Marco Marelli

An amorphous model for morphological processing in visual comprehension based on naive discriminative learning.

A 2-layer symbolic network model based on the equilibrium equations of the Rescorla-Wagner model (Danks, 2003) is proposed. The study first presents 2 experiments in Serbian, which reveal for sentential reading the inflectional paradigmatic effects previously observed by Milin, Filipović Đurđević, and Moscoso del Prado Martín (2009) for unprimed lexical decision. The empirical results are successfully modeled without having to assume separate representations for inflections or data structures such as inflectional paradigms. In the next step, the same naive discriminative learning approach is pitted against a wide range of effects documented in the morphological processing literature. Frequency effects for complex words as well as for phrases (Arnon & Snider, 2010) emerge in the model without the presence of whole-word or whole-phrase representations. Family size effects (Moscoso del Prado Martín, Bertram, Häikiö, Schreuder, & Baayen, 2004; Schreuder & Baayen, 1997) emerge in the simulations across simple words, derived words, and compounds, without derived words or compounds being represented as such. It is shown that for pseudo-derived words no special morpho-orthographic segmentation mechanism, as posited by Rastle, Davis, and New (2004), is required. The model also replicates the finding of Plag and Baayen (2009) that, on average, words with more productive affixes elicit longer response latencies; at the same time, it predicts that productive affixes afford faster response latencies for new words. English phrasal paradigmatic effects modulating isolated word reading are reported and modeled, showing that the paradigmatic effects characterizing Serbian case inflection have crosslinguistic scope.

SemEval-2014 Task 1: Evaluation of Compositional Distributional Semantic Models on Full Sentences through Semantic Relatedness and Textual Entailment

This paper presents the task on the evaluation of Compositional Distributional Semantics Models on full sentences organized for the first time within SemEval2014. Participation was open to systems based on any approach. Systems were presented with pairs of sentences and were evaluated on their ability to predict human judgments on (i) semantic relatedness and (ii) entailment. The task attracted 21 teams, most of which participated in both subtasks. We received 17 submissions in the relatedness subtask (for a total of 66 runs) and 18 in the entailment subtask (65 runs).

Affixation in semantic space: Modeling morpheme meanings with compositional distributional semantics.

The present work proposes a computational model of morpheme combination at the meaning level. The model moves from the tenets of distributional semantics, and assumes that word meanings can be effectively represented by vectors recording their co-occurrence with other words in a large text corpus. Given this assumption, affixes are modeled as functions (matrices) mapping stems onto derived forms. Derived-form meanings can be thought of as the result of a combinatorial procedure that transforms the stem vector on the basis of the affix matrix (e.g., the meaning of nameless is obtained by multiplying the vector of name with the matrix of -less). We show that this architecture accounts for the remarkable human capacity of generating new words that denote novel meanings, correctly predicting semantic intuitions about novel derived forms. Moreover, the proposed compositional approach, once paired with a whole-word route, provides a new interpretative framework for semantic transparency, which is here partially explained in terms of ease of the combinatorial procedure and strength of the transformation brought about by the affix. Model-based predictions are in line with the modulation of semantic transparency on explicit intuitions about existing words, response times in lexical decision, and morphological priming. In conclusion, we introduce a computational model to account for morpheme combination at the meaning level. The model is data-driven, theoretically sound, and empirically supported, and it makes predictions that open new research avenues in the domain of semantic processing. (PsycINFO Database Record

Meaning is in the beholder’s eye: Morpho-semantic effects in masked priming

A substantial body of literature indicates that, at least at some level of processing, complex words are broken down into their morphemes solely on the basis of their orthographic form (e.g., Rastle, Davis, & New, Psychonomic Bulletin and Review 11:1090–1098, 2004). Recent evidence has shown that this process might not be obligatory, as indicated by the fact that morpho-orthographic effects were not found in a cross-case same–different task—that is, when lexical access was not necessarily required (Duñabeitia, Kinoshita, Carreiras, & Norris, Language and Cognitive Processes 26:509–529, 2011). In this study, we employed a task that required understanding a series of words and, thus, implied lexical access. Masked primes were shown very briefly right before the appearance of the target word; prime–target pairs entertained a morpho-semantic (dealer–DEAL), a morpho-orthographic (corner–CORN), or a purely orthographic (brothel–BROTH) relationship. Eye fixation times clearly indicated facilitation for transparent pairs, but not for opaque pairs (or for orthographic pairs, which were used as a baseline). Conversely, the usual morpho-orthographic pattern was found in a control experiment, employing a lexical decision task. These results indicate that the access to a morpho-orthographic level of representation is not always necessary for lexical identification, which challenges models of visual word identification that cannot account for task-induced effects.

Semantic transparency in free stems: The effect of Orthography-Semantics Consistency on word recognition.

A largely overlooked side effect in most studies of morphological priming is a consistent main effect of semantic transparency across priming conditions. That is, participants are faster at recognizing stems from transparent sets (e.g., farm) in comparison to stems from opaque sets (e.g., fruit), regardless of the preceding primes. This suggests that semantic transparency may also be consistently associated with some property of the stem word. We propose that this property might be traced back to the consistency, throughout the lexicon, between the orthographic form of a word and its meaning, here named Orthography-Semantics Consistency (OSC), and that an imbalance in OSC scores might explain the “stem transparency” effect. We exploited distributional semantic models to quantitatively characterize OSC, and tested its effect on visual word identification relying on large-scale data taken from the British Lexicon Project (BLP). Results indicated that (a) the “stem transparency” effect is solid and reliable, insofar as it holds in BLP lexical decision times (Experiment 1); (b) an imbalance in terms of OSC can account for it (Experiment 2); and (c) more generally, OSC explains variance in a large item sample from the BLP, proving to be an effective predictor in visual word access (Experiment 3).

Multimodal Word Meaning Induction From Minimal Exposure to Natural Text

By the time they reach early adulthood, English speakers are familiar with the meaning of thousands of words. In the last decades, computational simulations known as distributional semantic models (DSMs) have demonstrated that it is possible to induce word meaning representations solely from word co-occurrence statistics extracted from a large amount of text. However, while these models learn in batch mode from large corpora, human word learning proceeds incrementally after minimal exposure to new words. In this study, we run a set of experiments investigating whether minimal distributional evidence from very short passages suffices to trigger successful word learning in subjects, testing their linguistic and visual intuitions about the concepts associated with new words. After confirming that subjects are indeed very efficient distributional learners even from small amounts of evidence, we test a DSM on the same multimodal task, finding that it behaves in a remarkable human-like way. We conclude that DSMs provide a convincing computational account of word learning even at the early stages in which a word is first encountered, and the way they build meaning representations can offer new insights into human language acquisition.

SICK through the SemEval glasses. Lesson learned from the evaluation of compositional distributional semantic models on full sentences through semantic relatedness and textual entailment

AbstractThis paper is an extended description of SemEval-2014 Task 1, the task on the evaluation of Compositional Distributional Semantics Models on full sentences. Systems participating in the task were presented with pairs of sentences and were evaluated on their ability to predict human judgments on (1) semantic relatedness and (2) entailment. Training and testing data were subsets of the SICK (Sentences Involving Compositional Knowledge) data set. SICK was developed with the aim of providing a proper benchmark to evaluate compositional semantic systems, though task participation was open to systems based on any approach. Taking advantage of the SemEval experience, in this paper we analyze the SICK data set, in order to evaluate the extent to which it meets its design goal and to shed light on the linguistic phenomena that are still challenging for state-of-the-art computational semantic systems. Qualitative and quantitative error analyses show that many systems are quite sensitive to changes in the proportion of sentence pair types, and degrade in the presence of additional lexico-syntactic complexities which do not affect human judgements. More compositional systems seem to perform better when the task proportions are changed, but the effect needs further confirmation.

Framing effects reveal discrete lexical-semantic and sublexical procedures in reading: an fMRI study

According to the dual-route model, a printed string of letters can be processed by either a grapheme-to-phoneme conversion (GPC) route or a lexical-semantic route. Although meta-analyses of the imaging literature support the existence of distinct but interacting reading procedures, individual neuroimaging studies that explored neural correlates of reading yielded inconclusive results. We used a list-manipulation paradigm to provide a fresh empirical look at this issue and to isolate specific areas that underlie the two reading procedures. In a lexical condition, we embedded disyllabic Italian words (target stimuli) in lists of either loanwords or trisyllabic Italian words with unpredictable stress position. In a GPC condition, similar target stimuli were included within lists of pseudowords. The procedure was designed to induce participants to emphasize either the lexical-semantic or the GPC reading procedure, while controlling for possible linguistic confounds and keeping the reading task requirements stable across the two conditions. Thirty-three adults participated in the behavioral study, and 20 further adult participants were included in the fMRI study. At the behavioral level, we found sizeable effects of the framing manipulations that included slower voice onset times for stimuli in the pseudoword frames. At the functional anatomical level, the occipital and temporal regions, and the intraparietal sulcus were specifically activated when subjects were reading target words in a lexical frame. The inferior parietal and anterior fusiform cortex were specifically activated in the GPC condition. These patterns of activation represented a valid classifying model of fMRI images associated with target reading in both frames in the multi-voxel pattern analyses. Further activations were shared by the two procedures in the occipital and inferior parietal areas, in the premotor cortex, in the frontal regions and the left supplementary motor area. These regions are most likely involved in either early input or late output processes.

Picking buttercups and eating butter cups: Spelling alternations, semantic relatedness, and their consequences for compound processing

Semantic transparency (ST) is a measure quantifying the strength of meaning association between a compound word (buttercup) and its constituents (butter, cup). Borrowing ideas from computational semantics, we characterize ST in terms of the degree to which a compound and its constituents tend to share the same contexts in everyday usage, and we collect separate measures for different orthographic realizations (solid vs. open) of the same compound. We can thus compare the effects of semantic association in cases in which direct semantic access is likely to take place (buttercup), vis-a-vis forms that encourage combinatorial procedures (butter cup). ST effects are investigated in an analysis of lexical decision latencies. The results indicate that distributionally based ST variables are most predictive of response times when extracted from contexts presenting the compounds as open forms, suggesting that compound processing involves a conceptual combination procedure focusing on the merger of the constituent meanings.

Understanding the mental lexicon through neglect dyslexia: a study on compound noun reading

The present study employs neglect dyslexia (ND) as an experimental model to study compound-word processing; in particular, it investigates whether compound constituents are hierarchically organized at mental level and addresses the possibility of whole-word representation. Seven Italian-speaking patients suffering from ND participated in a word naming task. Both left-headed (pescespada, swordfish) and right-headed (astronave, spaceship) Italian compound nouns were used as stimuli. Non-existent compounds, which were generated by substituting the leftmost constituent of a compound with an orthographically similar word (e.g., *pestespada, *plaguesword), were also employed. A significant headedness effect emerged in the group analysis: patients read left-headed compounds better than right-headed compounds. A significant lexicality effect was also found: the participants read real compounds better than their non-existent compound pairs. Moreover, logit mixed-effects analyses indicated a left-hand constituent frequency effect. Results are discussed in terms of hierarchical representation of compounds and direct access to compound lemma nodes.