Wait...What'd I Just Say?

Several months ago, an acquaintance decided to try out a speech recognition software program to cure writer's block. He found it to be quite "intelligent" for a $50 piece of software - the calibration process was copious but effective, and the program supposedly "learns" based on corrected mistakes in order to eliminate future mistakes before they occur. By his account, the program was difficult to adjust to and absolutely mangled European names, but on the whole he was happy with it.

Speech recognition programs fall within the realm of natural language processing (NLP), which combines computer science, linguistics, and artificial intelligence principles with an eye toward reliable interaction between computers and natural human language. Like other AI fields, NLP has made great progress but still leaves a lot to be desired. Language is heavily rule-based but is also capable of great ambiguity, which of course leads to difficulties even in human-to-human communication. Throughout history, lexicographers and great thinkers alike have developed ambiguous example sentences to demonstrate the subtlety and craftiness which makes human language unique and baffling.

Ambiguity often depends on punctuation, as witnessed by recent memes demonstrating the importance of commas using the statements "Let's eat, grandma!" and "Let's eat grandma!" One of the earliest examples of this type of lexical ambiguity dates to 1327. Shortly after the murder of Edward II of England in that year, one of the king's prison officers received a note which read "Edwardum occidere nolite timere bonum est." This statement could mean one of two very different things depending on how it's punctuated. Place a semicolon after nolite and it means "Do not kill Edward; it is good to be afraid of doing this", while a semicolon after timere renders it "Do not be afraid to kill Edward; doing so is good." The meaning of this note has been widely discussed for centuries.

Homonyms - words which share identical spelling and (usually) pronunciation but have different meanings - also contribute significantly to ambiguity. This phrase was crafted by William Rapaport, a professor at the University of Buffalo, in 1972:

"Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo."

Through the use of homonyms, Rapaport designed this grammatical sentence to appear nonsensical. The sentence relies on three different forms of the word "buffalo": as a proper noun (the city in New York), a noun (the large ungulates also known as "bison"), and a verb ("to bully"). By interpreting these usages we can translate the sentence as "Buffalo bison [that other] Buffalo bison bully, [themselves] bully Buffalo bison."

The most common thread behind miscommunication is syntactic ambiguity, or the simple fact that identical sentences and words may carry different contextual meanings. A famous example which plays on the words "fly" and "like" is the sentence pair "Time flies like an arrow. Fruit flies like a banana."

Other instances include:

"We saw her duck."

"Flying planes can be dangerous."

"He saw that gas can explode."

Logical statements can be made even more confusing by combining multiple ambiguities. For example, the statement "The old man the boats" appears to be nonsensical because of the common qualified noun "old man." By playing into our immediate assumption that "old" qualifies "man," this sentence leads us to believe that it's gibberish, but by rereading it and considering "old" as a noun and "man" as a verb we find that it's grammatical. This sentence in particular serves as a kind of verbal optical illusion: by blocking out "the boats" we see the noun "old man," but considering the whole sentence we interpret something completely different.

NLP will probably follow the same old AI destiny: it can make inanimate objects appear intelligent, but they still won't understand meaning or context. (Even Siri's legendary snarkiness is just programming, after all.) And, as this post hopefully shows, even human-to-human speech can be impossibly confusing based on misinterpretation of meaning and context. Besides, eschewing ambiguity would also result in the loss of scores of classic jokes; ask Groucho about that one.

Image credits: Information Insights | Speculative Grammarian