9.15: Language Comprehension and Production

Language features – Syntax and Semantics

In this chapter the main question will be “how do we understand sentences?”. To find an answer to that problem it is necessary to have a closer look at the structure of languages. The most important properties every human language provides are rules which determine the permissible sentences and a hierarchical structure (phonemes as basic sounds, which constitute words, which in turn constitute phrases, which constitute sentences, which constitute texts). These feature of a language enable humans to create new unique sentences. The fact that all human languages have a common ground even if they developed completely independent from one another may lead to the conclusion that the ability to process language must be innate. Another evidence of a inborn universal grammar is that there were observations of deaf children who were not taught a language and developed their own form of communication which provided the same basic constituents. Two basic abilities human beings have to communicate is to interpret the syntax of a sentence and the knowledge of the meaning of single words, which in combination enables them to understand the semantic of whole sentences. Many experiments have been done to find out how the syntactical and semantical interpretation is done by human beings and how syntax and semantics works together to construct the right meaning of a sentence. Physiological experiments had been done in which for example the event-related potential (ERP) in the brain was measured as well as behavioristic experiments in which mental chronometry, the measurement of the time-course of cognitive processes, was used. Physiological experiments showed that the syntactical and the semantical interpretation of a sentence takes place separately from each other. These results will be presented below in more detail.

Physiological Approach

Semantics

Semantical incorrectness in a sentence evokes a N400 in the ERP

Semantical incorrectness in a sentence evokes an N400 in the ERP The exploration of the semantic sentence processing can be done by the measurement of the event-related potential (ERP) when hearing a semantical correct sentence in comparison to a semantical incorrect sentence. For example in one experiment three reactions to sentences were compared:

Semantically correct: “The pizza was too hot to eat.” Semantically wrong: “The pizza was too hot to drink.” Semantically wrong: “The pizza was too hot to cry.”

In such experiments the ERP evoked by the correct sentence is considered to show the ordinary sentence processing. The variations in the ERP in case of the incorrect sentences in contrast to the ERP of the correct sentence show at what time the mistake is recognized. In case of semantic incorrectness there was observed a strong negative signal about 400ms after perceiving the critical word which did not occure, if the sentence was semantically correct. These effects were observed mainly in the paritial and central area. There was also found evidence that the N400 is the stronger the less the word fits semantically. The word “drink” which fits a little bit more in the context caused a weaker N400 than the word “cry”. That means the intensity of the N400 correlates with the degree of the semantic mistake. The more difficult it is to search for a semantic interpretation of a sentence the higher is the N400 response.

Syntax

Syntactical incorrectness in a sentence can evoce an ELAN (early left anterior negativity) in the electrodes above the left frontal lobe after 120ms.

To examine the syntactical aspects of the sentence processing a quite similar experiment as in the case of the semantic processing was done. There were used syntactical correct sentences and incorrect sentences, such as (correct:)“The cats won´t eat…” and (incorrect:)“The cats won´t eating…”. When hearing or reading a syntactical incorrect sentence in contrast to a syntactical correct sentence the ERP changes significantly on two different points of time. First of all there a very early increased response to syntactical incorrectness after 120ms. This signal is called the ‘early left anterior negativity’ because it occurs mainly in the left frontal lobe. This advises that the syntactical processing is located amongst others in Broca's area which is located in the left frontal lobe. The early response to syntactical mistakes also indicates that the syntactical mistakes are detected earlier than semantic mistakes.

The other change in the ERP when perceiving a syntactical wrong sentence occurs after 600ms in the paritial lobe. The signal is increasing positively and is therefore called P600. Possibly the late positive signal is reflecting the attempt to reconstruct the grammatical problematic sentence to find a possible interpretation.

Syntactical incorrectness in a sentence evokes after 600ms a P600 in the electrodes above the paritial lobe.

To summarize the three important ERP-components: First of all there occurs the ELAN at the left frontal lobe which shows a violation of syntactical rules. After it follows the N400 in central and paritial areas as a reaction to a semantical incorrectness and finally there occurs a P600 in the paritial area which probably means a reanalysis of the wrong sentence.

Behavioristic Approach – Parsing a Sentence

Behavioristic experiments about how human beings parse a sentence often use syntactically ambiguous sentences. Because it is easier to realize that sentence-analysing mechanisms called parsing take place when using sentences in which we cannot automatically constitute the meaning of the sentence. There are two different theories about how humans parse sentences. The syntax-first approach claims that syntax plays the main part whereas semantics has only a supporting role, whereas the interactionist approach states that both syntax and semantics work together to determine the meaning of a sentence. Both theories will be explained below in more detail.

The Syntax-First Approach of Parsing The syntax-first approach concentrates on the role of syntax when parsing a sentence. That humans infer the meaning of a sentence with help of its syntactical structure (Kako and Wagner 2001) can easily be seen when considering Lewis Carroll´s poem ‘Jabberwocky’:

"Twas brillig, and the slithy toves Did gyre and gimble in the wabe: All mimsy were the borogoves, And the mome raths outgrabe."

Although most of the words in the poems have no meaning one may ascribe at least some sense to the poem because of its syntactical structure.

There are many different syntactic rules that are used when parsing a sentence. One important rule is the principle of late closure which means that a person assumes that a new word he perceives is part of the current phrase. That this principle is used for parsing sentences can be seen very good with help of a so called garden-path sentence. Experiments with garden-path sentences have been done by Frazier and Fayner 1982. One example of a garden-path sentence is: “Because he always jogs a mile seems a short distance to him.” When reading this sentence one first wants to continue the phrase “Because he always jogs” by adding “a mile” to the phrase, but when reading further one realizes that the words “a mile” are the beginning of a new phrase. This shows that we parse a sentence by trying to add new words to a phrase as long as possible. Garden-path sentences show that we use the principle of late closure as long it makes syntactically sense to add a word to the current phrase but when the sentence starts to get incorrect semantics are often used to rearrange the sentence. The syntax-first approach does not disregard semantics. According to this approach we use syntax first to parse a sentence and semantics is later on used to make sense of the sentence.

Apart from experiments which show how syntax is used for parsing sentences there were also experimens on how semantics can influence the sentence processing. One important experiment about that issue has been done by Daniel Slobin in 1966. He showed that passive sentences are understood faster if the semantics of the words allow only one subject to be the actor. Sentences like “The horse was kicked by the cow.” and “The fence was kicked by the cow.” are grammatically equal and in both cases only one syntactical parsing is possible. Nevertheless the first sentence semantically provides two subjects as possible actors and therefore it needs longer to parse this sentence. By measuring this significant difference Daniel Slobin showed that semantics play an important role in parsing a sentence, too.

The Interactionist Approach of Parsing

The interactionist approach ascribes a more central role to semantics in parsing a sentence. In contrast to the syntax-first approach, the interactionist theory claims that syntax is not used first but that semantics and syntax are used simultanuasly to parse the sentence and that they work together in clearifying the meaning. There have been made several experiments which provide evidence that semantics are taking into account from the very beginning reading a sentence. Most of these experiments are working with the eye-tracking techniques and compare the time needed to read syntactical equal senences in which critical words cause or prohibit ambiguitiy by semantics. One of these experiments has been done by John Trueswell and coworkers in 1994. He measured the eye movement of persons when reading the following two sentences:

The defendant examined by the lawyer turned out to be unreliable. The evidence examined by the lawyer turned out to be unreliable.

He observed that the time needed to read the words “by the lawyer” took longer in case of the first sentence because in the first sentence the semanics first allow an interpretation in which the defendant is the one who examines, while the evidence only can be examined. This experiment shows that the semantics also play a role while reading the sentence which supports the interactionist approach and argues against the theory that semantics are only used after a sentence has been parsed syntactically.

Inferences Creates Coherence

Coherence is the semantic relation of information in different parts of a text to each other. In most cases coherence is achieved by inference; that means that a reader draws information out of a text that is not explicitly stated in this text. For further information the chapter Psychology and Cognitive Neuroscience/Situation Models and Inferencing#Neuropsychology of Inferencing Neuroscience of Text Comprehension should be considered.

Situation Model

A situation model is a mental representation of what a text is about. This approach proposes that the mental representation people form as they read a story does not indicate information about phrases, sentences, paragraphs, but a representation in terms of the people, objects, locations, events described in the story (Goldstein 2005, p. 374)

For a more detailed description of situation models, see Psychology and Cognitive Neuroscience/Situation Models and Inferencing Situation Models