Early morphological processing is sensitive to morphemic meanings: Evidence from processing ambiguous morphemes

https://doi.org/10.1016/j.jml.2012.11.003Get rights and content

Abstract

In three priming experiments, we investigated whether the meanings of ambiguous morphemes were activated during word recognition. Using a meaning generation task, Experiment 1 demonstrated that the dominant meaning of individually presented ambiguous morphemes was reported more often than did other less frequent meanings. Also, participants tended to produce responses that were consistent with the morphemic meaning of the subliminally presented prime words. Experiment 2 employed a masked priming lexical decision task (prime display duration = 40 ms) and showed that the recognition of targets which took the dominant meaning of ambiguous morphemes was facilitated by all morpheme-sharing primes, regardless of their intended interpretation. In contrast, morphological priming for subordinate targets was observed only in the subordinate priming condition. Using an unmasked priming task (prime display duration = 100 ms). Experiment 3 revealed that lexical decision responses were facilitated only when the morphemic interpretations in primes and targets were matched. These data indicate that the different meanings of an ambiguous morpheme are activated early during word recognition and that it takes time to select the appropriate morphemic interpretation. The results are discussed with reference to a modified lemma model of word recognition.

Highlights

► We examined the role of morpho-semantics in word recognition using ambiguous morphemes. ► Priming of dominant targets was independent of morphemic meaning in primes. ► Priming of subordinate targets was observed only when primes take the subordinate meaning. ► Morphemic meaning was activated at the early stage of word recognition. ► We modified the lemma model of word recognition to accommodate the present results.

Introduction

Human language is characterized by an arbitrary linkage between symbols and meanings. For instance, there are no obvious reasons for the letter sequence “c-o-o-k” to be associated with the action of preparing food. Morphemes, however, can help regularize the form-meaning association, as words that share the same morpheme are usually related in meaning (e.g., “cooker”, “cookery”, and “cookbook”; Lavric et al., 2007, Spencer, 2001). Indeed, language users can even combine existing morphemes in creative ways to express novel ideas, such as “facebook” (“face” + “book”) and “unfriend” (“un-” + “friend”). Previous studies have found a reliable morpheme effect on word recognition (e.g., Crepaldi et al., 2010, Marslen-Wilson et al., 1994, Rastle et al., 2004, Taft, 1994, Tsang and Chen, 2010). Moreover, the effects of morphemes cannot be reduced to formal or conceptual processing at the lexical level (Feldman, 2000, Rastle et al., 2000), supporting a unique and independent role of morphemes in recognizing words. Although it is now a standard practice to incorporate morphemes into theories of word recognition, many details about morphological processing are unsettled. For instance, it remains controversial whether early morphological processing relies purely on surface orthographic structure (i.e., the morpho-orthographic view) or morphemic meaning (i.e., the morpho-semantic view; Bertram et al., 2011, Davis and Rastle, 2010, Feldman et al., 2009, Rastle and Davis, 2008).

One popular approach in testing the relative importance of morpho-orthographic and morpho-semantic processing in word recognition is to investigate the semantic transparency effect. This is commonly done by comparing the strength of morphological priming produced by transparent words (e.g., “departure”; the stem “depart” is semantically related to the whole-word) against that produced by opaque words (e.g., department; the stem “depart” is semantically unrelated to the whole-word). The presence of semantic transparency effect has been taken as evidence of morpho-semantic processing during word recognition. Employing a masked priming lexical decision procedure, Frost, Forster, and Deutsch (1997) demonstrated that prior exposure to both transparent and opaque words could facilitate subsequent target word recognition in Hebrew. Moreover, the strengths of facilitation in the two conditions were statistically equivalent, indicating the absence of morpho-semantic constraints. Similar results have also been obtained in English (Marslen-Wilson et al., 2008, Rastle et al., 2004), French (Longtin, Segui, & Hallé, 2003), Spanish (Duñabeitia, Perea, & Carreiras, 2008), and Russian (Kazanina, Dukova-Zheleva, Geber, Kharlamov, & Tonciulescu, 2008), using not just behavioral (reaction times and error rates), but also neurophysiological measures (fMRI: Gold & Rastle, 2007; ERP: Lavric et al., 2007). The absence of morpho-semantic effects in these studies led Rastle and Davis (2008; also see Taft & Forster, 1975) to propose an obligatory morphological analysis driven entirely by surface morpho-orthographic structure in word recognition: Letter sequences that look like morphemes will be segmented, regardless of whether they contribute to word meanings or not. Therefore, both transparent and opaque words will be decomposed, but monomorphemic words like “brothel” is analyzed holistically because “-el” is not a legal suffix in English.

According to Rastle and Davis (2008), morphemic meaning only plays a role in lexical access at the later stage of word recognition, where “illegal” or meaningless combinations of morphemes (i.e., opaque words) will terminate morphological processing and trigger semantic re-analysis (Lavric, Rastle, & Clapp, 2011). This hypothesis appears to agree well with studies that revealed a significant semantic transparency effect. For example, Marslen-Wilson et al. (1994) employed the cross-modal priming paradigm to compare the strengths of morphological priming by transparent and opaque primes (e.g., “departure” vs. “department”). In six experiments, facilitation was found only in the transparent condition. Opaque primes either did not influence target word detection or led to inhibition. Morpho-semantic constraints were also found in the comprehension of novel French words, such that morphological priming occurred in processing interpretable novel words (e.g., “rapidifier” and “installage), but not non-interpretable ones (e.g., “sportation”; Meunier & Longtin, 2007). In addition, morpho-semantic activation could be observed in a visual-visual intra-modal priming experiment (Sandra, 1990), in which an explicit semantic relationship between the constituent morpheme and the prime word (e.g., “bread–buttercup”) led to facilitation in lexical decision. Contrary to the masked priming experiments, participants in these studies were fully aware of the primes because they were presented either auditorily or visually for a longer duration (>60 ms). Differences between transparent and opaque words in these studies were thus reflective to later stages of word recognition, confirming the two-stage mechanism that Rastle and Davis proposed.

On the other hand, a number of recent studies have shown that the early stage of morphological processing does not necessarily involve morpho-orthography only. For instance, masked morphological priming appears to survive orthographic changes between primes and targets (e.g., bought – buy, Crepaldi et al., 2010; fetish – fete, McCormick, Rastle, & Davis, 2008). More importantly, in some experiments, the strength of facilitation produced by transparent words was stronger than that by opaque words (Diependaele et al., 2005, Morris et al., 2007). Indeed, as illustrated by Feldman, O’Connor, and Moscoso del Prado Martín (2009), in at least 75% of the published masked priming experiments, there is a consistent trend for stronger facilitation in the transparent than in the opaque condition. When data across studies were pooled, the difference between transparent and opaque conditions actually reached statistical significance (but see Davis & Rastle (2010) for alternative interpretations). These data thus provide support for the idea that early morphological processing is not just a morpho-orthographic event but also a morpho-semantic one.

The involvement of both morphemic form and meaning in early morphological processing is consistent with the distributed connectionist approach to morphology (e.g., Plaut and Gonnerman, 2000, Seidenberg and Gonnerman, 2000), which suggests that a morphological representation emerges due to the stable correlation between form and meaning among words within the same morphological family. In other words, morphology can be coded as an intermediate level of representation between orthography and semantics. While being a localist model in nature, a similar idea is adopted in the hierarchical framework of word recognition proposed by Taft (2004; Taft & Nguyen-Hoan, 2010; also see Crepaldi et al., 2010). According to this model, morphology is coded at the “lemma” level, an abstract layer of representations situated between form and functional information (e.g., semantic and syntactic features). Morpho-semantic priming can thus be attributed to the pre-activation of the same lemma shared between prime and target. Moreover, given the emerging nature of these morphological representations, the model hypothesizes the following: (a) Ambiguous morphemes (e.g., “stick”, which means “adhere” or “twig” depending on the context) have distinct lemmas for different meanings. (b) The interpretation of an ambiguous morpheme will be biased towards the previously activated meaning during prime word processing. (c) The morphological processing of a target word will be facilitated more strongly by the prior exposure to a prime word containing the ambiguous morpheme with the same interpretation (i.e., sharing both form and lemma) than that with a different interpretation (i.e., sharing form only).

These predictions have received support in the meaning generation task by Taft and Nguyen-Hoan (2010). In their experiment, participants were instructed to report the first meaning that came to their mind towards isolated ambiguous morphemes like “express” (related to transportation or communication). Each target was preceded by a masked prime, which was derived from one of the alternative morphemic meanings (e.g., “expression”), a semantic associate of the same interpretation (e.g., “phrase”), or an unrelated control. Compared with the baseline control, participants were more likely to produce a meaning related to the prime in the morphemic, but not in the semantic conditions. Given that the different meanings of an ambiguous morpheme cannot be differentiated on the basis of the surface form, and that the semantic prime did not produce a comparable effect as the morphemic prime, the morpho-semantic priming observed in the morphemic condition could only be attributed to the operations at the lemma level. While the lemma model was used to explain the morpho-semantic priming observed in the meaning generation task, in principle it can be extended to accommodate the results of opaque word facilitation observed in previous studies of masked priming lexical decision (e.g., Rastle et al., 2004). In particular, the model assumes that perceiving a pseudo-complex word (e.g., “corner”) or an opaque word (e.g., “department”) would activate the lemma representing the stem of the word (e.g., “corn” or “depart”). Subsequent processing of the stem or other transparent words containing the stem would thus be facilitated by the opaque word due to activation at the lemma level (Fig. 4 in Taft & Nguyen-Hoan, 2010).

Although the lemma model can presumably account for findings related to the manipulations of both semantic transparency and morphemic ambiguity, no existing studies have directly compared the morpho-semantic processing produced by the two manipulations. However, this line of research is important because a better understanding of the nature of morpho-semantic processing can be obtained through comparing the effects produced by different morpho-semantic manipulations. The lack of relevant research may be attributed to the difficulty in finding ambiguous morphemes that also appear in opaque words in English (e.g., triplets like “inbox”, “incapable”, and “increase”, which correspond to two distinct morphemic meanings and an opaque use of the prefix “in-”, are extremely rare). On the other hand, these cases are much more common in Chinese (see Materials in Experiment 1 and Appendix A), making it ideal for examining the issue. Note also that morphemic boundaries are clearly marked by spaces in written Chinese (see Chen, 1992 for a detailed description; also see Table 1 for examples). Consequently, morphological segmentation is a relatively easy and probably automatic task in Chinese word recognition, which may increase the salience of morpho-orthographic information. On the other hand, it may also speed up the processing of morphemic form and may make morpho-semantic processing to start relatively fast (see Zhou, Marslen-Wilson, Taft, & Shu, 1999). In order to better understand the processing of morpho-semantic information in word recognition and to develop a more universal model of language processing, the present study was conducted to examine whether the lemma model can explain the processing of ambiguous morphemes and opaque words in Chinese.

The present study also attempts to clarify the inconsistent conclusions and results that have been reported across studies. Specifically, while the meaning generation experiment with the masked priming procedure clearly illustrated the presence of morpho-semantic priming even when participants were unaware of the primes (Taft & Nguyen-Hoan, 2010), it was demonstrated by Badecker and Allen (2002) in a lexical decision experiment that the recognition of Spanish words containing ambiguous morphemes was facilitated by prime words that contained the same morphemic form regardless of the correct interpretations. For example, “cerrar” (to close) could facilitate the recognition of the target “cerro” (hill) even when the stem “cerr-” takes different interpretations in the two words. Their results thus appear to be inconsistent with those of Taft and Nguyen-Hoan and instead appear to provide support for the pure form-based account of early stage morphological priming.1 The inconsistency in the previous findings may be due to the fact that different tasks were adopted in the mentioned studies. In order to assess the possible influence of task demand on morphemic processing, the present study adopted both the meaning generation task (in Experiment 1) and the lexical decision task (in Experiments 2 and 3) and compared and contrasted the results obtained with the two tasks.

Another potential source of contradictory results on morpho-semantic processing concerns the correct interpretations of ambiguous materials. Although Badecker and Allen (2002) did not provide much information about this, the primes in Taft and Nguyen-Hoan (2010) were biased towards the subordinate morphemic meaning. Indeed, one of the most frequently reported findings in the ambiguity resolution literature is the meaning frequency effect, which refers to a strong activation of the dominant meaning even when context supports the subordinate one (Duffy et al., 1988, Tsang and Chen, 2010). If the target words in Badecker and Allen were mainly biased towards the dominant morphemic meaning, their materials could be less able to differentiate morpho-semantic effects in the various conditions because the prior exposure to an ambiguous morpheme may activate the dominant meaning even when it is not the intended interpretation. In order to evaluate the possible influences of meaning frequency on morpho-semantic processing, the materials used in the present study were developed based on both dominant and subordinate meanings. The results so obtained could help connect the lemma model with the mechanism of ambiguity resolution, which has been extensively studied at the lexical level.

Finally, previous studies of semantic transparency (Rastle & Davis, 2008) have shown that different stages of processing can be examined by varying the time allowed for prime word processing. Therefore, the same materials in Experiment 2 were used in another lexical decision experiment (i.e., Experiment 3) with a longer duration of prime presentation (i.e., 100 ms vs. 40 ms in Experiment 2) to ensure conscious processing of the primes. Comparing the results of Experiments 2 and 3 should be able to provide important information about the time course of activation of the dominant and subordinate meanings of an ambiguous morpheme and how the appropriate meaning is selected (i.e., the process of morphemic ambiguity resolution; Tsang & Chen, 2010).

Section snippets

Experiment 1

The first experiment adopted the meaning generation task to investigate how morphemic ambiguity and semantic transparency affect morphological processing. In addition, the lemma model proposed by Taft, 2004, Taft and Nguyen-Hoan, 2010 was used as our theoretical framework. This model assumes that morphological priming originates not just from morpho-orthographic sharing but also from morpho-semantic sharing at the lemma level. Based on this model, we expected that meaning generation would

Experiment 2

The results of Experiment 1 are consistent with those of Taft and Nguyen-Hoan (2010) and the predictions of the lemma model. Given the importance of these results in developing models of word recognition and morphemic ambiguity resolution, it is desirable to obtain converging evidence of the morpho-semantic effect from another task. In Experiment 2, we employed the lexical decision task (Badecker & Allen, 2002) to examine whether the morpho-semantic priming and meaning frequency effect observed

Experiment 3

The results of Experiment 2 are consistent with those obtained in Experiment 1 in supporting the presence of morpho-semantic priming in word recognition and the idea that the frequency of an intended morphemic interpretation can modulate the pattern of priming in processing a compound word. While masked priming has been used to examine early stages of processing, it is also important to specify the time window within which the different meanings of an ambiguous morpheme are available.

General discussion

The relative importance of morpho-orthography and morpho-semantics at the early stage of word recognition remains a controversial issue in the psycholinguistic literature. While some researchers argued that only morphemic form is relevant (e.g., Frost et al., 1997, Rastle and Davis, 2008, Rastle et al., 2004), others have shown that morphemic meaning is also an important component (e.g., Diependaele et al., 2005, Feldman et al., 2009, Morris et al., 2007). These studies have focused mainly on

Acknowledgments

We thank Sally Andrews, Marcus Taft, and two anonymous reviewers for constructive comments and suggestions on an earlier version of the manuscript. This research was supported by the research Grant 2021016 from The Chinese University of Hong Kong.

References (56)

  • F. Meunier et al.

    Morphological decomposition and semantic integration in word recognition

    Journal of Memory and Language

    (2007)
  • J. Rodd et al.

    Making sense of semantic ambiguity: Semantic competition in lexical access

    Journal of Memory and Language

    (2002)
  • M.S. Seidenberg et al.

    Explaining derivational morphology as the convergence of codes

    Trends in Cognitive Sciences

    (2000)
  • D.A. Swinney

    Lexical access during sentence comprehension: (Re)consideration of context effects

    Journal of Verbal Learning and Verbal Behavior

    (1979)
  • M. Taft et al.

    Lexical storage and retrieval of prefixed words

    Journal of Verbal Learning and Verbal Behavior

    (1975)
  • R.H. Baayen

    Analyzing linguistic data: A practical introduction to statistics

    (2008)
  • R. Bertram et al.

    The role of context in morphological processing: Evidence from Finnish

    Language and Cognitive Processes

    (2000)
  • R. Bertram et al.

    Morphology in language comprehension, production, and acquisition

    Language and Cognitive Processes

    (2011)
  • R. Bertram et al.

    The balance of storage and computation in morphological processing: The role of word formation type, affixal homonymy, and productivity

    Journal of Experimental Psycholgoy: Learning, Memory, and Cognition

    (2000)
  • S.W. Brown

    Polysemy in the mental lexicon

    Colorado Research in Linguistics

    (2008)
  • Q. Cai et al.

    SUBTLEX-CH: Chinese word and character frequencies based on film subtitles

    PLoS ONE

    (2010)
  • M.H. Davis et al.

    Form and meaning in early morphological processing: Comment on Feldman, O’Connor, and Moscoso del Prado Martín (2009)

    Psychonomic Bulletin & Review

    (2010)
  • K. Diependaele et al.

    Masked cross-modal morphological priming: Unravelling morpho-orthographic and morpho-semantic influences in early word recognition

    Language and Cognitive Processes

    (2005)
  • J.A. Duñabeitia et al.

    Does darkness lead to happiness? Masked suffix priming effects

    Language and Cognitive Processes

    (2008)
  • L.B. Feldman

    Are morphological effects distinguishable from the effects of shared meaning and shared form?

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2000)
  • L.B. Feldman et al.

    Early morphological processing is morphosemantic and not simply morpho-orthographic: A violation of form-then-meaning accounts of word recognition

    Psychonomic Bulletin & Review

    (2009)
  • K.I. Forster et al.

    DMDX: A Windows display program with millisecond accuracy

    Behavior Research Methods, Instruments, & Computers

    (2003)
  • R. Frost

    Towards a universal model of reading

    Behavioral and Brain Sciences

    (2012)
  • Cited by (48)

    • Semantic Transparency in Chinese Compound Word Processing: Evidence from Mismatch Negativity

      2022, Neuroscience
      Citation Excerpt :

      Support for holistic processing has come from auditory lexical decision experiments, where word frequency, but not morpheme frequency, effects were found (Zhou and Marslen-Wilson, 1994). In contrast, results from morphological priming and visual-world experiments showed that morphemes play a role in Chinese word recognition (Tsang and Chen, 2010, 2013; Huang et al., 2020). Further research is needed to examine whether the psychological validity of morphemes in spoken word processing is language universal.

    • The role of semantics and orthography in modulating conscious access to Chinese words

      2022, Lingua
      Citation Excerpt :

      Thus, we assume that the absence of priming in our study is caused by the larger cognitive load of compound words with shared morphemes. As indicated in psycholinguistic studies, the processing of Chinese double-character words involves both the morphemic form and meaning in each character, as well as the spatial or temporal integration of two characters and the interaction between a single character and the whole word (Tsang and Chen, 2013; Tsang et al., 2013; Tsang et al., 2014; Zhou et al., 1999). However, a large number of backward masking studies have demonstrated that the priming effect of the identical-morpheme condition [S+O+, 华贵-华丽 (luxurious-gorgeous)] is larger than that of the homomorphic condition [S-O+, 华贵-华裔 (luxurious- ethnic Chinese)] (Wu et al., 2017).

    View all citing articles on Scopus
    View full text