Lexico-syntactic constraints influence verbal working memory in sentence-like lists

Abstract

We test predictions from the language emergent perspective on verbal working memory that lexico-syntactic constraints should support both item and order memory. In natural language, long-term knowledge of lexico-syntactic patterns involving part of speech, verb biases, and noun animacy support language comprehension and production. In three experiments, participants were presented with randomly generated dative-like sentences or lists in which part of speech, verb biases, and animacy of a single word were manipulated. Participants were more likely to recall words in the correct position when presented with a verb over a noun in the verb position, a good dative verb over an intransitive verb in the verb position, and an animate noun over an inanimate noun in the subject noun position. These results demonstrate that interactions between words and their context in the form of lexico-syntactic constraints influence verbal working memory.

Appendices

Appendix A

Table 1 Critical stimuli of Experiment 1 and Experiment 2A

Table 2 Means and standard deviations of control factors

Table 3 Critical stimuli of Experiment 2B

Appendix A4

Similar to our procedures in Experiment 1, we then extracted and parsed all sentences containing target intransitive verbs across the same portion of COCA. We again extracted coarse-grained dependency information associated with these verbs, heuristically identifying sentences containing intransitive verbs as those that did not contain direct object, dative, or passive dependencies. We also excluded sentences in which the subject of the sentence was the object of a relative clause, such as, The bill Taylor paid… because the verb is transitive; the sentence contains a direct object (bill) before the verb. We then tabulated the frequencies for each sentence containing the candidate verbs that met these criteria and excluded those verbs that were not used in strictly intransitive contexts. Finally, the second and fourth authors further excluded those verbs that they judged could accept other kinds of noun phrases in the next position (e.g., a wink in The kids hardly slept a wink).

Appendix A5

To align transitive and intransitive words, we first estimated the mean Euclidean distance between all dative verbs to their noun controls from Experiment 1, as well as the standard deviation of these distances (mean = 2.18, standard deviation = 1.51) to establish a baseline for dative–intransitive control pairs. For all 65 × 52 (= 3,380) candidate dative-intransitive pairs, we then calculated the Euclidean distance between all four features from the dative verbs to the intransitive verbs. We then excluded all pairs that fell outside plus or minus one standard deviation of the dative-noun mean, which eliminated 50% of the possible dative–intransitive verb pairs. Of the remaining 1,699 verb pairs, we wished to identify sufficiently similar verb pairs in a computationally efficient way, so we applied a greedy allocation algorithm. For each of the intransitive verbs we selected the closest (Euclidean distance) dative verb; this dative verb was then removed from the pool of verbs available as potential matches for the remaining intransitive verbs. This process continued until all verbs were allocated a closest neighboring verb.

Appendix A6

For every animate noun, the Euclidean distance to every other animate noun on the dimensions of character length, frequency, and contextual diversity was calculated. All dimensions were z-scored to ensure they were weighed equally. The 10 closest inanimate nouns were then selected for each animate noun for further comparison based upon semantic similarity measures. The semantic similarity between target animate nouns and their foil inanimate nouns was defined as the cosine similarity of the word vectors pulled from the spaCy natural language processing Python package (version 2.1.8; specifically the model ‘en_core_web_lg’). Animate–inanimate noun pairs were generated by taking the 10 inanimate nouns with the highest cosine similarity for each animate noun and assigning them in a greedy fashion. The final animate–inanimate pairs were identified by selecting the 52 pairs with the highest cosine similarity.

Appendix A7

To allow comparison of the results of Experiment 2A with the other experiments and to test the robustness of the effects in Experiment 2A, a series of mixed effect logistic regression models were fit on recall across all list positions using the data from Experiment 2A. These additional models were fit using the same fixed and random effects as the models reported in the body of the paper. We report the models on three types of scoring: strict serial scoring to assess overall recall, lenient scoring to assess item memory, and conditionalized order scoring to assess order memory.

Verb biases on overall recall. Participants had higher recall when presented with a dative verb in Position 3 relative to an intransitive verb in Position 3, as indicated by a significant effect of verb type, β = 0.11, χ²(1) = 6.35, p < .05. There was no effect of counterbalance list condition on recall, χ²(1) = 0.23, p = .63, nor was there an interaction between the effect of verb type and counterbalance list condition, χ²(1) = 0.15, p = .70. There was a significant linear trend of list position on recall, β = 1.92, χ²(1) = 2123.70, p < .001, as well as a significant quadratic trend of list position on recall, β = 1.28, χ²(1) = 89.11, p < .001.

Verb biases on item memory. Participants were more likely to recall words irrespective of the position in which they were recalled when presented with a dative verb in Position 3 relative to an intransitive verb in Position 3, β = 0.13, χ²(1) = 13.15, p < .001. There was no effect of counterbalance list condition, χ²(1) = 0.28, p = .60. Further, there was no interaction between type of verb and counterbalance list condition, χ²(1) = 0.04, p = .85. There was a significant linear trend of list position on item memory, β = χ1.50, χ²(1) = 1364.68, p < .001, and there was a significant effect of the quadratic trend of list position on item memory, β = 1.06, χ²(1) = 61.58, p < .001.

Verb biases on order memory. Participants were no different at recalling all words in a list when presented with a ditransitive verb in Position 3 compared with when an intransitive verb was in Position 3, χ²(1) = 0.08, p = .78. Additionally, there was no effect of counterbalance list condition on recall, χ²(1) = 0.04, p = .85, nor was there an interaction between verb type and counterbalance list condition, χ²(1) = 0.77, p = .38. There was an effect of the linear trend of list condition on order memory, β = −3.86, χ²(1) = 657.41, p < .001, as well as an effect of the quadratic trend of list position on order memory, β = 4.58, χ²(1) = 127.17, p < .001.