Basic processes in reading: A critical review of pseudohomophone effects in reading aloud and a new computational account

There are pervasive lexical influences on the time that it takes to read aloud novel letter strings that sound like real words (e.g.,brane frombrain). However, the literature presents a complicated picture, given that the time taken to read aloud such items is sometimes shorter and sometimes longer than a control string (e.g.,frane) and that the time to read aloud is sometimes affected by the frequency of the base word and other times is not. In the present review, we first organize these data to show that there is considerably more consistency than has previously been acknowledged. We then consider six different accounts that have been proposed to explain various aspects of these data. Four of them immediately fail in one way or another. The remaining two accounts may be able to explain these findings, but they either make counterintuitive assumptions or invoke a novel mechanism solely to explain these findings. A new account is advanced that is able to explain all of the effects reviewed here and has none of the problems associated with the other accounts. According to this account, different types of lexical knowledge are used when pseudohomophones and nonword controls are read aloud in mixed and pure lists. This account is then implemented in Coltheart, Rastle, Perry, Langdon, and Ziegler’s (2001) dual route cascaded model in order to provide an existence proof that it accommodates all of the effects, while retaining the ability to simulate three standard effects seen in nonword reading aloud.     

The structure of lexical concepts

Jerry Fodor (Concepts: Where cognitive science went wrong. New York: Oxford University Press, 1998) famously argued that lexical concepts are unstructured. After examining the advantages and disadvantages of both the classical approach to concepts and Fodor’s conceptual atomism, I argue that some lexical concepts are, in fact, structured. Roughly stated, I argue that structured lexical concepts bear a necessary biconditional entailment relation to their structural constituents. I develop this account of the structure of lexical concepts within the framework of Pavel Tichy’s (The foundations of Frege’s logic. Berlin, New York: De Gruyter, 1988) theory of constructions. I argue that concepts are constructions which can be combined by way of Tichy’s construction-forming operations of composition and closure and an additional operation, simplification, which I propose in section 6. The last of these construction-forming operations plays a central role in my account of lexical concept structure. Stated generally, structured lexical concepts are a result of simplifying their structural constituents.     

A model for evidence accumulation in the lexical decision task

We present a new model for lexical decision, REM-LD, that is based on REM theory (e.g., ). REM-LD uses a principled (i.e., Bayes' rule) decision process that simultaneously considers the diagnosticity of the evidence for the 'WORD' response and the 'NONWORD' response. The model calculates the odds ratio that the presented stimulus is a word or a nonword by averaging likelihood ratios for lexical entries from a small neighborhood of similar words. We report two experiments that used a signal-to-respond paradigm to obtain information about the time course of lexical processing. Experiment 1 verified the prediction of the model that the frequency of the word stimuli affects performance for nonword stimuli. Experiment 2 was done to study the effects of nonword lexicality, word frequency, and repetition priming and to demonstrate how REM-LD can account for the observed results. We discuss how REM-LD could be extended to account for effects of phonology such as the pseudohomophone effect, and how REM-LD can predict response times in the traditional 'respond-when-ready' paradigm.     

On the control of visual word recognition: Changing routes versus changing deadlines

Two different accounts have been proposed to explain the fact that (1) an effect of word frequency is present when readers of transparent orthographies read only words aloud and (2) the effect of word frequency is eliminated when subjects name words and nonwords mixed together in a single block. In the route-shifting account, subjects shift from using a lexical route that can read only words to using a nonlexical route that can read both words and nonwords via the use of sublexical spelling-sound correspondences (hence, no word frequency effect). The essence of the second, time criterion account is that the elimination of the word frequency effect is determined by the speed with which the nonwords are processed, because subjects attempt to homogenize the point in time at which they release an articulation. These two different accounts are pitted against each other in a series of naming experiments utilizing the transparent Turkish orthography. A word frequency effect persists even when words and nonwords are mixed together, provided that nonword sets are matched so as to be named as quickly as the high-frequency words and as slowly as the low-frequency words, respectively. This result is argued to be consistent with the time criterion account, but not with the unadorned route-shifting account.     

There goes the neighbourhood: contextual control over the breadth of lexical activation when reading aloud

There are currently two computational accounts of how the time to read pseudohomophones (like BRANE) and their nonword controls (like FRANE) varies with changes in context. In Reynolds and Besner's (2005) account, readers vary the breadth of lexical activation in response to changes in context. A competing account proposed by Kwantes and Marmurek (2007) and independently by Perry, Ziegler, and Zorzi (2007) has readers varying their response criterion in response to changes in context. The present work adjudicates between these two accounts by examining how the effect of neighbourhood density changes as a function of list context when reading pseudohomophones aloud. The results of an experiment and simulations from a leading computational model support the lexical breadth account, but are inconsistent with the response criterion account.     

Looks aren’t Everything: Pseudohomophones Prime Words but Nonwords do not

After examining literature that deals with phonological and orthographic effects associated with pseudohomophones, the current effort deviates from the norm by using fewer pseudohomophones (20) and extending the lags between primes and targets (M=8). Word and pseudohomophone primes were found to facilitate lexical decision response latencies to word targets. Response latencies to word targets were not influenced by nonword primes, however. The presence of pseudohomophone effects was demonstrated by longer response latencies and higher error rates for pseudohomophones (e.g., DREEM) that were equated in orthography to nonword controls (e.g., DROAM). Despite the frequency effect observed for base words, the pseudohomophones did not exhibit an effect of base word frequency. The results suggest that phonological codes exert an influence on lexical representation but are not frequency sensitive.     

Toward a scalable holographic word-form representation

Phenomena in a variety of verbal tasks—for example, masked priming, lexical decision, and word naming—are typically explained in terms of similarity between word-forms. Despite the apparent commonalities between these sets of phenomena, the representations and similarity measures used to account for them are not often related. To show how this gap might be bridged, we build on the work of Hannagan, Dupoux, and Christophe, Cognitive Science 35:79-118, (2011) to explore several methods of representing visual word-forms using holographic reduced representations and to evaluate them on their ability to account for a wide range of effects in masked form priming, as well as data from lexical decision and word naming. A representation that assumes that word-internal letter groups are encoded relative to word-terminal letter groups is found to predict qualitative patterns in masked priming, as well as lexical decision and naming latencies. We then show how this representation can be integrated with the BEAGLE model of lexical semantics (Jones & Mewhort, Psychological Review 114:1–37, 2007) to enable the model to encompass a wider range of verbal tasks.     

Second-language experience modulates first- and second-language word frequency effects: Evidence from eye movement measures of natural paragraph reading

We used eye movement measures of first-language (L1) and second-language (L2) paragraph reading to investigate whether the degree of current L2 exposure modulates the relative size of L1 and L2 frequency effects (FEs). The results showed that bilinguals displayed larger L2 than L1 FEs during both early- and late-stage eye movement measures, which are taken to reflect initial lexical access and postlexical access, respectively. Moreover, the magnitude of L2 FEs was inversely related to current L2 exposure, such that lower levels of L2 exposure led to larger L2 FEs. In contrast, during early-stage reading measures, bilinguals with higher levels of current L2 exposure showed larger L1 FEs than did bilinguals with lower levels of L2 exposure, suggesting that increased L2 experience modifies the earliest stages of L1 lexical access. Taken together, the findings are consistent with implicit learning accounts (e.g., Monsell, 1991), the weaker links hypothesis (Gollan, Montoya, Cera, Sandoval, Journal of Memory and Language, 58:787–814, 2008), and current bilingual visual word recognition models (e.g., the bilingual interactive activation model plus [BIA+]; Dijkstra & van Heuven, Bilingualism: Language and Cognition, 5:175–197, 2002). Thus, amount of current L2 exposure is a key determinant of FEs and, thus, lexical activation, in both the L1 and L2.     

Orthographically Influenced Abstract Phonological Representation: Evidence from Non-rhotic Speakers

It is typically assumed that when orthography is translated silently into phonology (i.e., when reading silently), the phonological representation is equivalent to the spoken form or, at least, the surface phonemic form. The research presented here demonstrates that the phonological representation is likely to be more abstract than this, and is orthographically influenced. For example, the claim is made that the word corn has an underlying /r/ in its phonological representation, even in non-rhotic dialects. The evidence comes from difficulties observed in judgements about the homophony with a target word of a pseudohomophone whose phonology does not match the putative abstract representation of that word. For example, it is hard to say that the pseudohomophone cawn is homophonic with corn. The conclusion that orthography can shape phonological representation is antithetical to both computational models of the conversion of print to sound and linguistic accounts of phonology.The research reported in this paper was supported by a grant awarded to the author by the Australian Research Council.     

Frequency,not age of acquisition,affects Italian word naming

The purpose of this study was to evaluate the claim that age of acquisition (AoA) and word frequency effects are reduced or nonexistent in languages that have very regular letter-to-sound mappings, like Italian. The first two experiments (Exp. 1, Exp. 2) showed that frequency variables affect reading aloud and lexical decision in Italian. Variables interpretable as pertaining to a semantic component, including AoA, affected lexical decision but not reading aloud. In Experiments 3 and 4, a measure of frequency—child written word frequency (ChFreq)—and AoA were manipulated. Reading performance was affected by word frequency but not by AoA (Exp. 3), whereas lexical decision was affected by both variables (Exp. 4). In Experiments 5 and 6, ChFreq and AoA were manipulated orthogonally. Only frequency affected reading aloud, with no main effect or interaction involving AoA (Exp. 5). The effects of AoA and frequency interacted in Experiment 6 for lexical decision due to a larger effect of AoA for low frequency words than high frequency words. These results show that in languages with a transparent orthography word frequency may affect reading aloud in the absence of an effect of AoA because Italian readers employ lexical nonsemantic reading aloud. The effect of child written frequency points to the efficiency of the mappings between those orthographic and phonological word forms that were frequently encountered when learning to read.     

Entropy,semantic relatedness and proximity

Although word co-occurrences within a document have been demonstrated to be semantically useful, word interactions over a local range have been largely neglected by psychologists due to practical challenges. Shannon’s (Bell Systems Technical Journal, 27, 379–423, 623–665, 1948) conceptualization of information theory suggests that these interactions should be useful for understanding communication. Computational advances make an examination of local word–word interactions possible for a large text corpus. We used Brants and Franz’s (2006) dataset to generate conditional probabilities for 62,474 word pairs and entropy calculations for 9,917 words in Nelson, McEvoy, and Schreiber’s (Behavior Research Methods, Instruments, & Computers, 36, 402–407, 2004) free association norms. Semantic associativity correlated moderately with the probabilities and was stronger when the two words were not adjacent. The number of semantic associates for a word and the entropy of a word were also correlated. Finally, language entropy decreases from 11 bits for single words to 6 bits per word for four-word sequences. The probabilities and entropies discussed here are included in the supplemental materials for the article.     

Expectations from preceding prosody influence segmentation in online sentence processing

Previous work examining prosodic cues in online spoken-word recognition has focused primarily on local cues to word identity. However, recent studies have suggested that utterance-level prosodic patterns can also influence the interpretation of subsequent sequences of lexically ambiguous syllables (Dilley, Mattys, & Vinke, Journal of Memory and Language, 63:274–294, 2010; Dilley & McAuley, Journal of Memory and Language, 59:294–311, 2008). To test the hypothesis that these distal prosody effects are based on expectations about the organization of upcoming material, we conducted a visual-world experiment. We examined fixations to competing alternatives such as pan and panda upon hearing the target word panda in utterances in which the acoustic properties of the preceding sentence material had been manipulated. The proportions of fixations to the monosyllabic competitor were higher beginning 200 ms after target word onset when the preceding prosody supported a prosodic constituent boundary following pan-, rather than following panda. These findings support the hypothesis that expectations based on perceived prosodic patterns in the distal context influence lexical segmentation and word recognition.     

Revisiting backward recall and benchmark memory effects: a reply to Bireta et al. (2010)

When participants are asked to recall lists of items in the reverse order, known as backward recall, several benchmark memory phenomena, such as the word length effect, are abolished (Bireta et al. Memory & Cognition 38:279–291, 2010). Bireta et al. (Memory & Cognition 38:279–291, 2010) suggested that in backward recall, reliance on order retention is increased at the expense of item retention, leading to the abolition of item-based phenomena. In a subsequent study, however, Guérard and Saint-Aubin (in press) showed that four lexical factors known to modulate item retention were unaffected by recall direction. In a series of five experiments, we examined the source of the discrepancy between the two studies. We revisited the effects of phonological similarity, word length, articulatory suppression, and irrelevant speech, using open and closed pools of words in backward and forward recall. The results are unequivocal in showing that none of these effects are influenced by recall direction, suggesting that Bireta et al.’s (Memory & Cognition 38:279–291, 2010) results are the consequence of their particular stimuli.     

Reading aloud: when the effect of stimulus quality distinguishes between cascaded and thresholded components

A central feature of many formal accounts of reading aloud, and of Coltheart and colleagues dual-route cascaded model in particular, is that activation across various modules is cascaded. Evidence is reviewed that this assumption is problematic in a particular context, along with a solution that involves thresholding the output of the letter level to the nonlexical routine. Consideration of the known effects of repetition leads to the prediction of a three-way interaction between stimulus quality, repetition, and lexicality in which repetition and stimulus quality interact when reading aloud exception words, but produce additive effects when reading aloud nonwords. The result of such an experiment confirms this prediction, and appears consistent with the localized dual-route model. Implications for other accounts are briefly noted.     

Semantic priming occurs for word but not location pronunciation in the postcue task

Semantic priming refers to the finding that a word response is facilitated if it is preceded by a related word compared to when it is preceded by an unrelated word. Dallas and Merikle (Can J Psychol 30: 15–21 1976a; Bull Psychon Soc 8: 441–444 1976b) demonstrated that semantic priming occurred under conditions in which a pair of simultaneously displayed words was previewed for over a second prior to the onset of a cue indicating which of the words should be pronounced aloud (postcue task). In contrast, semantic interference effects have been reported for postcue picture-naming tasks (Dean et al. in J Exp Psychol Learn Mem Cogn 27: 733–743, 2001; Humphreys et al. in J Exp Psychol Learn Mem Cogn 21: 961–980, 1995). According to Dean et al., the semantic interference effects in postcue picture naming occur because the integration of the object and the cued attribute in memory is more difficult for categorically related pictures than for unrelated pictures. The aim of this experiment was to determine whether this idea was true for postcue word pronunciation tasks. Participants completed two postcue tasks, one requiring pronunciation of the target word indicated by a locational cue and another requiring pronunciation of the location of a centrally presented word. Results indicated a semantic priming effect only for the locational cue condition suggesting that the integration of the cue and identity information was unaffected by word context. These data suggest that priming in a postcue word pronunciation task may be due to feedback from residual activation within the semantic system facilitating access to the target word’s phonology.     

Reading without phonology?

Responses to items such as brane are slower and/or more error prone than responses to items such as slint in lexical decision (is this string spelt like a real word?). The received view is that this “pseudohomophone” effect is attributable to phonological receding. Taft (1982) has challenged this view, offering instead a grapheme-grapheme account which assumes that graphemes that map onto a common phoneme develop the ability to activate each other without reference to phonological mediation.

Taft's grapheme-grapheme account is tested in two experiments. Experiment 1 shows that the presentation of a pseudohomophone facilitates the response to a subsequently presented word (e.g., groce-gross). Experiment 2 shows that nonword letter strings that are translatable into words by the application of putative grapheme-grapheme rules (e.g., gloce-gloss) produce no facilitation. These results are consistent with the notion of a phonological influence but inconsistent with the grapheme-grapheme account. Loci for this pseudohomophone priming effect are discussed.     

I said, you said: the production effect gets personal

Saying a word out loud makes it more memorable than simply reading it silently. This robust finding has been labeled the production effect and has been attributed to the enhanced distinctiveness of produced relative to unproduced items (MacLeod et al. Journal of Experimental Psychology: Learning, Memory, and Cognition, 36, 671–685, 2010). Produced items have the additional information that they were spoken aloud encoded in their representations, and this information is useful during retrieval in certifying prior encoding. The present study explored whether production must be self-performed to be beneficial, or whether another person’s production also makes an item more memorable. In two experiments, the production effect was shown to be reliable when production was done by someone other than the rememberer (i.e., by the experimenter or by another participant), but substantially smaller than the benefit from self-performed production. Intriguingly, the effect was intermediate when production was done by both the rememberer and another person. Distinctiveness—and hence the production effect—is greatest to the extent that it is personal.     

Does the familiarity bias hypothesis explain why there is no masked priming for “NO” decisions?

Traditional "activation" views of masked priming explain the identity priming effect in terms of facilitation due to ‘pre-activation’ of stored representations. Norris and Kinoshita's (2008) Bayesian Reader theory of masked priming instead explains priming in terms of the evidence that the prime contributes towards the decision required to the target. In support of the Bayesian Reader account, Norris and Kinoshita showed that the absence of priming for nonwords in the lexical decision task and for targets requiring a Different decision in the same–different match task can be explained based on a single principle. Against this, Bowers (2010) argued that the absence of priming should be explained instead by a combination of sublexical priming and "familiarity bias". As evidence, Bowers cited Bodner and Masson’s (1997) finding that nonword priming did emerge with targets presented in visually unfamiliar cAsE-AlTeRnAtEd format. We present evidence that this finding was due to the use of an ambiguous letter in case-alternated format; when using unambiguous letters, we consistently failed to find priming of case-alternated nonwords. We suggest that the Bayesian Reader, rather than the familiarity bias hypothesis, explains the absence of priming.     

Morphological decomposition and the reverse base frequency effect

If recognition of a polymorphemic word always takes place via its decomposition into stem and affix, then the higher the frequency of its stem (i.e., base frequency) the easier the lexical decision response should be when frequency of the word itself (i.e., surface frequency) is controlled. Past experiments have demonstrated such a base frequency effect, but not under all circumstances. Thus, a dual pathway notion has become dominant as an account of morphological processing whereby both decomposition and whole-word access is possible. Two experiments are reported here that demonstrate how an obligatory decomposition account can handle the absence of base frequency effects. In particular, it is shown that the later stage of recombining the stem and affix is harder for high base frequency words than for lower base frequency words when matched on surface frequency, and that this can counterbalance the advantage of easier access to the higher frequency stem. When the combination stage is crucial for discriminating the word items from the nonword items, a reverse base frequency effect emerges, revealing the disadvantage at this stage for high base frequency words. Such an effect is hard for the dual-pathway account to explain, but follows naturally from the idea of obligatory decomposition.     

Anti-Luck Epistemology and Safety’s (Recent) Discontents

Anti-luck epistemology is an approach to analyzing knowledge that takes as a starting point the widely-held assumption that knowledge must exclude luck. Call this the anti-luck platitude. As Duncan Pritchard (2005) has suggested, there are three stages constituent of anti-luck epistemology, each which specifies a different philosophical requirement: these stages call for us to first give an account of luck; second, specify the sense in which knowledge is incompatible with luck; and finally, show what conditions must be satisfied in order to block the kind of luck with which knowledge was argued to be incompatible. What I’ll show here is that the modal account of luck offers a plausible story at the first stage and leads naturally to equally plausible lines to take at the second and third stages, at which a safety condition on knowledge is squarely motivated. There are, however, recent challenges—advanced by Jonathan Kvanvig (Philosophy and Phenomenological Research 77: 272–281, 2008); Kelly Becker (2007); and Jennifer Lackey (Australasian Journal of Philosophy 86(2):255–267, 2008), among others—to the plausibility of the safety-based anti-luck project I’ve sketched here at each of its three stages of development. Once I’ve made precise the challenges, I’ll show why none implies that we abandon the commitments of the safety-based anti-luck project at any of its stages. What we should conclude, then, is that a safety-condition on knowledge is motivated by independently defensible accounts of (1) what luck is; and (2) just how knowledge should be thought incompatible with it.