Single-Stage Prediction Models Do Not Explain the Magnitude of Syntactic Disambiguation Difficulty

The disambiguation of a syntactically ambiguous sentence in favor of a less preferred parse can lead to slower reading at the disambiguation point. This phenomenon, referred to as a garden-path effect, has motivated models in which readers initially maintain only a subset of the possible parses of the sentence, and subsequently require time-consuming reanalysis to reconstruct a discarded parse. A more recent proposal argues that the garden-path effect can be reduced to surprisal arising in a fully parallel parser: words consistent with the initially dispreferred but ultimately correct parse are simply less predictable than those consistent with the incorrect parse. Since predictability has pervasive effects in reading far beyond garden-path sentences, this account, which dispenses with reanalysis mechanisms, is more parsimonious. Crucially, it predicts a linear effect of surprisal: the garden-path effect is expected to be proportional to the difference in word surprisal between the ultimately correct and ultimately incorrect interpretations. To test this prediction, we used recurrent neural network language models to estimate word-by-word surprisal for three temporarily ambiguous constructions. We then estimated the slowdown attributed to each bit of surprisal from human self-paced reading times, and used that quantity to predict syntactic disambiguation difficulty. Surprisal successfully predicted the existence of garden-path effects, but drastically underpredicted their magnitude, and failed to predict their relative severity across constructions. We conclude that a full explanation of syntactic disambiguation difficulty may require recovery mechanisms beyond predictability.     

Expectation-based syntactic comprehension

This paper investigates the role of resource allocation as a source of processing difficulty in human sentence comprehension. The paper proposes a simple information-theoretic characterization of processing difficulty as the work incurred by resource reallocation during parallel, incremental, probabilistic disambiguation in sentence comprehension, and demonstrates its equivalence to the theory of Hale [Hale, J. (2001). A probabilistic Earley parser as a psycholinguistic model. In Proceedings of NAACL (Vol. 2, pp. 159-166)], in which the difficulty of a word is proportional to its surprisal (its negative log-probability) in the context within which it appears. This proposal subsumes and clarifies findings that high-constraint contexts can facilitate lexical processing, and connects these findings to well-known models of parallel constraint-based comprehension. In addition, the theory leads to a number of specific predictions about the role of expectation in syntactic comprehension, including the reversal of locality-based difficulty patterns in syntactically constrained contexts, and conditions under which increased ambiguity facilitates processing. The paper examines a range of established results bearing on these predictions, and shows that they are largely consistent with the surprisal theory.     

Effects of Surprisal and Locality on Danish Sentence Processing: An Eye-Tracking Investigation

An eye-tracking experiment in Danish investigates two dominant accounts of sentence processing: locality-based theories that predict a processing advantage for sentences where the distance between the major syntactic heads is minimized, and the surprisal theory which predicts that processing time increases with big changes in the relative entropy of possible parses, sometimes leading to anti-locality effects. We consider both lexicalised surprisal, expressed in conditional trigram probabilities, and syntactic surprisal expressed in the manipulation of the expectedness of the second NP in Danish constructions with two postverbal NP-objects. An eye-tracking experiment showed a clear advantage for local syntactic relations, with only a marginal effect of lexicalised surprisal and no effect of syntactic surprisal. We conclude that surprisal has a relatively marginal effect, which may be clearest for verbs in verb-final languages, while locality is a robust predictor of sentence processing.     

Interaction of stimulus and contextual information during reading: Identifying words within sentences

In general, studies on the effects of a sentence context on word identification have focused on how context affects the efficiency of processing a single target word, presented separately from the context. Such studies probably would be incapable of measuring contextual facilitation resulting from cascaded or parallel processing of neighboring words within a sentence. To measure these and other types of facilitation, we presented entire phrases and sentences for subjects to read as fast as possible and to monitor for nonwords. Subjects read at rates representative of natural reading. Experiment 1 demonstrated a large contextual facilitation effect on decision time. Experiment 2 showed that facilitation is caused by specific semantic information and, perhaps to a greater degree, by nonpredictive syntactic information. Experiment 3 showed that the amount of facilitation is greater than could be accounted for by separate contributions from autonomous word level and sentence level processes. These results present difficulties for an autonomous model of reading, but are consistent with interactive models, in which the results of ongoing sentential analyses are combined with stimulus information to identify words.     

Context effects in sentence comprehension: A study of the subjective lexicon

These studies explore the role of context in determining what information about the meanings of words is activated in memory at the time a word is encountered in a sentence. Using a color-naming paradigm, it was shown that both meanings of a word that has two distinct meanings are activated in memory at the time the word is heard in a sentence. This activation occurs even when there is sufficient contextual information to indicate which meaning was intended by the speaker. These results support the hypothesis that there exists in memory an isolable subjective lexicon. They suggest that context which is effective in disambiguating lexical ambiguities in the language has its effect only at a relatively late stage in the cognitive processing involved in language comprehension.     

The influence of lexical and conceptual constraints on reading mixed-language sentences: Evidence from eye fixations and naming times

In two experiments, we explored the degree to which sentence context effects operate at a lexical or conceptual level by examining the processing of mixed-language sentences by fluent Spanish-English bilinguals. In Experiment 1, subjects’ eye movements were monitored while they read English sentences in which sentence constraint, word frequency, and language of target word were manipulated. A frequency × constraint interaction was found when target words appeared in Spanish, but not in English. First fixation durations were longer for high-frequency Spanish words when these were embedded in high-constraint sentences than in low-constraint sentences. This result suggests that the conceptual restrictions produced by the sentence context were met, but that the lexical restrictions were not. The same result did not occur for low-frequency Spanish words, presumably because the slower access of low-frequency words provided more processing time for the resolution of this conflict. Similar results were found in Experiment 2 using rapid serial visual presentation when subjects named the target words aloud. It appears that sentence context effects are influenced by both semantic/conceptual and lexical information.     

Uncertainty and Expectation in Sentence Processing: Evidence From Subcategorization Distributions

There is now considerable evidence that human sentence processing is expectation based: As people read a sentence, they use their statistical experience with their language to generate predictions about upcoming syntactic structure. This study examines how sentence processing is affected by readers' uncertainty about those expectations. In a self‐paced reading study, we use lexical subcategorization distributions to factorially manipulate both the strength of expectations and the uncertainty about them. We compare two types of uncertainty: uncertainty about the verb's complement, reflecting the next prediction step; and uncertainty about the full sentence, reflecting an unbounded number of prediction steps. We find that uncertainty about the full structure, but not about the next step, was a significant predictor of processing difficulty: Greater reduction in uncertainty was correlated with increased reading times (RTs). We additionally replicated previously observed effects of expectation violation (surprisal), orthogonal to the effect of uncertainty. This suggests that both surprisal and uncertainty affect human RTs. We discuss the consequences for theories of sentence comprehension.     

Data from eye-tracking corpora as evidence for theories of syntactic processing complexity

We evaluate the predictions of two theories of syntactic processing complexity, dependency locality theory (DLT) and surprisal, against the Dundee Corpus, which contains the eye-tracking record of 10 participants reading 51,000 words of newspaper text. Our results show that DLT integration cost is not a significant predictor of reading times for arbitrary words in the corpus. However, DLT successfully predicts reading times for nouns. We also find evidence for integration cost effects at auxiliaries, not predicted by DLT. For surprisal, we demonstrate that an unlexicalized formulation of surprisal can predict reading times for arbitrary words in the corpus. Comparing DLT integration cost and surprisal, we find that the two measures are uncorrelated, which suggests that a complete theory will need to incorporate both aspects of processing complexity. We conclude that eye-tracking corpora, which provide reading time data for naturally occurring, contextualized sentences, can complement experimental evidence as a basis for theories of processing complexity.     

Contextual override of pragmatic anomalies: evidence from eye movements

Readers typically experience processing difficulty when they encounter a word that is anomalous within the local context, such as 'The mouse picked up the dynamite...'. The research reported here demonstrates that by placing a sentence in a fictional scenario that is already well known to the reader (e.g., a Tom and Jerry cartoon, as a context for the example sentence above), the difficulty usually associated with these pragmatic anomalies can be immediately eliminated, as reflected in participants' eye movement behaviour. This finding suggests that readers can rapidly integrate information from their common ground, specifically, their cultural knowledge, whilst interpreting incoming text, and provides further evidence that incoming words are immediately integrated within the global discourse.     

Do individuals with autism process words in context? Evidence from language-mediated eye-movements

It is widely argued that people with autism have difficulty processing ambiguous linguistic information in context. To investigate this claim, we recorded the eye-movements of 24 adolescents with autism spectrum disorder and 24 language-matched peers as they monitored spoken sentences for words corresponding to objects on a computer display. Following a target word, participants looked more at a competitor object sharing the same onset than at phonologically unrelated objects. This effect was, however, mediated by the sentence context such that participants looked less at the phonological competitor if it was semantically incongruous with the preceding verb. Contrary to predictions, the two groups evidenced similar effects of context on eye-movements. Instead, across both groups, the effect of sentence context was reduced in individuals with relatively poor language skills. Implications for the weak central coherence account of autism are discussed.     

Similarity information versus relational information: differences in the time course of retrieval

Two experiments are reported that examine the time course of retrieval in a sentence matching procedure. Subjects learned lists of active and passive sentences and were tested with sentences in active or passive, correct or incorrect versions; for example, if "John hit Bill" was a studied sentence, "Bill hit John" would be an incorrect active test sentence. A response signal procedure was used so that accuracy could be measured as a function of time. The data show that sentences containing words from studied sentences are discriminable early in processing from sentences containing all new words, but discrimination of correct from incorrect versions of studied sentences occurs only later in processing (after 600-700 ms). These results demonstrate that different kinds of information are available at different points during the time course of retrieval and so suggest that modifications are required of models that provide only a unitary value for the amount of match between a test probe and information in memory. Early in processing, the growth of accuracy can be explained by a simple model that assumes independent contributions to total amount of match for each of the content words of a sentence, but this independent processing model cannot account for discrimination later in processing. Several, more general, memory models are examined with respect to their abilities to produce independent item information early in processing and relational information later in processing.     

WORKING MEMORY CAPACITY DISSOCIATES LEXICAL AND SENTENTIAL CONTEXT EFFECTS

Abstract— Semantically associated and unassociated word pairs were embedded in normal meaningful sentences and in sentences that were semantically anomalous throughout The influence of lexical context was isolated via comparison of responses to the second words of the associated and unassociated pairs. The influence of sentence-level context was isolated by comparing responses to the same words in the two sentence types Subjects of high, medium and low working memory capacity (as evaluated by the reading span test) showed modulations of event-related brain potentials in response so lexical context. In contrast, only the high- and medium-capacity groups were responsive to purely sentence-level semantic context. The results demonstrate that sentential context influences the processing of words in intermediate sentence positions at normal reading speeds but that the on-line utilization of this context is more demanding of working memory than single-word contexts.     

Taboo words: The effect of emotion on memory for peripheral information

In three experiments, we examined memory for peripheral information that occurred in the same context as emotion-inducing information. In the first two experiments, participants studied either a sentence (Experiment 1) or a pair of words (Experiments 2A—2C) containing a neutral peripheral word, as well as a neutral, negative-valence, or taboo word, to induce an emotional response. At retrieval, the participants were asked to recall the neutral peripheral word from a sentence fragment or emotion-inducing word cue. In Experiment 3, we presented word pairs at encoding and tested memory with associative recognition. In all three experiments, memory for peripheral words was enhanced when it was encoded in the presence of emotionally arousing taboo words but not when it was encoded in the presence of words that were only negative in valence. These data are consistent with priority-binding theory (MacKay et al., 2004) and inconsistent with the attention-narrowing hypothesis (Easterbrook, 1959), as well as with object-based binding theory (Mather, 2007).     

语义相关性对副中央凹-中央凹效应的影响

阅读时词汇加工的注意分配是序列的还是平行的,是所有眼动控制模型需解答的关键问题。对此,可通过分析副中央窝-中央窝效应（POF效应）进行解释。以往研究结果的争议在于POF效应是否为误差眼跳。本研究为控制眼跳误差,采用分心阅读范式,考察词n与词n－1的语义相关性对词n－1加工的影响,探讨中文阅读的POF效应。实验以26名大学生为被试,选取高频双字词为词n－1,低频双字词为词n。结果发现当词n－1与词n语义相关性高时,词n－1的凝视时间更短,再注视率更低,出现同向的POF效应。本研究在一定条件下支持SWIFT模型为代表的平行加工假设。     

Grammatical structure and recall: A function of the space in immediate memory or of recall delay?

Savin and Perchonock suggest that when a sentence and additional unrelated words are presented for immediate recall, the number of words recalled in addition to that sentence represents the space left over in a limited memory store after the sentence and its transformational tag have been stored. They report error data and word-recall data in support of a transformational model of sentence processing. In two replications of that study, error data failed to support the transformational model. Furthermore, the word-recall data could be attributed to differential word-recall delays-the more complex the sentence, the longer the delay between word presentation and word recall. These findings, as well as others in the literature. fail to support a transformational interpretation of sentence processing.     

Inhibitory Control During Sentence Reading in Dyslexic Children

The present study focused on the nature of the reading disability of children with the guessing subtype of dyslexia (who read fast and inaccurately). The objective was to separate the excitatory account of their reading disturbance (i.e., in guessers the words’ resting levels of activation are oversensitive to semantic context) from the inhibitory account (i.e., guessers tend to react prematurely to (false) candidate words that are activated in the lexicon).

To disentangle the above accounts, guessers and normal readers were presented with a sentential priming task (SPT). In the SPT, subjects had to determine whether the final word of a sentence was semantically congruent or incongruent with the sentence, but had to inhibit their ‘congruent’ or ‘incongruent’ response in case of an occasionally presented pseudoword. To evoke guessing, each pseudoword closely resembled either a valid congruent or incongruent word. Guessing referred to prematurely accepting a pseudoword as a word that either appropriately or inappropriately completed the sentence. The extent to which subjects guessed at word meaning was evidenced by the false recognition rates (FRR) of the misspelled terminal words.

Analyses on the FRRs of the pseudowords showed that guessers had significantly more difficulty in suppressing the ‘go tendency’ triggered by the pseudowords. It was concluded that the impulsive reading style of guessers should be ascribed to a less efficient suppression mechanism rather than to excessive reliance on contextual information. Specifically, the data were explained by assuming that the availability of the pseudoword’s candidate meaning activated the hand to respond with, and that guessers found difficulty in suspending this response until they analyzed all letters in the stimulus and they could be sure of its spelling.     

The influence of contextual constraints on recall for words within sentences

Two experiments examined predictions of the relational-distinctive processing view to account for the influence of sentence constraints on memory of target words. In Experiment 1, congruous expected words were recalled better than incongruous words. Words appearing in high-constraint sentences were recalled better than words appearing in low-constraint sentences. In Experiment 2, word expectancy and sentence constraint interacted so that unexpected congruous words appearing in high-constraint sentences were recalled better than their expected counterparts, but this difference was not present for low-constraint sentences. A hybrid model including aspects of the featural restriction model of sentence constraint and the relational-distinctive processing view is proposed.     

Direct Evidence of Memory Retrieval as a Source of Difficulty in Non‐Local Dependencies in Language

Linguistic dependencies between non‐adjacent words have been shown to cause comprehension difficulty, compared with local dependencies. According to one class of sentence comprehension accounts, non‐local dependencies are difficult because they require the retrieval of the first dependent from memory when the second dependent is encountered. According to these memory‐based accounts, making the first dependent accessible at the time when the second dependent is encountered should help alleviate the difficulty associated with the processing of non‐local dependencies. In a dual‐task paradigm, participants read sentences that did or did not contain a non‐local dependency (i.e., object‐ and subject‐extracted cleft constructions) while simultaneously remembering a word. The memory task was aimed at making the word held in memory accessible throughout the sentence. In an object‐extracted cleft (e.g., It was Ellen whom John consulted…), the object (Ellen) must be retrieved from memory when consulted is encountered. In the critical manipulation, the memory word was identical to the verb's object (ELLEN). In these conditions, the extraction effect was reduced in the comprehension accuracy data and eliminated in the reading time data. These results add to the body of evidence supporting memory‐based accounts of syntactic complexity.     

Semantic similarity, predictability, and models of sentence processing

The effects of word predictability and shared semantic similarity between a target word and other words that could have taken its place in a sentence on language comprehension are investigated using data from a reading time study, a sentence completion study, and linear mixed-effects regression modeling. We find that processing is facilitated if the different possible words that could occur in a given context are semantically similar to each other, meaning that processing is affected not only by the nature of the words that do occur, but also the relationships between the words that do occur and those that could have occurred. We discuss possible causes of the semantic similarity effect and point to possible limitations of using probability as a model of cognitive effort.     

Presentation format and its effect on working memory

In three experiments, we examined the separate cognitive demands of processing and storage in working memory and looked at how effective the coordination was when items for storage varied in format/modality. A sentence verification task involving arithmetic facts was combined with a span task involving two to six items presented in picture, printed word, or spoken word format. The first two experiments were the same, except for the added requirement of articulation of the math sentence in Experiment 2. Experiment 3 varied the length of the span item and compared recall with recognition performance. The results showed that both spoken words and picture produced superior recall and recognition, as compared with printed words, and are consistent with Baddeley and Logie's (1999) and Mayer's (2001) models of working memory. Also, the differences in processing performance across spans varied with the difficulty of the task but showed the strongest support for the resource allocation model (Foos 1995).