User:Kaylavel/garden path

Syntactic Ambiguity
Syntactic ambiguity refers to a property of a string of words that have more than one possible syntactic structure. This ambiguity can apply to whole sentences and persist even after the sentence is processed (permanent ambiguity) or it can refer to ambiguity that is resolved as the sentence is being processed (temporary ambiguity).

Sentence Processing
After a sentence is heard or read, it must be parsed, a process through which the syntactic structure of the sentence is determined by the listener to extract the meaning. This is accomplished through a cognitive mechanism termed the Human Sentence Processing Mechanism (HSPM). A sentence that is syntactically ambiguous is one that can be parsed in multiple ways, leading to confusion in the listener, especially in the absence of contextual cues. The ways in which a sentence can be parsed can be visualized by hierarchical diagrams called “parsing trees” in which the syntactic relations and grammatical roles of words in a sentence are displayed in a network of nodes (See Figure 1 and FIgure 2).

Examples of Permanent Ambiguity
Permanently ambiguous sentences do not contain information that can disambiguate them by the end of processing.
 * 1) Milk drinkers are turning to powder.
 * 2) Iraqi head seeks arms.
 * 3) Grandmother of eight make hole in one.
 * 4) Kids make nutritious snacks.
 * 5) Drunk gets nine months in violin case.

Examples of Temporary Ambiguity
Temporarily ambiguous sentences refer to sentences whose ambiguity can be resolved at some time during sentence processing.


 * 1) I saw the man with the binoculars that I had just picked up
 * 2) Pour the egg in the bowl over the flour

A Garden Path Sentence is a special case of a temporarily ambiguous sentence where there is a strongly preferred initial interpretation that is later confounded by additional information.

Garden Path Sentences
A garden path is a sentence that has more than one possible syntactic structure, which creates a temporary ambiguity. The point at which the correct structure of the garden path becomes clear is called the disambiguation point.

Examples of Garden Paths
One often-cited example of a garden path sentence is The horse raced past the barn fell. In this sentence, the reader builds an initial parse, in which there is a horse and it is being raced past a barn. However, upon reaching the word “fell," the reader is forced to re-evaluate the sentence and form a new structure that is consistent with this new information. Then the reader forms the structure in which there is a specific horse, the one that is being raced past the barn, and it fell. This situation arises from the ambiguity in the phrase “raced past the barn." Specifically, “the horse” can act as the subject, “raced” as the verb, and “past the barn” as the prepositional phrase. However, when the word “fell” must be incorporated, this structure no longer works, and a new one, in which “raced past the barn” modifies “horse," must be formed. Some other examples of garden path sentences: Garden path sentences can often be found in newspaper headlines. For example:
 * “Delays Dog Deaf-Mute Murder Trial”
 * “British Banks Soldier On”
 * “Family Leave Law a Landmark Not Only for Newborn’s Parents”

Models of Processing
There are two main theories that explain how a person might parse a garden path sentence.

Garden Path Model
The Garden Path Model relies on a serial processing system in two stages.

Stage One: In the first stage, only syntactic information is used to process the sentence. The parser identifies the lexical categories of the words (noun, verb, preposition, etc.) and begins to build the structure. At this point, the parser does not care about what the specific words are. Instead, s/he uses processing heuristics to decide on a structure.

Minimal attachment stipulates that new information should be incorporated into the existing structure in such a way as to create the simplest grammatical structure available, with the fewest nodes possible.

Rayner and Frazier (1987) tracked eye movements and found that sentences that violated minimal attachment took people longer to process. People spend more time looking at the disambiguating region of these sentences. For example, in the sentence The criminal confessed his sins harmed many people, minimal attachment says that ‘sins’ is the direct object of ‘confessed,’ but when the parser reaches the word ‘harmed,’ they have to reassess their initial parse because the sentence actually requires the more complicated structure for the correct parse.

Late closure stipulates that new material should be attached low, and incorporated into the phrase currently being processed. Late closure is not a consistent principle in all languages. Spanish speakers, for example, prefer higher attachment. This could be due to several reasons. Spanish may be an exception to this processing technique, or the frequency of correct parses with high attachment might be greater than those with low attachment.

Frazier and Rayner (1982) found that sentences that violate late closure are more difficult to process, such as the sentence Since Alice always jogs a mile and a half seems a short distance to her. The parser tries to attach “a mile and a half” to the structure as the direct object of “jog,” but has to reassess their decision when they get to “seems."

Note: If minimal attachment and late closure are in conflict with one another, minimal attachment is the stronger heuristic.

Main assertion states that the reader should structure the sentence in such a way that new information relates to the main assertion of the sentence.

Stage Two: If the reader encounters an unresolved ambiguity in their first parse of the sentence, they will launch stage two to resolve the ambiguity. In the second stage, the structure is checked against the rest of the information available in the sentence, including semantics (meaning) and pragmatics (context). The parser then uses this information to revise their original interpretation.

Constraint-Based Model
The Constraint-Based Model uses multiple sources of information simultaneously to activate all possible structures in parallel and decide on the correct one. Some of the information used includes the frequency of a particular structure, the specific verb-information (e.g. transitivity), and the fit of the word with the rest of the sentence. The referential context account, which is a specific version of constraint-based theory, suggests that when there is more than one choice in structure available, the reader should build a structure consistent with the semantic information available, and this structure should be one that allows for the least amount of ambiguity possible. Also, one should rank the likelihood of certain structures based on what types of syntactic structures are encountered most commonly.

Summary
The difference between the two theories can be summarized by the following three questions:
 * Does parsing take place in one or in two stages?
 * Does non-syntactic information affect the initial parse?
 * Do garden path sentences require a modular or connectivist explanation?

Testing Methods
There are a few prominent techniques that are used to study garden path sentences. Xhe quick brown xxxxxxxxxxxxxxxxxxxxxxxxxxxx Xxxxxxxxx brown fox jumxxxxxxxxxxxxxxxxxxxxx
 * 1) Rapid Serial Visual Presentation: The words of a sentence are shown individually in succession, controlling the amount of time that can be spent processing each individual word.
 * 2) Self-paced reading: Subjects read the words of a sentence off of a computer screen one word at a time and press a key to advance to the next word. This allows researchers to measure how much time is spent processing each word. Longer processing times may indicate that the reader is experience a garden path effect, especially if time is spent on the disambiguating region.
 * 3) Moving window: The viewing window of a sentence is sized to include a specific number of characters to either side of a word. For example, the sentence The quick brown fox jumped over the lazy dog may appear as follows:
 * 1) Eye movements: Eye-tracking can be used to measure fixations and regressive movements to specific disambiguation regions of the sentence.
 * 2) Visual world paradigm: Participants in an eye-tracker listen to sentences while presented with either real objects or images. By measuring where participants are looking while listening to sentences related to these objects, researchers can establish whether or not participants are experiencing a garden path effect.

Evidence for Garden Path Theory
Reduced Relatives

In Ferreira and Clifton (1986), the researchers investigated the effects of the animacy of the subject on whether or not the verb was interpreted as the main verb of the sentence or the beginning of a reduced relative clause. They compared responses to the following sentences: They found that people showed similar processing difficulties for both sentences, suggesting that semantic information such as the animacy of the noun is not processed in the initial parse of the sentence. Direct Object Plausability
 * 1) The defendant examined by the lawyer turned out to be unreliable.
 * 2) The evidence examined by the lawyer turned out to be unreliable.

In Ferreira and Henderson (1990), the researchers examined two sentences that differed only in the plausibility of interpreting the verb as having a direct object. They compared responses to the following sentences: In sentence one, it is implausible that “Jill” is the direct object of the verb “hoped," but, in sentence two, it is plausible that “Jill” is the direct object of “wrote." If semantic information is included in the initial parse, one would expect to find ambiguity effects in sentence two, but not in sentence one. However, minimal attachment says that “Jill” should be initially interpreted as a direct object in both sentences, as semantics are not involved in the initial parse. Ferreira and Henderson found a garden path effect for both sentences, meaning that “Jill” was interpreted as a direct object, regardless of semantic information, supporting minimal attachment and garden path theory.
 * 1) Bill hoped Jill arrived safely today.
 * 2) Bill wrote Jill arrived safely today.

Electrical Activity

Electrodes can be used to measure brain activity in response to events. These measures are called event-related potentials (ERPs). After reading a semantic anomaly in a sentence, a N400, a negative amplitude occurring 400 milliseconds after the event, can be observed in certain brain areas. After reading a syntactic anomaly, a P600, a positive amplitude occurring 600 milliseconds after the event, will occur. These effects occur in different parts of the brain based on if the anomaly is semantic or syntactic. This suggests that semantic and syntactic information may be processed separately, and at different times, supporting garden path theory.

Evidence for the Constraint-Based Model
A Response to Animacy

Trueswell, Tanenhaus, and Garnsey (1994) attempted to replicate Ferreira and Clifton (1986), who found that the animacy of a noun does not influence the garden path effect, and address some concerns with the methods and materials of the original study. By altering their materials, Trueswell et al. found that animacy did affect processing; sentences with inanimate nouns showed reduced garden path effects. This suggests that semantics do have some effects in the first stage of processing. However, Clifton et. al. (2003) responded to this finding, arguing that preview effects, or the ability to see the disambiguating region before fixating on it, allowed them to enter reanalysis early, causing the observed effect.

Verb Bias and Plausibility

Garnsey, Pearlmutter, Myers, and Lotocky (1997) investigated ambiguities involving different types of verbs. They then manipulated the plausibility of the object nouns to follow in each category. Their materials included three conditions: They found plausibility effects in the DO-bias condition but not the SC-bias condition. This suggests that semantic information such as plausibility can have effects on the initial parse, supporting the constraint-based model.
 * DO-bias: verbs usually followed by a direct object. Example: The talented photographer accepted the money/fire...
 * SC-bias: verbs usually followed by an embedded sentence. Example: The ticket agent admitted the mistake/airplane...
 * EQ-bias: verbs that were equally likely to be followed by a direct object or an embedded sentence. Example: The proud mother announced the wedding/flowers...

Visual Context Effects

Tanenhaus, Spivey-Knowlton, Eberhard, and Sedivy (1995) used a visual world paradigm to examine ambiguity resolution. Participants were evaluated with an eye-tracker while they listened to sentences such as Put the apple on the towel in the box. In this sentence, the attachment of “on the towel” is ambiguous. It may attach high, as the location of the apple, or low, as a modifier of the word apple. While hearing this sentence, participants were exposed to one of two pictures on a monitor. In the first picture, participants saw an apple sitting on a towel, an empty towel, a pencil, and an empty box. In the second picture, participants saw the same scene, except an apple sitting on a napkin replaced the pencil. These conditions elicited different types of eye movements. In the one-apple condition, participants looked to the apple and then to the empty towel, suggesting that participants went with the high attachment interpretation, where “on the towel” refers to where the apple should be placed. In the two-apple condition, participants looked at the two apples but not the towel, suggesting that they had chosen the low attachment interpretation, where “on the towel” modifies “apple." This suggests that context information available in the visual world can bias attachment preference, even increasing a preference for a more complex structure, which is inconsistent with garden path theory, where context does not affect the initial parse.

Context

Taraban and McClelland (1988) examined the effect of semantic properties and context by examining two similar sentences. They compared reading time on the following: They found that people took longer to read the second sentence, even though it is grammatically simpler. This suggests that people took into account information about the verb “stole” and found the context of “in the museum” to be more plausible than “in the night." This is evidence of semantic information affecting the initial parse, which is inconsistent with garden path theory.
 * 1) The thieves stole all the paintings in the museum while the guard slept.
 * 2) The thieves stole all the paintings in the night while the guard slept.

Garden Paths in Speech
Garden paths are less common in speech because of prosodic cues. Prosody refers to the rhythm, intonation, and stress of speech.

There are two types of prosodic cues:
 * Cooperating prosody refers to prosodic cues that direct listeners toward the appropriate structure in a garden path sentence.
 * Conflicting prosody refers to prosodic cues that direct listeners toward the inappropriate structure in a garden path sentence.

Kjelgaard & Speer (1999) looked to see how prosody affected a parser's interpretation of a garden path sentence. They tested statements of the following type:
 * 1) When Roger leaves the house is dark.
 * 2) When Roger leaves the house it’s dark.

The first sentence leads the parser down a garden path because "the house" is initially interpreted as the direct object of the verb when it is actually the subject of the main clause. The second sentence does not lead the parser down a garden path.

The first sentence should be easier to understand if there it is pronounced with a pause after “leaves,” and altered pitch, length, and duration of some words.

Kielgaard & Speer used two different types of tasks to see if adjusting the prosody of a garden path sentence facilitated parsing. First, they asked subjects to press a button when they thought they understood the meaning of the sentence. They also asked subjects to “respond to a visual target word" (e.g. they asked the participant to say ‘is’ after ‘house,' which should be easier to say if s/he had been building the correct syntactic structure all along). They found that performed faster in both tasks when the sentence had cooperating prosody than conflicting prosody.