Bookshelf 📖

Structural and semantic constraints on the resolution of pronouns and reflexives (Kaiser et al., 2009)

Citation

Kaiser, E., Runner, J. T., Sussman, R. S., & Tanenhaus, M. K. (2009). Structural and semantic constraints on the resolution of pronouns and reflexives. Cognition, 112(1), 55-80.

My thoughts

• Whether the source/perceiver should be a semantic, thematic role or a discourse-level/pragmatic manipulation
• the mutilple constraint hypothesis = different weights for constraints and the anaphor resolution is not infleucned by one singles constriant but interactions ebtween multiple constraint

Summary

总结

• 在“picture NP”（如 Andrew’s picture of him/himself）里，代词和反身词的指称不是单靠结构说了算；语义/话语因素（谁是信息的“来源”source、谁是“感知者”perceiver）也会早早介入，而且两类指称形式对这些约束的权重不一样
• 反身词更吃结构，代词结构与语义势均力敌。
• 在非属格的 picture-NP 里（Experiments 1/2b），反身词对source有稳定但偏弱的偏好：眼动里大约在200–250 ms就冒头；更晚的1400–1800 ms时间窗里，动词操纵带来显著效应，直接支持“反身词偏 source、代词偏 perceiver”这套说法。
• 在反身词条件下，当宾语是 source（heard）时，被试看宾语图的比例更高于 told，F(1,8)=6.44, p<.05。也就是说，source 身份会把注意力往“非结构偏好”的方向拽一点。
• 到有属格的 picture-NP（Experiment 3）：反身词几乎被possessor 约束钉死（选择里 possessor ≈90–92%），也明说在这种结构下source 约束太弱，基本检测不到。
  • 属格结构： 属格结构就是名词短语里带“谁的”那个壳。句法上是 genitive/possessor，表示所属或与名词的紧密关系。X’s N（Lisa’s picture），或 the N of X（the picture of Lisa）。

方法总览（操纵与任务）

• 操纵：用动词切换信息角色
• told：主语=source，对象=perceiver；heard (from)：主语=perceiver，对象=source。既测无属格PNP（Experiments 1, 2a, 2b），也测有属格PNP（Experiment 3）。 任务：
• E1 图片核对（yes/no）；E2a 纸面二选一；E2b 眼动+二选一；E3 眼动+三选一（主语图片/宾语图片/属主图片）。

实验分解

• Experiment 1（无属格 PNP，图片核对）
• 被试：24。材料：每屏两个人物+中间一张人物照片；交叉动词(told/heard) × 指称形式(him/himself) × 照片是谁(主/宾)。

结论（方向）

• 反身词偏向主语；代词偏向宾语（结构主效应）。
• 动词操纵有效：反身词对source有偏好、代词对perceiver有偏好。

关键统计

• 主/宾图片主效应：b = −0.82, Z = −3.9, p < .001。
• 主/宾 × 指称形式：b = 5.52, Z = 12.94, p < .001。
• 代词：宾像 > 主像，b = 1.89, Z = 7.44, p < .001。
• 反身词：主像 > 宾像，b = −3.50, Z = −10.5, p < .001。
• 三因子交互（主/宾×指称×动词）显著：b = 3.84, Z = 4.6, p < .001。

以“结构期望/非期望”回应为因变量： • 指称主效应：b = −0.80, Z = −3.92, p < .001（代词更容易给出结构上“非期望”的主语解释）。 • 动词主效应：b = 0.95, Z = 4.62, p < .001。 • 代词的动词效应强：b = 1.23, Z = 4.74, p < .001；反身词也有较弱效应：b = 0.88, Z = 2.41, p < .05。

• 小翻译：反身词主要听语法（主语约束），但也会被“谁是信息来源”轻飘飘拽一下；代词则被“别指本句主语”的结构约束和“找感知者”的语义约束拉扯得旗鼓相当。

Experiment 2a（无属格 PNP，纸面二选一）

• 被试：24。任务：圈出句子里提到的那张照片（主/宾两张同时在场）。

结论（方向）

• 反身词仍强主语偏好；代词出现清晰的perceiver偏好。

关键统计与比例

• 反身词：总体89.6%选主像；选宾像 heard 14% vs told 7%。
• 代词：总体55%选宾像；主像 told 31.7% vs heard 58%。
• 主效应：指称 b = −2.22, Z = −7.99, p < .001；动词 b = 1.09, Z = 3.96, p < .001；交互不显著。
• 条件内：代词的动词效应 b = 1.52, Z = 4.80, p < .001；反身词的动词效应边缘显著 b = 0.82, Z = 1.822, p = .069。

直白点：把两种解释硬扔到你面前时，反身词几乎“铁头”指主语；代词会明显受“谁在感知”驱动。

Experiment 2b（无属格 PNP，眼动+二选一）

• 被试：16。指标：主像优势分（看主像比例−看宾像比例），按时间窗分析。

结论（时间进程）

• 反身词：主像优势从一开始就有；对source的偏好大约在200–250 ms后冒头，600 ms暂时减弱，约1200 ms再增强。

• 代词：基本相反的轨迹，后段出现perceiver偏好。1000–1400 ms 动词效应边缘显著，1400–1800 ms 动词主效应显著；计划比较显示：代词条件动词效应 F(1,8)=29.67, p=.001；反身词 F(1,8)=7.65, p<.05。两类形式对 source/perceiver 的早期敏感性约200 ms后就能看到。

• 动作依赖分析（只看最后选了结构偏好图的试次）：依然能看到动词的语义效应，说明语义不是“结构失败才启用的补救策略”。
• 总结：眼动把话挑明了——语义约束不是晚到的救火队，开口就掺和，而且代词/反身词的“权重表”从早期就不一样。

Experiment 3（有属格 PNP，眼动+三选一）

• 范式：句子含属主（如 Lisa’s picture of her/herself），屏幕有主/宾/属主三张图。

结论（方向）

• 反身词：几乎被属主约束死死拽住；动词操纵（source/perceiver）基本不起作用；仍有少量主语选择（两动词约 8–10%），显示主语约束的弱痕迹。
• 代词：清晰的反属主结构约束 + perceiver语义效应；眼动里先有短暂“看属主”的早期峰值，真正的 perciever 效应要到约 1400 ms才显著。

关键统计（选择与眼动） 选择： • 选属主图：指称主效应 b = −9.46, Z = −6.84, p < .001（反身词≫代词）。 • 选主图：指称主效应 b = 2.79, Z = 6.45, p < .001；动词主效应 b = −1.54, Z = −3.64, p < .001；动词×指称交互 b = −3.52, Z = −4.15, p < .001（代词在 heard 条件下主像更多）。 • 选宾图：反身词几乎为零（heard 0%，told 0.4%）。

• 眼动：反身词比代词显著更多看“属主图”（时间窗 1400–1800 ms：F(1,8)=31.55, p=.001；总体 p<.001）；代词条件下早期“看属主”很短暂，约 400 ms 后下滑，随后才出现晚到的 perceiver 效应。

粗暴解读：有属格时，反身词几乎“写着属主的名字”；代词先被属主的视觉/提及近邻性晃一下眼，真正在意“谁在感知”得等一会儿。

理论贡献

• 形式特异的多重约束：代词与反身词对结构/语义的权重不同；不能用“所有结构约束同权”的一刀切模型。
• 反身词更“结构主义”，代词更“折中”：在无属格 PNP，反身词主要受主语约束，source 只是调味；代词的反主语结构约束和 perceiver 语义约束势均力敌。有属格 PNP 里，属主约束（反身词）和反属主约束（代词）比对应的主语/反主语约束还强。
• 时间进程证据反对“先语法后语义”的两阶段模型：source/perceiver 效应约 200 ms 就出现，说明处理是并行整合多约束，而非“先绑定再修复”。

话糙理不糙的小结

• 反身词：基本“向结构投诚”，source 说两句管点用。
• 代词：一边嫌主语太近（结构），一边追着感知者跑（语义），摇摆但高效。
• 有属格时，属主的存在像“场上灯塔”，把反身词牢牢吸走，把代词先晃眼再放人。

They conducted four experiments on the interpretation of pronouns and reflexives in picture noun phrases with and without possessors, and 2 offline and 2 visual world eye-tracking

• They found that the interpretation of pronouns and reflexives in the picture NP condition is sensitive not to purely structural information, but also to semantic information
• Pronouns and reflexives differ in the degree of sensitivity they exhibit to different kinds of information
• Indicate a form-specific multiple-constraints appraoch
• referential forms have asymmetrical sensisitives to the different constraints guiding reference resolution and also applies within-sentence domain

Andrew’s picture of him/himself The picture of him/himself

2 offline 2 visual world

Research Questions

• how syntactic and semantic factors will guide the interpretation of pronouns and reflexives
• How is different information integrated during real-time reference resolution

Introduction

• It’s common to assume that the choice of the referent for a reflexive like herself is specified by the structural configuration of the sentence
• Many research have argued that semantic informaiton infleucnes the interpreation of pronouns with cross-clasual antecedent
• Stevenson and colleagues (Stevenson et al., 1994) found that the thematic role of an enity (agent, goal, source, experiencer) infleunces its likelyhood of being the antecedent for a subsequent pronoun (also see Arnold, 2001)
• which suggested that referential expressions tend to fall into two groups, one is the strucutre princple and another one is the non-strcuture factors such as semantics and discourse.
• multiple-constraints frame-work, first introduced in Kaiser (2003) (see also Brown-Schmidt, Byron, & Tanenhaus, 2005; Kaiser, 2005; Kaiser & Trueswell, 2008)
  • The anaphor resolution is not determined by a single constraint but rather is the result of the interaction of multiple constraints
  • allow for multiple constraints that guide reference resolution to be weighted differently for different referntial forms
  • example: Finnish pronouns and demonstratives can be used to refer to human antecedent mentioned in the preceding clause, but do not show the same level of sensitity to the antecedent’s syntactic role and linear position

Picture NP

• the reflelxive and pronouns in the picture NP are exempt from the binding principle, and they should be taken account with semantics and discrouse roles
• Reflexives in possessed PNPs can refer to the subject of the sentence, which is contrary to the predicitions of the strucutral appraoch
• 2 appraoches
• Agument Binding Theory in order to maintain a primarily strcutural account of the distribution of pronouns and refelxvies
• treat some anaphors (especially reflexives in possessorless PNPs) are being exempt from Binding Theory altoghether and guided by semantic and discrouse constraints

1.2. Semantic and discourse factors

• Reflexives in possessorless PNPs have been argued to be sensitive to factors such as the potential antecedents’ point of view, degree of awareness and semantic role (e.g., Kuno, 1987; Pollard & Sag, 1992).
• Kuno, 1987 mentioned that reflexive can felicitously refer to Mary becuase she is the one who provides John the informaiton, which makes Mary of source of the sentence, and pronouns and refelxvies are both sensitive to the point of view, which means they have a preference for antecedents that are either source or the perceivers of informaiton.
• a purely structural approach is insufficient to capture their referential properties, especially in pos-sessorless PNPs.

1.3. General aims of this paper

1. how strcutural and semantic constrqints infleucnes pariticiapnts’ final interpreations of pronouns and reflexvies
2. the time-course with which different constraints infleucne the processing of these forms
3. complement and extend exisiting work by testing whether the source and perceiver hypotheses are supported by experimental data for the on-line interpreation of posseorless PNP as well as for PNP
4. test the asymmetrical sensitvitiy that the form-specific multiple-constraints appraoch permits also applies to the within-sentence domain and whether strcutural and semantic constraints show wualitativiely dfferet behavior during real-time processing
5. the relative weights of syntactic and semantic informaiton: how much of an infleucne does semantic informaiton exert on the interpretation of pronouns and refelxvies, as compared to the infleucne of syntactic information?

(7a) Peter told John about the picture of himself/him on the wall (7b) Peter heard from John about the picture of himself/ him on the wall

• create situations in which structural constraints and non-structure constraints are pitted against each other as well as they are aligned and favor the same antecedent

1.4. Nature of the relation between structural and semantic constraints

• If both syntactic and semantic constraints play a role, and if their relative weights are the same for pronouns as for reflexives, then we predict that in sentences like (7a) and (7b):

(i) with pronouns, there will be more subject interpretations with hear than with tell (based on the perceiver preference hypothesis in (6)) (ii) with reflexives, there will be more object interpretations with hear than with tell (based on the source preference hypothesis in (4)) (iii) if the relative weights of structural and semantic information are the same for pronouns and reflexives, the proportion of subject and object choices triggered by one form should be ‘mirrored’ by the proportion of subject and object choices triggered by the other form. In other words, whatever levels of sensitivity reflexives show to structural and to semantic information, pronouns will show the same levels of sensitivity.

• real-time language processing is continuously guided by multiple weighted constraints
• we expect to see early effects of both structural and semantic constraints and can observe when potential asymmetries in the constraint weights for pronouns and reflexives emerge during real-time processing

Method

2. Experiment 1: picture verification

• 5 male and 5 female characters
• participants were familiarized with the names of the characters
• a total of 32 target items and 32 filler items
• target items, crossed verb type (told/heard), anaphoric form (himself/him), and visual display (picture of subject/picture of object)
  • 8 conditions

Predictions for Experiment 1

• If structural constraints (the subject constraint for reflexives and the anti-subject constraint for ponouns) fully determine reference resolution, differences in verb semantics should not lead to differences in anaphor resolution in picture NPs
• the preferred antecedents for pronouns and reflexvies should remain the same regardless of whether the sentence contains hear or tell
• If semantic constraints modulate the effects of structural constraints for both pronouns and reflexives, then a main effect of the verb conditions is predicted.
• whereas the structural and semantic constraints converge in the tell conditions, they conflict in the hear conditions
• The magnitude of the verb effect depends on the weights of the structural and semantic constraints. If the weights of structural and semantic constraints are the same, then the prediction is that the magnitude of the verb effect will be the same for reflexives and pronouns
• If the form-specific multiple-constraints approach applies in the within-clause domain = then syntactic and semantic constraints might not be weighted equally for pronouns and reflexives

Results and discussion

• mixed-effect logistic regression
• reflexvie are referring to the subject, across verb types, there were more yes answers when the subject is pictured than object is pictured
• the proportion of subject and object choices was modulated by the verb manipulation
• pronoun referring to the objec, across verbs, there were more yes answers when the object is pictured than subject is pictured
• mixed-effect regression to analyze the proportion of yes answers as a function of anaphor type, verb and picture with paritcipant and item as random effects
  • significant main effect of picture
  • significant picture x anaphor interaction
  • no main effect of anaphor
  • significant picture x anaphor x verb interaction
    • shown that with a picture of the object, there were more yes answers with tell than hear, but when a picture of the subject was shown, there were more yes answers with hear than tell
    • weaker but significant picture x verb interaction in the reflexvie conditions
    • with a picture of the object, there were more yes responses with hear than tell, but when a picture of subject was shown, the verb manipulation did not have a very strong effect
• Overall, the results of experiment 1 showed signficiant effect of strcuture, with pronouns perferring obejcts and reflexvies preferrring subejcts
• verb manipulation also had an effect on both anaphoric forms, with reflexvies preferring sources and pronoun preferring perceivers
• verb manipulation had a greater effect on pronouns than on reflexvies

Experiment 2: picture choosing

• off-line picture-choosing experiment to investigate the effects of a design switch from picture verification to picture choice
  • aim: whether the effects of the source/perceiver verb manipulation presist in a context where people need to choose betweent two directly competing interpreations
• visual world eye-tracking paradigm

Predictions

• (i) the interpretation of reflexives will be guided by a strong subject constraint, modulated by a relatively weaker source constraint
• (ii) the interpretation of pronouns is guided by two more evenly-matched constraints, i.e., the anti-subject constraint and the perceiver constraint
• In the reflexive conditions, participants are predicted to generally choose the subject picture over the object picture, but to choose the object picture more often with hear than tell.

Experiment 2a Results and Discussion

• two main effect: anaphor = reflexives triggering more structurally-expcted choices than pronouns
• main effect of verb = source.perceiver manipulation infleucnes paritipants picture choices
• no significant anaphor-verb interaction
• indicating the degree of sensitivity to the verb manipulation does not differ for the two anaphor types
• in pronoun conditions, paritiapants’ choice of subject vs object was strongly infleucned by the verb

Experiment 2b: eye tracking of possessorless PNP processing

Conclusion and discussion

• set out to investigate how different kinds of information interact during reference resolution
• at least in picture no without possessors, pronouns exhibit a preference for perceivers of infroamtion, and refelxvies for sources of information
• the interpretation of pronouns and refelxives is guided by both strcutural and semantic constraints
• pronouns and reflexvies differ from each other in the degree of sensitivity they exhibit to structural and semantic constraints
• The interpretation of regfelxvis is guided by a structural subject constraint that is weighted more heavily than a semantic source constraint
• eye-movement patterns shown that both strcutural and semantic constriants infleucne processing of pronouns and reflexvies very early on, and semantic constraints have an effect even when pariticpants interpret the anaphoric form in the strually preferred manner
• whehther constraints belonging to a pariticualr linguistic class (strucutral vs semantic) are all weighted the same?
  • for reflexvies, the strucural possessor constraint outranks both the subject constraint and source constraint
  • for pronouns, we saw fairly even competition between strcutural and semantic constraints, but experiment 3 reveals a situation where the strutual anti-possesoor cosntraint is stronger than the semantic source constriant and the struccutal anti-subject constraint
  • both pronouns and refelxvies can be subejct to powerful strctural constraints, depending on the structural configuration
  • different strcutural constraints can be weighted differently, suggesting that the lingusitic status of a constraint is not what determines its weight
• our results suggest that anaphor resolution is guided by multiple factors which are weighted differently for different anaphoric forms, we leave open the question of how this is realized in the processing system.
• source and preciver It may also be possible to connect the source vs. perceiver effects to the notion of point-of-view, something that has been argued to play a crucial role in guiding the use and interpretation of reflexives
• Off-line experiments on PNPs in German and Dutch reveal striking parallels with English: Reflexives are subject to a strong subject preference, modulated by a weaker source preference, whereas pronouns show a stronger sensitivity to perceiver of information (Kaiser & Runner, 2008)