Prosody and parsing
Recall that proponents of the syntax-first approach to parsing claimed autonomous syntactic processing, but note also that they based their studies almost exclusively on reading tasks, i.e. on written sentence processing. We saw in earlier chapters that speech carries cues to syntactic structure, as well as to other types of sentence and utterance organisation, such as informational structure and focus. These are prosodic cues, which include intonation patterns as well as the stress and rhythm patterns of words, phrases and sentences. Our concern in this section is with how such cues might be used in the comprehension of sentences.
There are many well-attested effects of prosodic structure on processing, especially in word recognition, as shown also in slips-of-the-ear see Chapter 8. But how might prosody help the parser?
Late Closure predicts that the NP following a verb like mending in (11.48) will be interpreted as the object of that verb. sing punctuation, in the form of a comma after the verb, signals the Early Closure interpretation, with the following NP then being in a new clause, as in (11.49). One way in which prosody might influence structural analysis is therefore by providing comma intonation’ (i.e. over mending the in (11.49).
Is there similar prosodic marking for Minimal Attachment A comma is less likely to distinguish the pairs in (11.50) and (11.51) than those in (11.48) and (11.49), but what about prosody?
The Syntax and Speech’ studies introduced in Chapter 2 considered the acoustic speech properties of a range of ambiguities, including these types, and showed that there are reliable contrasts between different versions of the ambiguity, in pausing, rhythm, amplitude and pitch con tours.
The question then is whether listeners can use such information during their processing of sentences, and if so, when and how they use it. Simple continuation experiments, in which participants listen to spoken fragments of sentences and have to choose completions, have confirmed that prosodic information is used (Schafer, Speer, Warren & White, 2000; Speer et al., 2011. But does prosodic information determine the initial analysis of a sentence, or is it only used in some later post-perceptual analysis to address this, studies have used a variety of online tasks.
Marslen-Wilson et al. (1992) ran a cross-modal naming experiment using a set of Minimal Attachment sentences previously extensively studied in reading experiments (Frazier & Rayner, 1982; Holmes, Kennedy & Murray, 1987; Rayner & Frazier, 1987). In the cross-modal naming experiment, a spoken utterance fragment is followed by a visual probe word, and the participants’ task is to read aloud the visual probe. If the probe word is related to the utterance fragment e.g. provides a good continuation of that fragment, then participants are able to initiate their naming response more rapidly. The experiment tested whether prosodic cues can resolve ambiguity in spoken versions of minimal attachment sentences such as (11.53), just as the complementiser (that) resolves it in (11.52).
Spoken sentence fragments were presented up to the point marked by and followed immediately by the presentation of the visual probe word. For these examples this was the word was. Note that the sentence fragments up to the l in (11.53) and (11.54) contain identical word strings.
Their spoken forms in the experiment differed, however, because 11.53 was a recording from a non-minimal attachment version of the sentence, while 11.54 was from a minimal attachment version.
Marslen-Wilson et al. found that naming times after the minimal attachment fragment (11.54) were significantly slower than after the non-minimal attachment fragment in (11.53), even though the fragments contained identical word strings. Since the only difference between the two sentence fragments is in how they were pronounced, i.e. in their prosody, this result indicates that prosodic information is used online to resolve potential temporary ambiguities. In addition, comparison of response times for the two non-minimal attachment versions i.e. (11.52) and (11.53) showed that it made no difference whether or not a complementiser was present. This shows that prosodic information is just as effective in online parsing as explicit syntactic marking.
Similar results were found in an experiment using utterances involving the Late Closure strategy (Warren, Grabe & Nolan, 1995). Warren e al used sentence fragments like (11.55), followed by a visually presented continuation word for naming e.g. a se that was only compatible with the early closure interpretation of the fragment (i.e. closure after Hong Kong than, rather than after problems). When the spoken version of the utterance was taken from a late closure reading, naming times were significantly slower.
By manipulating the recordings that were used, Warren e al also showed that participants were sensitive to the presence or absence of phonological processes like stress shift. Stress shift is a phenomenon in English and in some other languages that prevents two stressed syllables from being too close together or clashing. For instance, the main stress that is on the second element of Hong Kong would be expected to shift to the first element when the word problems is in the same prosodic constituent, giving Hong Kong problems. It is considerably more likely that problems will be in the same constituent as Hong Kong in the late closure reading.
Structuring utterances through the placement of prosodic breaks can have a profound influence on the preferred analysis. For example, Pynte and Prieur (1996) ran a word monitoring task using sentences such as (11.56) and (11.57). The experiment was conducted in French, but the structures are equivalent. The target word was the final word in each sentence, i.e. the noun in a PP which is in a structurally ambiguous relationship to the preceding material, since it could modify either the verb or the object noun. Note that participants did not know that the target word was going to be the last one in the utterance, and indeed in filler trials in the experiment the target word appeared in many different positions, and possibly not at all. Word monitoring times in Pynte and Prieur’s experiment showed that a prosodic break after the verb made the attachment of the PP to the noun easier to process. This is because the break forces the linking of the PP to the noun guards rather than to the verb informed by putting it in the same prosodic phrase as guards, and separating it from informed.
The effects of prosody also depended on the preferred argument structure of the verb. A verb like inform seems to expect a prepositional phrase as part of the sentence i.e. we normally inform someone of or about some thing, but a verb like choose does not. So a prosodic break after informed made a major contribution to the interpretation of the sentence in (11.57), but a prosodic break after chose in (11.58) played a less significant role, as the preference was already to interpret the PP as a modifier of the object noun which works for (11.58) but not for (11.59).
Other research shows that listeners are sensitive to the use of prosody and intonation to mark focus, and that focus can influence structural attachments (Schafer, Carter, Clifton & Frazier, 1996). Recall the relative clause RC attachment ambiguities discussed in Chapter 10. These are NP1-of NP2-RC structures, where the RC can modify either NP1 or NP2 – see the underlined portion of (11.60). The preference shown in reading time studies with English materials is for the RC to modify the second noun phrase so the mechanic was repairing the plane. Schafer e al showed that this preference can be altered in spoken versions of these sentences by stress placement. That is, the relative clause is most likely to be taken to modify whichever of the NPs (the propeller or the plane) is in focus, as marked by stress on that NP.
These findings indicate that the input to the sentence processing system can be enriched by the prosodic features of spoken utterances, and that these features interact with structural and lexical preferences in interesting but complex ways. Interestingly, further research has indicated that prosodic structure may play a role in sentence processing even when participants are reading silently. That is, there is implicit prosody that we hear in our heads’ when reading, and which can help structure the input (Fodor, 2002).
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
