Psychology 215
Running Head: SEMANTIC AND SYNTACTIC INTERACTIONS
Semantic and Syntactic Interactions:
Processes in Sentence Parsing
A sentence is a set of words that expresses a thought. It can convey a statement, a question, an exclamation or a command. The basic structure of a sentence consists of a noun phrase (subject) and a verb phrase (predicate). Words a naturally grouped by phrases. The phrase structure of a sentence also guides its interpretation. It transmits critical information about who did what to whom. Studies involving patients suffering with Broca's Aphasia have given support to the role of semantics in sentence parsing. Patients with this condition would be confused about who did what to whom with a sentence that said, "the boy was kicked by the girl." However, they could easily comprehend a sentence that said, "the ball was kicked by the girl," as they would know that a ball is something that is usually kicked and a person is usually the one doing the kicking. Syntax plays a role in sentence parsing as well. The sentence, "the girl kicked the boys," tells the reader that there was more than one boy being kicked. Changing it to, "the girl kicker the boy," gives the reader the feeling that the sentence is now incomplete. Sentences can become very complicated as well. It is quite possible to say, "the boys who dined at nine was kicked by the girl kicker who devoured the chicken." In order for a person to understand a sentence, he or she needs to parse the sentence and figure out the phrase structure. Sentence parsing involves an interaction between syntactic analysis and semantic analysis.
Sometimes there is more than one phrase structure is compatible with a sentence. When this happens, there will be more than one way to interpret the sentence and more than one meaning of the sentence. This phenomenon is called a phrase-structure ambiguity and it can be used to study how people process the phrase structure of a sentence. An example of phrase-structure ambiguity can be seen in the sentence, "I saw the chicken in my bathrobe." In one interpretation the sentence could be parsed into the two phrases, "I saw," and, "the chicken in my bathrobe." This would imply that the chicken was wearing the narrator's bathrobe. The second interpretation would be the phrases, "I saw the chicken," and, "in my bathrobe." This suggests that the narrator was wearing a bathrobe while seeing the chicken. Parsing is influenced by function words that appear in a sentence. Re-writing the previous sentence to say, "I saw the chicken while in my bathrobe," or, "I saw the chicken that was in my bathrobe," would eliminate the ambiguity. The prefixes and suffixes in a word also signal a syntactic role. In the sentence, "he smurfly smurfed the smurfy smurf", the first "smurf" ends in a "ly", indicating that is an adverb. The second "smurf" ends with an "ed" indicating that it's a verb. The third one ends with a "y", signaling an adjective. Since an adjective describes a noun, the last "smurf" must be a noun. (Note: the meaning of profane words can also be derived from it's syntactical position in a sentence, though I have yet to find any studies which have done this.) From these examples, it is clear to see that syntax is used to guide sentence parsing. English speakers tend to assume that the sentences that they are hearing or reading will be in active from (e.g. "the girl kicked the boy") rather than passive form (e.g. "the boy was kicked by the girl"), Further evidence comes from a study by Hornby (1974). Thirty college students listened to a pre-recorded sentence followed one second later by a
brief tachistoscopic presentation of a picture for 50 msec. The subjects' task was to identify whether or no the picture matched what was being described by the sentence. There were 30 trials in total and 24 of them had a discrepancy between the picture and the sentence. The subjects would be divided into two groups; one group would hear active versions of the sentence while the other would hear passive versions. The mean number of times that the misrepresentation was over looked was measured. For active sentences it was 1.51, for passive sentences it was 2.39, F(1,29)=18.10, p<.001. Thus English speakers understand active sentences better than passive ones.
According to Trueswell, Tanenhaus, and Kello (1995), in addition to using syntactic rules, people tend to use a strategy of minimal attachment. At points of ambiguity, the listener or reader will seek out the simplest, grammatically correct phrase. For example, in the sentence, "the old man the ships," a temporary ambiguity will occur because people will initially interpret, "the old man," as the noun phrase. However, when they read the whole sentence, people will then re-interpret the sentence into the phrases, "the old," and, "man the ships." The syntax of a sentence plays an important role in sentence-parsing decisions.
Part of the reason for the existence of ambiguity inside the sentence, "the old man ships," is that the word "man" usually refers to a male human being, not a verb that means "to operate." Most people would also assume, "the old man," is the noun phrase because "old" is usually used as an adjective. This suggests that sentence parsing is guided by semantic factors. Some sentences are reversible. That is, the object and subject can switch places in the sentence. "The boy kicked the girl," or, "the girl kicked the boy," are both acceptable sentences. However, "the girl kicked the ball," is not reversible because balls cannot kick people. Some words can only be used in certain contexts. Some verbs require a subject. For example, it is correct to say, "we dined at nine," but not so to say, "we dined the chicken." Other verbs may require an object to act upon. Using a similar situation as the previous example, it would be grammatically correct to say, "we devoured the chicken," but wrong to say, "we devoured at nine." Some verbs demand that both a subject and an object be present. Some don't need either at all (e.g. "we eat.") Every time such a verb is encountered the reader must consult his or her own knowledge of the word and interpret the appropriate context it is used in. In a study by Trueswell, Tanenhaus and Kello (1995), thirty-six undergraduate students listened to a sentence fragment on a computer. Their task was to say out loud the appropriate word (from a choice of two words) that would complete the sentence into a microphone as fast as possible. Some of the sentences required an object or subject to act upon (e.g. "we devoured"). Other sentences used a verb that did not require something to act upon (e.g. "we ate"). There were 58 trials in total with 16 of them being the experimental ones. Results showed that the response time for the sentences that required something to act upon was 512 msec while other type was 526 msec, F(1,32)=7.03, p<.05. Thus the evidence supported that verb-based patterns are exploited by sentence parsing. Semantics, the interpretation and meaning of the sentence, is involved in sentence-parsing decisions.
Tyler and Marlen-Wilson (1995) also found that an interaction occurred between the semantics of a sentence and its syntax (passive versus active). Using the same subjects, and format, the task was to identify the correct verb type that would complete a noun phrase. There were two different types of noun phrases being looked at. Ones that were in a passive format (e.g., "by the river he") and others in an active format (e.g., "the subject"). The results were that passive format phrases were responded to quickly when followed by a verb that didn't require something to act on (83 msec) but when these phrases were followed by a verb that did require something to act on, the response time was slower (148 msec, F(1,20)=19.83, p<.01). For active phrases followed by verb required something to act on, the response time was 195 msec, but 95 msec for verbs that didn't require something to act on, F(1,20)=20.46, p<.01. According to Trueswell et al., the results indicate that the listener spontaneously retrieves verb-based information as the sentence is being heard, word-by-by word. By combining semantic and syntactic information, the listener will compose the phrase. According to this model, as soon the listener hears the first two words in the sentence, "the old man ships," the listener would naturally expect a noun to follow as "old" is initially interpreted to be an adjective. When the word "man" is heard, the listener has his/her mini-hypothesis confirmed, stores the first three words as a noun phrase and starts expecting to hear a verb phrase. When the word "ships" is registered, the listener initially interprets it as a verb, because it is consistent with his/her expectation. As soon as the listener is aware that, "the old man ships," is the complete sentence, the listener returns to the sentence and re-parses it into an acceptable noun phrase and verb phrase based on his/her semantic knowledge of the words "old, man," and, "ships." This demonstrates that syntactic information of the sentence is used as a guide for semantic analysis.
In 1997, Wurm was interested to see whether word processing utilized a whole-word routine that would consider the entire word or a decompositional routine which would break down the words and find the root word that combined with the prefix. He had 21 students listen to words and rate them on their "prefixedness." "Ablaze and blaze" are examples of prefixed and a non-prefixed word used in the study. Words that were pre-determined to be fully formed (with prefix) were recognized on average 39 msec (p<.05) faster than their root versions (same word, with no prefix). Since the whole word was recognized faster than its root, it would suggest that the words are processed in a strict left to right manner as previously thought. This means that words are accessed as complete units in the brain whether or not they contain suffixes or prefixes. An alternative explanation that prefixes were processed very early in the stage of recognition was dismissed because prefixes should not play a role in the early stages of processing if act of hearing the entire word had to be completed first. The implication from this is that the brain would have separate entries for words like "cover" and "discover" which would result in a huge redundancy in stored knowledge. But given the large capacity of the human brain, most linguists are relaxing their instance on maximally efficient storage principles. So when a word is read or heard, it is immediately recognized as a noun, verb, etc. The categorization of the word helps identify its syntactic position within a sentence and primes the listener or reader for the next word of the appropriate type. This speeds up recognition and demonstrates that the considerations of semantic meaning guide syntactic analysis.
Swinney and Cutler (1977) provide further evidence for the interactionist view one their study of idioms. Idioms are a group of words that have a meaning that cannot be deduced by the individual words that comprise it. For example the phrase "kick the bucket,' usually has nothing to do with either a "kick" or a "bucket". Twenty one undergraduate students sat in front of a computer that flashed stings of words for two seconds. The subjects' task was to identify whether or not the word strings were grammatical English phrases. Thirty of the trials featured word strings that were non-grammatical (e.g. "stranger is during"). Twenty-three of the trials featured grammatical idiom strings. There were twenty-three grammatical control strings that matched the structure of the idioms but had one of the words replaced (e.g. "kicked the can"). There were a total of 152 trials. The results were that the subjects responded faster to idioms (995 msec) than the control strings (1064 msec, F(1,19)=48.5, p<.0001). The results indicate that both the idiomatic and literal meanings of a phrase are simultaneously initiated when processing a sentence. This would imply that individual words are semantically checked as they are heard or read and compared with their syntactic location within the sentence.
All sentences are composed of a noun phrase and a verb phrase. In order to understand who is the "doer" and what is being "done", the listener or reader must be able to figure out which part of the sentence refers to the "doer" and which part refers to what is being "done." There are two important sources of information that can be used to derive meaning from a sentence. Syntax, the knowledge of general grammatical rules and how they are applied, is often used to figure out the semantic relationship between words. Semantics, the understanding of the meaning of the words, is used to figure out what type of word comes next. When parsing sentences, listeners or readers will use syntactic information to guide semantic interpretation and semantic information to guide syntactic analysis. There is an interaction between semantics and syntax in the sentence-parsing process.
References
Hornby, P. (1974). Surface structure and presupposition. Journal of Verbal Learning and Verbal Behavior, 13, 530-538.
May, C., Kane, M., & Hasher, L. (1995). Determinants of negative priming. Psychological Bulletin, 118, 35-54.
Swinney, D. & Cutler, A. (1977). The access of processing of idiomatic expressions. Journal of Verbal Learning and Verbal Behavior, 16, 523-554
Trueswell, J., Tanenhaus, M., & Kello, C. (1995). Verb-specific constraints in sentence processing: Separating effects of lexical preference from garden paths. Journal of Experimental Psychology: Learning, Memory and Cognition, 19, 528-553.
Tyler, L., & Marslen-Wilsen, W. (1977). The online effects of semantic context on syntactic processing. Journal of Verbal Learning and Verbal Behavior, 16, 683-692.
Wurm, L. (1997). Auditory processing of prefixed words. Journal of Verbal Learning and Verbal Behavior, 26, 438-459.
Back