About

Elsi Heilala Kaiser is a Professor in the Department of Linguistics at the University of Southern California, where she currently serves as the department chair. Her primary research focus is in psycholinguistics, particularly human sentence processing. She completed her dissertation at the University of Pennsylvania in 2003, which concentrated on reference resolution in Finnish, Dutch, and Estonian, utilizing a combination of psycholinguistic and corpus data. Kaiser holds a B.A. in Germanic Languages and Literatures from Princeton University and an M.A. in Psychology from the University of Pennsylvania. She was a post-doctoral researcher at the Center for Language Sciences at the University of Rochester from 2003 to 2005 before moving to Los Angeles in 2005. Her research interests encompass issues at the syntax, semantics, and pragmatics interface, including how information structure is encoded and processed, such as word order and prosody coherence relations in discourse, with particular attention to causality, perspective-taking, predicates of personal taste, subjectivity, and phenomena like free indirect discourse. She also investigates reference resolution and dependency formation, focusing on the comprehension and production of personal pronouns, demonstratives, reflexives, and reciprocals, as well as generic and impersonal reference. Her work extends to the representation and processing of linguistic and non-linguistic dependencies, including wh-dependencies and satiation effects, and explores the relations between language and music in dependency representation. Kaiser conducts cross-linguistic research involving typologically diverse languages, including Finnish, Estonian, French, German, Dutch, Bangla/Bengali, Hindi, Italian, Korean, Mandarin Chinese, and Vietnamese, often collaborating across linguistic boundaries.

Research topics

• Computer Science
• Political Science
• Artificial Intelligence
• Natural Language Processing
• Philosophy
• Psychology
• Social psychology
• World Wide Web
• Linguistics
• Epistemology
• Mathematics
• Programming language
• Internet privacy
• Law
• Medicine
• Business
• Advertising

Selected publications

• Event Knowledge Modulates Real‐Time Mental Representations of Object State‐Change

  Cognitive Science · 2026-01-01 · 1 citations

  articleOpen accessSenior author

  The present study examines how real-world event knowledge and grammatical aspect guide event comprehension. Specifically, we tested whether real-world knowledge about the likelihood of state-change (e.g., wine glasses usually crack when dropped but plastic cups do not) modulates the object state representations that people construct while reading perfective and imperfective sentences. Participants read "rebus" sentences in perfective and imperfective aspect, presented one word at a time, self-paced. In each sentence, the object was replaced by an image of the object that is either likely or unlikely to undergo state-change (e.g., Carlos was dropping/dropped a *wine glass*/*plastic cup* …), depicted in their initial (intact) or end (changed) states. Reaction times to images indicate that real-world knowledge about the likelihood of state-change is recruited when comprehenders construct mental models of events described as completed (perfective aspect, e.g., dropped) as well as events described as ongoing (imperfective aspect, e.g., was dropping). Results also indicate that perfective aspect increases the accessibility of both the initial and end states of objects, compared to imperfective aspect. Overall, these results demonstrate that both non-linguistic information grounded in real-world event knowledge as well as linguistic cues about the temporal structure of events guide how comprehenders dynamically update mental representations of object states in real-time.

  PublisherOA PDFDOI

• Deep learning characterizes depression and suicidal ideation in young adults from eye movements

  npj Digital Medicine · 2026-03-28 · 1 citations

  articleOpen access

  Objective biobehavioral markers for mental health conditions remain elusive, with diagnosis typically relying on self-reports and clinical interviews. We investigate eye tracking as a potential marker of attentional and mood biases associated with symptoms of depression and suicidal ideation from self-reported screening questionnaires. We analyze eye movements from 126 young adults during reading and responding to emotionally loaded sentences. A deep learning framework was designed to account for intra-trial and inter-trial variations in eye movements, achieving an AUC of 0.793 (95% CI: 0.766-0.819) for identifying depression/suicidality against healthy controls, and 0.826 (95% CI: 0.798-0.853) for suicidality specifically. The model also exhibited moderate accuracy in differentiating depressed from suicidal individuals (AUC: 0.609, 95% CI: 0.569-0.646). Discriminative patterns were more pronounced during response generation and for stimuli of negative sentiment. These findings suggest that eye tracking can provide objective markers of self-reported symptom severity by measuring the impact of emotional stimuli on oculomotor control.

  PublisherOA PDFDOI

• Uniform Information Density and Syntactic Reduction: Revisiting $\textit{that}$-Mentioning in English Complement Clauses

  ArXiv.org · 2025-09-05

  preprintOpen accessSenior author

  Speakers often have multiple ways to express the same meaning. The Uniform Information Density (UID) hypothesis suggests that speakers exploit this variability to maintain a consistent rate of information transmission during language production. Building on prior work linking UID to syntactic reduction, we revisit the finding that the optional complementizer $\textit{that}$ in English complement clauses is more likely to be omitted when the clause has low information density (i.e., more predictable). We advance this line of research by analyzing a large-scale, contemporary conversational corpus and using machine learning and neural language models to refine estimates of information density. Our results replicated the established relationship between information density and $\textit{that}$-mentioning. However, we found that previous measures of information density based on matrix verbs' subcategorization probability capture substantial idiosyncratic lexical variation. By contrast, estimates derived from contextual word embeddings account for additional variance in patterns of complementizer usage.

  PublisherOA PDFDOI

• On approximative degree morphology in Finnish: A comparison of two suffixes

  Proceedings from Semantics and Linguistic Theory · 2025-02-12

  articleOpen access1st authorCorresponding

  This paper discusses two approximative suffixal morphemes in Finnish, -hkO and -VhtAvA which, at first glance, both seem to resemble English -ish. However, I show that these two suffixes differ systematically in their distribution and semantics, and operate on different kinds of scales. I propose that -hkO, which attaches to gradable adjectives, signals proximity to a standard on a gradable scale provided by the adjective that -hkO modifies. In contrast, I claim that -VhtAvA, which attaches to nouns, signals proximity to a prototypical/canonical denotation of the noun that -VhtAvA modifies. Thus, the two approximative degree morphemes in Finnish wear their scale structure on their sleeve, so to speak. Evidence from comparatives and superlatives, as well as constraints on the order in which these suffixes can be stacked, supports the proposed analysis.

  PublisherOA PDFDOI

• On the Semantics and Pragmatics of ‘(An)other’ in Finnish: Comparing toinen and muu

  2025-11-13

  book-chapter1st authorCorresponding
  PublisherDOI

• Comparing form-based and meaning-based gender biases in pronoun resolution: Inferences from names and role nouns

  Proceedings of the Linguistic Society of America · 2025-06-18

  articleOpen access1st authorCorresponding

  Referring to people using only their name (e.g. Jones came in) is known to evoke the assumption that the individual is male. The same holds with some role nouns (e.g. mechanic, boxer). We explore these effects through the lens of pronoun interpretation in English. In two sentence-completion studies, we show that both form-based (last-name-only style) and meaning-based gender biases (from role nouns) are powerful enough to eliminate otherwise robust verb semantic effects on pronoun interpretation (implicit causality). In addition, the results provide initial evidence that meaning-based biases (at least the ones tested here) can be stronger than form-based biases, which may stem from differences in the form-function mapping.

  PublisherOA PDFDOI

• Effects of information structure on pronoun resolution: the number of pronouns matters

  Linguistics Vanguard · 2025-10-21

  articleOpen accessSenior author

  Abstract This paper addresses underexplored questions about referential structure impacting pronoun interpretation in multiple-pronoun and single-pronoun contexts: whether and how referential structure effects in pronoun resolution interact with information-structural factors such as topicality and focus. Regarding referential structure effects, it has been argued that demoting a higher-salience referent, which is a subject antecedent, to a lower-salience object position leads to less coherent transitions, resulting in different pronoun interpretation patterns by referential structure type. To investigate whether and how information structure interacts with referential structure effects, we disentangle information structure from subjecthood. We use a picture-writing task, where we manipulated referential structure (One-Pronoun vs. Two-Pronoun sentences) and information structure (Subject-Topic/Object-Focus vs. Subject-Focus/Object-Topic) in a dialogue setup. The results replicate previously observed referential structure effects on pronoun interpretation. Crucially, we find that both subjecthood and information structure affect these effects, but demoting a subject-focus antecedent is more dispreferred than demoting a subject-topic antecedent. We discuss how our results contribute to providing evidence of information structure contributing to referential structure effects in multiple-pronoun contexts, which have been largely overlooked in most prior work on pronoun interpretation.

  PublisherDOI

• Uniform Information Density and Syntactic Reduction: Revisiting *that*-Mentioning in English Complement Clauses

  2025-01-01

  articleOpen accessSenior author

  Speakers often have multiple ways to express the same meaning.The Uniform Information Density (UID) hypothesis suggests that speakers exploit this variability to maintain a consistent rate of information transmission during language production.Building on prior work linking UID to syntactic reduction, we revisit the finding that the optional complementizer that in English complement clauses is more likely to be omitted when the clause has low information density (i.e., more predictable).We advance this line of research by analyzing a large-scale, contemporary conversational corpus and using machine learning and neural language models to refine estimates of information density.Our results replicate the established relationship between information density and that-mentioning.However, we find that previous measures of information density based on matrix verbs' subcategorization probability capture substantial idiosyncratic lexical variation.By contrast, estimates derived from contextual word embeddings account for additional variance in patterns of complementizer usage. 1

  PublisherOA PDFDOI

• Dissimilation and referential form: Evidence from Finnish personal pronouns and anaphoric demonstratives

  Acta Linguistica Academica · 2025-10-03

  article1st authorCorresponding

  Abstract In many languages, two referentially disjoint personal pronouns can occur in the same domain (e.g., she saw her ), as expected given Binding Principle B. I present corpus-based and judgment-based evidence showing that Finnish patterns differently and is subject to a Pronoun Dissimilation Constraint: When two expressions in the same local domain (coarguments of the same verb) are referentially distinct, realizing them both as personal pronouns is dispreferred. In other words, sequences such as ‘She saw her’ sound very odd in Finnish, but replacing one of the personal pronouns with an anaphoric demonstrative yields an acceptable sentence. I present evidence showing that the Pronoun Dissimilation Constraint cannot be reduced to a pure disambiguation phenomenon, nor to linear proximity, phonological similarity, or the presence of another option in the language's anaphoric paradigm. After presenting large-scale corpus data assessing the robustness of the Pronoun Dissimilation Constraint, I return to the question about the source of this constraint, and explore links to other work on dissimilation phenomena and obviation patterns in Algonquian languages.

  PublisherDOI

• Anticipatory processing of cataphora is constrained by binding principles in L2 English

  Bilingualism Language and Cognition · 2024-03-22 · 3 citations

  articleSenior author

  Abstract Language processing studies show that native speakers anticipate linguistic elements before their occurrence. However, it is debated to what extent second language (L2) learners do the same. To address this question, this study examines the processing of cataphora by Chinese-speaking L2 English learners. Additionally, we query whether L2 learners’ expectations of upcoming antecedents are modulated by first language (L1) influence and constrained by Principle B of the Binding Theory (Chomsky, 1981). Two self-paced reading studies show that L1 English speakers’ anticipation of upcoming referents is active and strictly constrained by Principle B. Crucially, L2 English learners also actively predict upcoming referents and are sensitive to Principle B. However, L2 processing patterns suggest that Principle B competes with semantics at later processing stages. Together with data from L1 Chinese and English control participants, these results support the view that anticipatory processing in English is not fundamentally different between monolinguals and bilinguals.

  PublisherDOI

Recent grants

• Subjectivity in language processing: Linguistic and psychological aspects of perspective-taking

  NSF · $317k · 2018–2023

• Doctoral Dissertation Research: Prosodic Encoding of Informativity

  NSF · $20k · 2015–2017

• NIH Grant R01HD061457

  NIH · $460k · 2013

Frequent coauthors

• Jun Lyu

  15 shared

• Katja Reuter

  SUNY Upstate Medical University

  14 shared

• Jesse Storbeck

  Southern California University for Professional Studies

  14 shared

• NamQuyen Le

  University of Southern California

  12 shared

• Praveen Angyan

  12 shared

• Iris Chuoying Ouyang

  University of Southern California

  11 shared

• Jennifer B. Unger

  University of Southern California

  9 shared

• Francesco Pierini

  9 shared