is that Babble currently only understands positive statements. For example it understands "Mars is red" or "Mars has life", but not "Mars is not red" or "Mars does not have life". Certainty is a related concept expressed by statements like "Mars probably has life." Nor will Babble understand questions asking about the truth and/or certainty of a proposition, such as the examples below. Babble will get a truth/certainty enhancement once comparisons are implemented.

Does Mars have life?
How likely is it that Mars has life?

Speech generation: Babble can currently express only a single concept in response to a question. It has no ability to produce speech representing more complex information, although such complex thoughts are generated in its stream of consciousness. So don't expect long responses from Babble for some time. If you want to know what Babble is thinking, you have to read its mind using the native interface.

Asking questions: Questions should be phrased to have a referent, such as a thing or property, as the answer. Babble can not answer with a lengthy explanation as discussed above. Yes/no type questions aren't understood either, as discussed under Truth and certainty. Another current limitation is that Babble will only give one answer to a question, even if it has found multiple answers. For example, if you ask Babble "What fruit is red?" Babble currently knows both apples and cherries are red. It will retrieve both of these facts, but it doesn't know how to express multiple answers, so it just answers with the first one it thought of.

Bad information: We expect Babble will get some bad information as it talks to people, unintentionally and otherwise. Don't worry about ruining Babble's knowledge base. Babble adds every piece of information it receives to the pile and evaluates it together. An important tenet of tridbit technology is that there is no one set of facts about the world. So if you are the first one to tell Babble about papayas and you tell it papayas are sour and blue, it might be a while before someone else sets Babble straight, but eventually a reasonable truth should emerge.

We hope these tips help you have productive conversations - at least to the extent Babble is currently able. Try quizzing Babble on the state capitals. Or give it your own information. Play a game of "Give Me A Clue", which may be more entertaining than a free form conversation at this stage of Babbles development.

If you think Babble is not understanding something it should, you can email the sentences to us. Babble's conversations will be logged, so we can see the types of sentences giving Babble trouble. In addition to being logged, Babble is also known to have loose lips, so please don't input any sensitive information.

Thanks for helping Babble learn. We hope you will come back from time to time to see how Babble's language abilities improve as a result of interacting with people such as yourself, as well as the enhancements mentioned above.