Natural Language Understanding Wiki

TODO: Kehler et al. (2008)[1], Rohde (2008)[2], Kehler and Rhode (2013)[3]

From Grüter et al. (2015)[4]: "Psycholinguistic research has investigated the interaction of multiple probabilistic cues in referential processing in some depth, and within a variety of theoretical and experimental paradigms (e.g., Arnold, 2001, 2010; Grosz, Joshi, & Weinstein, 1995; Hobbs, 1979; Kaiser, 2011; Kehler, 2002; Stevenson, Crawley, & Kleinman, 1994) [...]

Tracking reference across sentences depends on a variety of factors—some related to properties of the available referents such as their grammatical or thematic roles (Arnold, 2001; Caramazza et al., 1977; McKoon, Greene, & Ratcliff, 1993; Stevenson et al., 1994), others related to properties of the unfolding discourse and the events being described (Hobbs, 1979; Kehler, 2002; Madden & Zwaan, 2003), and still others related to the available referring expressions in a language’s referential system and the felicity of those expressions for re-mentioning topical entities (Ariel, 1990; Gundel, Hedberg, & Zacharski, 1993; Grosz et al., 1995; Gatt, Krahmer, van Deemter, & van Gompel, 2014)."

Cues that people use to solve coreference[]

Superficial heuristics[]

Rhode et al. (2006)[5]: "The subject preference (Crawley, Stevenson, & Kleinman 1990) predicts that an ambiguous pronoun ought to be coreferential with the subject of the previous sentence, in this case the Source. The first-mention privilege (Gernsbacher & Hargreaves 1988) also points to the salience of the Source subject. The grammatical parallelism preference (Smyth 1994; Chambers & Smyth 1998) predicts that an ambiguous subject pronoun resolves preferentially to a subject antecedent. [...] thematic-role-level preference which amounts to a heuristic ranking Goals above Sources"


Main reference: Kehler (2002)[6]

Which cues are more important?[]

Rhode et al. (2006)[5]: "We found that sentences in the imperfective yielded significantly more Source resolutions than those in the perfective, supporting the event-structure hypothesis. Furthermore, we found that a significant proportion of Goal interpretations arise from continuations which use Occasion (i.e. narration-based) coherence relations. As such, the interpretation preferences are better explained at the level of event structure and discourse coherence, rather than by appeal to superficial heuristics."

TODO: Wolf et al. (2004)[7], Kertz et al. (2006)[8]


  1. Kehler, A., Kertz, L., Rohde, H., & Elman, J. L. (2008). Coherence and coreference revisited. Journal of Semantics, 25(1 SPEC. ISS.), 1–44.
  2. Rohde, H. (2008). Coherence-Driven Effects in Sentence and Discourse Processing.
  3. Kehler, A., & Rohde, H. (2013). A probabilistic reconciliation of coherence-driven and centering-driven theories of pronoun interpretation. Theoretical Linguistics, 39(1–2), 1–37.
  4. Grüter, T., Rohde, H., & Schafer, A. J. (2015). Coreference and discourse coherence in L2: The roles of grammatical aspect and referential form. Linguistic Approaches to Bilingualism, 1–28.
  5. 5.0 5.1 Rohde, H., Kehler, A., & Elman, J. L. (2006). Event Structure and Discourse Coherence Biases in Pronoun Interpretation. R. Sun. (ED.), Proceedings of the 28th Annual Conference of the Cognitive Science Society. Mahwah, NJ: Lawrence Erlbaum Associates., 112(2), 617–622.
  6. Kehler, A. (2002). Coherence, reference, and the theory of grammar. Stanford, CA: CSLI Publications.
  7. Wolf, F., Gibson, E. & Desmet, T. (2004) Coherence and pronoun resolution. Language and Cognitive Processes, 19(6), 665-675
  8. Kertz, L., Kehler, A., & Elman, J. (2006) Grammatical and coherence-based factors in pronoun interpretation. In the Proceedings of the 28th Annual Conference of the Cognitive Science Society (this volume), Vancouver, July 26-29, 2006.