Language is complex and rich, and the ability to acquire it is nothing short of remarkable. In an attempt to help explain how learners manage to do so, researchers have looked to the speech we hear, to see whether it contains any information that can help us along the way. In this post, I will talk about the way that we can look to information contained within speech itself to help us learn a particular linguistic feature: recursive grammar.
Imagine one day, you have a talk with your friend, and she says, "The cat the dog chased meowed". This is an example of recursive grammar; structures in this type of grammar are composed of a set of related elements (“The cat meowed”), separated by another set of related elements (“The dog chased”), which combine to produce a sentence with a new meaning. These kinds of structures are referred to as centre-embedded structures. Recursive grammar has been the focus of much study, as it is often quite difficult to learn and understand, although the example above this should have been quite easy. Why? Well, you know that dogs chase cats, and that cats meow. This is what we call a semantic dependency, and these help to make one interpretation more plausible than the other. In other words, our knowledge of the subject matter provides a quick, cheap and reliable basis for comprehension, meaning learners’ don’t have to puzzle out the structure of the sentence.
What happens then if you have a sentence where the meanings are more ambiguous? Or, what if you have yet to learn these dependencies? The quirks of the way you say something can fill the gap. Fortunately, it is not just the content of what we say that counts, it is also the way we say it. A great example is pitch variation, and no example is better than the way you ask a question. Generally speaking, to turn a statement into a question, you simply use a higher inflection at the end of the statement. People can also use pitch to reveal the structure of a sentence. When saying, "The boy the girl kissed blushed", people tend to voice 'the girl' and 'kissed' (an embedded clause) at a similar register relative to 'the boy' and 'blushed'. Not only this, but people tend to say 'the boy' at the highest register, indicating the sentence has started, and 'blushed' in the lowest register, to indicate the statement is over1.
Pitch is not the only important vocal cue however; rhythm plays it’s part. Whilst speaking, individuals rarely pause between words, but they do between clauses. In a recent study, it was found that inserting short pauses in an artificial language helped infants learn it’s grammar2. In everyday speech, this is the difference between saying, "StevemetDaniRogerwasannoyed" vs. "StevemetDani Rogerwasannoyed"; adding a pause makes it more obvious that there are two statements, and their meaning helps you recognize that they are related.
To assess the role of these natural language cues in recursive grammar learning, Padraic Monaghan, Rebecca Frost and I trained adults on an artificial language that contained lots of new words, which were put into sentences that followed the structure of a recursive grammar. Participants’ learning was assessed with a short test, which asked them to judge whether a new set of sentences followed or violated the rules of the language they had just learnt. Crucially, we manipulated the presence of cues over five conditions, to assess how they affected learning: learners in the baseline condition received a language containing just the words, with no additional cues to the structure. Learners in the pitch condition received the baseline grammars, with the addition of natural language pitch cues, where dependent non-words were given the same pitch, with each deeper level of embedding uses a lower pitch, and the sentence initial and sentence final elements using the highest and lowest register respectively. In the rhythm condition, the short pauses were inserted between clauses in the speech used for the baseline condition. In the phonological similarity condition, grammatical pairings would always start with the same sound (e.g. “Pa” and “Po”), mimicking the semantic similarities of natural language. Finally, a combined condition utilised the similarity, pitch and rhythmic cues.
So, which cue best-helped learners? Preliminary analysis of our data indicates that participants initially benefit hugely from the presence of multiple information sources (the combined condition), though over time individual cues can become comparably reliable, particularly when two dependent elements are. So, once you know that “the cat meows”, you can use that alone to process a sentence. What this essentially means is that for the early language learner, the way you say something is incredibly important; it supports sentence processing. With better knowledge of language, the linguistic input itself is enough for successful processing. Moving forward, we intend to extend these findings from an artificial language setting into natural language; will these cues support the comprehension of recursive grammars? Will different cues be more important in a natural setting? Keep your eye on this space, as we will hopefully have those answers soon.
1. Fery, C. & Schubö, F. (2010). Hierarchical prosodic structures in the intonation of center-embedded relative clauses. The Linguistic Review, 27, 293-317.
2. Hawthorne, K. & Gerken, L. (2014). From pauses to clauses: Prosody facilitates learning of syntactic constituency. Cognition, 133, 420-428.