Lossy-Context Surprisal: An Information-Theoretic Model of Memory Effects in Sentence Processing

A key component of research on human sentence processing is to characterize the processing difficulty associated with the comprehension of words in context. Models that explain and predict this difficulty can be broadly divided into two kinds, expectation-based and memory-based. In this work, we present a new model of incremental sentence processing difficulty that unifies and extends key features of both kinds of models. Our model, lossy-context surprisal, holds that the processing difficulty at a word in context is proportional to the surprisal of the word given a lossy memory representation of the context—that is, a memory representation that does not contain complete information about previous words. We show that this model provides an intuitive explanation for an outstanding puzzle involving interactions of memory and expectations: language-dependent structural forgetting, where the effects of memory on sentence processing appear to be moderated by language statistics. Furthermore, we demonstrate that dependency locality effects, a signature prediction of memory-based theories, can be derived from lossy-context surprisal as a special case of a novel, more general principle called information locality.     

阅读中文时信息整合的即时性

通过两个实验对中文语篇理解中相关背景信息的激活与整合到底是即时发生抑或延时进行这一问题进行了探讨。研究所用的阅读材料都包含一个目标句,它与局部语境一致,但与前文提供的背景信息冲突或无关。实验一采用自定步速逐行阅读技术,考察了一致性条件以及句子结束的标记（即句号）是否存在对目标句阅读时间的影响。结果发现,句子结束的标记是否存在并不影响目标句整合加工的进行。实验二运用眼动追踪技术着重探讨信息整合加工能否在目标句关键区中即时进行,实验条件仍为当前信息与背景信息的一致性,重点考察其对目标句中关键前区、关键区及关键后区的作用。结果表明,自变量的操纵导致关键区中反映初次加工的指标产生可靠的效应。这一结果表明中文语篇理解中整合加工的过程开始得非常迅速,是一个即时引发、而非延时进行的过程     

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.     

Assessing resource demands during sentence processing in Parkinson's disease

Patients with Parkinson's disease (PD) have sentence comprehension difficulty, but it is unclear whether this is due to a deficit in grammatical processing or to an executive resource limitation. To assess grammatical processing in PD while minimizing task-related demands, PD patients and healthy control subjects performed a word detection procedure that assesses sensitivity to grammatical agreements in sentences in an "on-line" fashion. With this technique, we found that control subjects and PD patients are equally sensitive to grammatical agreement violations in sentences. A traditional, resource-demanding measure of sentence comprehension was also administered to the same PD patients. In comparison to healthy controls, PD patients were significantly impaired in their relative comprehension of sentences containing object-gap subordinate clauses compared to subject-gap subordinate clauses. Performance on several executive resource measures was also impaired in PD, and this correlated with their comprehension performance. Sensitivity to grammatical agreements with the word detection procedure, in the context of sentence comprehension difficulty on a traditional measure, suggests that PD patients' executive resource limitations contribute to their sentence comprehension difficulty.     

语篇语境对句子理解的影响：来自N400的证据

N400作为语义违反的特异性ERP组分,其波幅反映了语义整合加工过程的难度,因此可将其作为研究语篇语境意义建构过程的一项重要指标。语篇语境可分为语言语境和非语言语境。前人一系列基于语篇水平的N400研究表明,从语义的角度上看,语言语境对句子理解具有制约和协调作用,其机制涉及词汇语义启动、场景适宜性、文字内容信息以及因果推理等方面;非语言语境对句子理解具有制约和补充作用,但其机制尚不明确。两种不同来源的语义信息在实时理解中是同时得到加工的     

Producing Pronouns and Definite Noun Phrases: Do Speakers Use the Addressee’s Discourse Model?

We report two experiments that investigated the widely held assumption that speakers use the addressee's discourse model when choosing referring expressions (e.g., Ariel, 1990; Chafe, 1994; Givón, 1983; Prince, 1985), by manipulating whether the addressee could hear the immediately preceding linguistic context. Experiment 1 showed that speakers increased pronoun use (and decreased noun phrase use) when the referent was mentioned in the immediately preceding sentence compared to when it was not, even though the addressee did not hear the preceding sentence, indicating that speakers used their own, privileged discourse model when choosing referring expressions. The same pattern of results was found in Experiment 2. Speakers produced more pronouns when the immediately preceding sentence mentioned the referent than when it mentioned a referential competitor, regardless of whether the sentence was shared with their addressee. Thus, we conclude that choice of referring expression is determined by the referent's accessibility in the speaker's own discourse model rather than the addressee's.     

The time-course of lexical and structural processes in sentence comprehension

Online sentence comprehension involves multiple types of cognitive processes: lexical processes such as lexical access, which call on the user's knowledge of the meaning of words in the language, and structural processes such as the integration of incoming words into an emerging representation. In this article, we investigate the temporal dynamics of lexical access and syntactic integration. Unlike much previous work that has relied on temporary ambiguity to investigate this question, we manipulate the frequency of the verb in unambiguous structures involving a well-studied syntactic complexity manipulation (subject- vs. object-extracted clefts). The results demonstrate that for high-frequency verbs, the difficulty of reading a more structurally complex object-extracted cleft structure relative to a less structurally complex subject-extracted cleft structure is largely experienced in the cleft region, whereas for low-frequency verbs this difficulty is largely experienced in the postcleft region. We interpret these results as evidence that some stages of structural processing follow lexical processing. Furthermore, we find evidence that structural processing may be delayed if lexical processing is costly, and that the delay is proportional to the difficulty of the lexical process. Implications of these results for contemporary accounts of sentence comprehension are discussed.     

Event-related potential indices of ambiguous sentence processing

An event-related potential (ERP) probe was used to examine various models of ambiguous sentence processing. ERPs to light flashes were recorded during and immediately after auditorily presented ambiguous and unambiguous target sentences. Each target sentence was preceded by either a relevant or a neutral context sentence. Principal component analyses of the ERPs indicated that although certain components varied as a function of ambiguity, none of the components varied as a function of preceding context. These findings provided some support for a postdecision model of ambiguity processing which suggests that both meanings of an ambiguity are always processed, even when prior disambiguating context is available.     

Semantic activation of noun concepts in context

A modified Stoop procedure was used to examine the role that context plays in guiding semantic access of unambiguous nouns in sentence contexts. The sentences either emphasized a high- or a low-dominant property of a noun that was the last word in the sentence or were control sentences. Each sentence was followed by the relevant high- or low-dominant property either immediately or after a 300-or 600-ms delay. There was significant color-naming interference (relative to control) for high-dominant properties regardless of biasing context in the immediate and delayed conditions. There was also significant color-naming interference for low-dominant properties in the immediate condition regardless of context. However, in the delayed conditions, the low-dominant properties led to color-naming interference only when preceded by sentence contexts biasing interpretation toward the low-dominant property. It was concluded that high-dominant properties function as core, or invariant, aspects of meaning and that initial semantic access is context independent.     

A longitudinal study of sentence context effects in second-grade children: Tests of an interactive-compensatory model

During the first half of the school year and at the end of the school year second-grade children (mean age: 7 years, 4 months at the first testing) completed a task in which they read words preceded by either a congruous sentence context, an incongruous sentence context, or a neutral context. Prior to the first testing, each child was given practice at recognizing one-half of the words in isolation. Word difficulty was varied orthogonally with practice and context condition. The effect of context on reading times decreased with development and practice, and increased with word difficulty. The results were interpreted as supporting an interactive-compensatory model of the development of reading fluency.     

语篇背景在语义整合中的作用

通过眼动实验考察了语篇背景对不同类型句子语义整合过程的影响。句子类型有3个水平：(1)宾语违反了动词的选择性限制; (2)宾语符合选择性限制, 但是违反了世界知识; (3)无违反。语篇背景有2个水平：(1)中性背景; (2)校正背景。在中性语篇背景下, 句子(1) (2)是不合适的; 而在校正背景下, 三种句子都是合理的。24名大学生被试自然阅读的眼动数据表明, 违反动词选择性限制的目标词上的首次注视时间、回视路径时间和总注视时间, 在中性背景下比在校正背景下长, 注视点个数也更多; 而在非选择性限制违反条件下, 语篇背景的效应只在总注视时间和注视点个数两个指标上显著。这些结果表明, 当句子中的宾语违反了动词的选择性限制时, 语篇背景影响了宾语的早期和晚期加工。而当宾语只违反世界知识时, 语篇背景的作用体现在晚期阶段。语篇背景对当前信息的整合是动态的过程, 并不是只能在固定某个阶段起作用。     

What this construction needs is generalized

What is the nature of the knowledge that readers acquire as they learn to comprehend a novel construction in their native language? This question was explored in two experiments that examined the extent to which readers who learn a new construction are able to extend it to sentence contexts other than the one in which it initially appeared. Readers were trained on the needs construction (“The meal needs cooked given that dinner is in an hour”) and then were presented with examples of the construction in a different sentential context (“John thinks that what the meal needs is cooked given that dinner is in an hour”). The results show that readers have little difficulty comprehending the needs construction in the new context, even when the construction is used with a different verb. These findings are discussed with respect to theories of structural priming and sentence processing.     

The generalizability of context effects on word recognition: A reconsideration of the roles of parafoveal priming and sentence context

In three experiments, subjects named target words preceded by congruous, incongruous, or neutral sentence contexts. There was no evidence that the recognition of the target word was affected by the semantic characteristics of a word presented immediately to the right of it. The nature of the preceding sentence context did affect target-naming speed. However, the magnitude of the context effect was considerably smaller in these experiments, in which nonterminal target words were used, than in previous experiments in which the target word was always the final word of the sentence, was highly predictable from the context, and was often semantically related to words in the sentence. The implications of these two findings for theories of reading and context effects are discussed.     

Word-finding as a function of stimulus context: Children compared with aphasic adults

A study of object naming in 202 children, aged 5–11 years, demonstrated that varying the stimulus context affected the level of response accuracy. From age 6 to 10, completing a sentence with a noun (auditory) yielded the lowest error scores, naming objects upon hearing a definition (also auditory) produced most errors, while responding with names to pictured objects was of intermediate difficulty. This is the same order of object naming difficulty found in adult aphasic patients. Children older than 10 do not appear to be affected by stimulus context in their object naming. Results are discussed in terms of the relative syntactic difficulty of the two auditory conditions, the unexpected sex differences, and the “aphaseoid” naming errors of children.     

Memory connections between thematically similar episodes

Recent theories about the representation of thematic information in memory propose that two episodes that share a theme are connected together through a thematic structure. We investigated the use of such cross-episode connections in comprehension and memory in six experiments. Experiments 1 and 2 used a priming technique; it was found that verification time for a test sentence from one story was speeded by an immediately preceding test sentence from a thematically similar story but only when subjects were given instructions to rate the similarities of the stories. In the remaining experiments, a single test sentence was presented immediately after a story was read, with timing controlled by presenting the story one word at a time. Response time for a test sentence from a previously read story was facilitated if the immediately preceding story was thematically similar, but only if the previously read story was extensively prestudied. We conclude that, during reading of an episode, thematic information may be encoded so as to lead to activation of similar episodes and formation of connections in memory between episodes, but such encoding is not automatic and depends on subjects' strategies and task difficulty.     

Sentence context modifies compound word recognition: Evidence from eye movements

Two eye movement experiments are reported that examine the influence of sentence context on morphological processing. English compound words which vary in beginning lexeme frequency (Experiment 1) and ending lexeme frequency (Experiment 2) were embedded into sentence contexts that were either predictive of the compound word or were neutral with respect to the compound. A predictable sentence context reduced the effect of beginning lexeme frequency on first fixation and single fixation durations. However, sentence context did not modify effects of beginning and ending lexeme frequency in later fixation measures. These results further support the theoretical position that morphology plays a role at multiple levels within readers' mental lexicons. In addition, these results suggest that access to early morpho-orthographic processes can be influenced by sentence context, a finding that suggests an interactive relationship between sentence context and word recognition.     

A distinction between the effects of sentential speaking rate and semantic congruity on word identification

It is well known that numerous aspects of sentential context can influence the manner in which a word within the sentence is identified. We investigated two such contextual effects, that of the speaking rate of the sentence in which the target word occurs and that of the semantic congruence between the sentence and the target word. We observed that although the two effects are similar on the surface, in that each is realized as a change in the identification of acoustically ambiguous (but not unambiguous) items along a speech series, they are strikingly different in their susceptibility to changes in task demands. Specifically, changes in the task that readily eliminate the semantic congruity effect do not serve to eliminate the rate effect, suggesting that the two effects arise at different stages of analysis. The implications of this finding for models of speech processing are discussed.     

汉语歧义句的加工

通过两个实验,对汉语歧义句多种解释的意义频率和语境位置在句子歧义解析过程中的效应进行了考察。实验结果表明：（１）前语境为被试提供了某种预期信息,它对句子歧义解析的效应大于后语境;（２）在句子歧义解析过程中,不管有语境或无语境,不管是前语境或后语境,被试提取歧义句的主要意思比提取次要意思快;（３）歧义句的多种解释的激活,有一个时间进程,在短延迟间隔条件下,歧义句的主要意思可以被激活并形成表征,但次要意思的激活和表征则需要较长的时间     

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.     

工作记忆、语境限制强度和句子长度对二语词汇学习的影响

本研究选取初二、高二、大三三个年龄段的被试,考察在句子阅读过程中,工作记忆容量、语境限制强度和句子长度对第二语言词汇学习的影响。研究结果发现,在句子阅读条件下,工作记忆容量是影响第二语言词汇学习的重要因素,高工作记忆容量促进了第二语言词汇的学习;在中等和长句子的条件下,语境的限制性越强,词汇学习的成绩越好;相比初二年龄组被试,高二和大三年龄组被试,在低语境限制性的情况下,阅读的句子越长,第二语言词汇学习效果越差。研究结果说明,个体工作记忆的容量,以及阅读材料等变量,都是影响自然阅读条件下二语词汇学习的重要因素。