Local Associations and Semantic Ties in Overt and Masked Semantic Priming

English. Distributional semantic models (DSM) are widely used in psycholinguistic research to automatically assess the degree of semantic relatedness between words. Model estimates strongly correlate with human similarity judgements and offer a tool to successfully predict a wide range of language-related phenomena. In the present study, we compare the state-of-art model with pointwise mutual information (PMI), a measure of local association between words based on their surface cooccurrence. In particular, we test how the two indexes perform on a dataset of sematic priming data, showing how PMI outperforms DSM in the fit to the behavioral data. According to our result, what has been traditionally thought of as semantic effects may mostly rely on local associations based on word cooccurrence. Italiano. I modelli semantici distribuzionali sono ampiamente utilizzati in psicolinguistica per quantificare il grado di similarità tra parole. Tali stime sono in linea con i corrispettivi giudizi umani, e offrono uno strumento per modellare un'ampia gamma di fenomeni relativi al linguaggio. Nel presente studio, confrontiamo il modello con la pointwise mutual information (PMI), una misura di associazione locale tra parole basata sulla loro cooccorrenza. In particolare, abbiamo testato i due indici su un set di dati di priming semantico, mostrando come la PMI riesca a spiegare meglio i dati comportamentali. Alla luce di tali risultati, ciò che è stato tradizionalmente considerato come effetto semantico potrebbe basarsi principalmente su associazioni locali di co-occorrenza lessicale.

[1]  K. Forster,et al.  REPETITION PRIMING AND FREQUENCY ATTENUATION IN LEXICAL ACCESS , 1984 .

[2]  Silvia Bernardini,et al.  The WaCky wide web: a collection of very large linguistically processed web-crawled corpora , 2009, Lang. Resour. Evaluation.

[3]  Stefan Evert,et al.  Corpora and collocations , 2007 .

[4]  Marco Marelli,et al.  Word-embeddings Italian semantic spaces: A semantic model for psycholinguistic research , 2017 .

[5]  Georgiana Dinu,et al.  Don’t count, predict! A systematic comparison of context-counting vs. context-predicting semantic vectors , 2014, ACL.

[6]  S. Kinoshita,et al.  The masked semantic priming effect is task dependent: Reconsidering the automatic spreading activation process. , 2015, Journal of experimental psychology. Learning, memory, and cognition.

[7]  Gabriel Recchia,et al.  More data trumps smarter algorithms: Comparing pointwise mutual information with latent semantic analysis , 2009, Behavior research methods.

[8]  T. McNamara Semantic Priming: Perspectives from Memory and Word Recognition , 2005 .

[9]  Nick Chater,et al.  Judging the Probability of Hypotheses Versus the Impact of Evidence: Which Form of Inductive Inference Is More Accurate and Time-Consistent? , 2016, Cogn. Sci..

[10]  M. Brysbaert,et al.  Explaining human performance in psycholinguistic tasks with models of semantic similarity based on prediction and counting : A review and empirical validation , 2017 .

[11]  Benjamin Rohaut,et al.  What are the boundaries of unconscious semantic cognition? , 2018, The European journal of neuroscience.

[12]  R. Bottini,et al.  The nature of semantic priming by subliminal spatial words: Embodied or disembodied? , 2016, Journal of experimental psychology. General.

[13]  Hinrich Schütze,et al.  Book Reviews: Foundations of Statistical Natural Language Processing , 1999, CL.

[14]  T. Landauer,et al.  A Solution to Plato's Problem: The Latent Semantic Analysis Theory of Acquisition, Induction, and Representation of Knowledge. , 1997 .

[15]  D. M. Green,et al.  Signal detection theory and psychophysics , 1966 .

[16]  Jeffrey Dean,et al.  Efficient Estimation of Word Representations in Vector Space , 2013, ICLR.

[17]  M. Marelli,et al.  Corpus-based estimates of word association predict biases in judgment of word co-occurrence likelihood , 2014, Cognitive Psychology.