[PDF] Seeing through issues of semantic transparency

View - Download Seeing through issues of semantic transparency

Previous PDF

Next PDF

Copyedited by RMD/KP, CJL/RCL 2005-088

, 13 April 2006 1 Everything is Psycholinguistics: Material and Methodological Considerations in the Study of Compound Processing

GARY LIBBEN

University of Alberta

The specific domain that I discuss in this paper is the processing of compound words. This research area is no more than 30 years old and my own involvement in it has been for about half that time. The study of how compound words are processed is a very specific sub-field, considerably narrower than the domains of which we would normally consider. Yet, for a variety of reasons which I discuss below, compound processing research finds itself associated with issues that are central to our views of the funda mental nature of lexical processing, and the enterprise of psycholinguistic research itself. These are certainly the domains of which we do want to take stock. My goal in this article is to address issues that are of general interest th rough the medium of research on the representation and processing of compound words. In considering the development, current state, and likely future trajectory of a specific research domain, this provides us with an opportunity to gain a broader perspective on the field as a whole and to step outside (and hopefully above) the normal course of research activity. This research reported here was supported by a Major Collaborative Research Initiative Grant from the Social Sciences and Humanities Research Council of Canada (SSHRC) to Gary Libben (Director), Gonia Jarema, Eva Kehayia, Bruce Derwing, and Lori

Buchanan.

Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 2

I begin by considering the role that the

construct of "morpheme" plays in lexical processing. Various kinds of evidence indicate that compound words are parsed into morphemes during processing, so that the mental lexicon can be said to have morphological representations (section 1). The finding that morphological parsing automatically activates all possible constituents challenges the traditional distinction between semantically transparent versus semantically opaque compounds (section 2). This in turn leads to a reappraisal of standard experimental techniques (section 3). Thus, the apparently parochial concerns that arise in the course of investigating the processing of compound words turn out to have dramatic consequences for what we take to be our object of study, and how we study it. 1. WHY ARE PSYCHOLINGUISTS INTERESTED IN COMPOUND WORDS? Compounds words are, by definition, multi-morphemic. As such they have a dual life: we can consider the meaning of the compound word as a whole ("whole-word meaning") or we can examine the meanings of the cons tituent parts of the compound ("constituent meaning"). Evidence from aphasia indicates that speakers manipulate both types of meanings (section 1.1), and this in turn has implications for the interplay between lexical storage and morphological computation (section 1.2). Of particular interest are semantically opaque compounds, whose whole-word meanings are incompatible with their constituent parts (section 1.3). Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 3

1.1. Whole word versus constituent meanings: A case study from aphasia

In Libben (1993, 1998), I discussed the case of a patient, RS, who had become aphasic after suffering a brain hemorrhage in the temporoparietal area of the left hemisphere. In a paraphrase task, RS was asked to provide meanings for a set of compound words that included items such as butterfly, yellowbelly, and dumbbell. These items can be described as semantically opaque because the meanings of the whole words are not aligned in a straightforward manner with the meanings of their constituent morphemes. RS's paraphrase patterns for these words were striking, and are given in (1). (1) RS'S PARAPHRASES FOR SOME COMPOUND WORDS

COMPOUND WORD RS'S PARAPHRASE

butterfly "a pretty fly...it's yellow" yellowbelly "a yellow stomach...a chicken" dumbbell "stupid weights" (followed by a reference to Arnold Schwarzenegger) In all these cases, it seemed clear that RS was accessing the whole-word meanings of the compounds. For non-entomologists, most flies are not pretty, but butterflies are. Only chicken, meaning coward, refers to the meaning of yellowbelly as a whole, not to the colour of a stomach. Only dumbbells, but not other types of bells, are types of "weights". However, in addition to this demonstrated access to the whole-word meanings of opaque compounds, RS's paraphrases also seemed to show access to the independent meanings of constituents. The association of "yellow" with butterfly seems independently related to the colour of butter. The association of "stomach" to yellowbelly is clearly related to the independent meaning of belly. Finally, the association of "stupid" to dumbbell seems also to indicate independent access to constituent meaning. (As for the Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 4 reference to Arnold Schwarzenegger, I leave it to the reader to decide whether this reflects activation of whole-word or constituent meaning.) The result of this blending of whole-word and constituent meanings had a negative effect on RS's lexical comprehension, and to that extent her lexical processing was clearly impaired. But what was not at all clear was whether the processes that she was employing in lexical access were qualitatively different from the processes that non- impaired native speakers of English employ in the everyday processing of such words. Is it possible, for example, that the processing of compounds always involves automatic and obligatory access to both whole-word and constituent meanings in a relatively independent manner? If this were the case, it would suggest that RS's problem was not that she created multiple incompatible representations, but rather that she simply did not resolve the conflict that they created, as non-impaired language users do.

1.2 The trade-off between storage and computation

The possibility that lexical processing involves automatic and obligatory online morphological parsing is what has drawn psycholinguists to the study of how multi- morphemic words, and compound words in particular, are processed. It should be mentioned at the outset that this new attention to multi-morphemic word processing comes at a methodological price. The current techniques of investigation in lexical processing are extremely sensitive to a great number of variables that need to be controlled in the design of experiments. Because multi-morphemic words are, by definition, more complex than monomorphemic words, they present new challenges to Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 5 experimental control, and accordingly make the design and analysis of experiments more difficult. Despite these methodological challenges, the study of multi-morphemic words opens up opportunities that cannot be ignored. First among these is the opportunity to understand the cognitive trade-offs that exist between lexical storage and morphological computation. Languages do not participate equally in the use of word-formation morphology. Those languages that do make extensive use of derivation and compounding derive benefit in terms of the language's root inventory size because new words can be created without having to cr eate new morphemes. Many psycholinguists who investigate the processing of multi-morphemic words ask whether morphological processing yields a similar benefit by obviating the need for all multi-morphemic words to have individual representations in the mental lexicon. Issues surrounding this question have received extensive attention, most notably in the domain of inflectional morphology (Pinker 1999; Pinker and Ullman 2002). The potential contribution of research on compound processing lies in its rather unique position at the crossroads of storage and computation (Libben

2006). On the one hand, there seems to be a way in which morphological parsing must

occur for compounds. It is an extremely productive word-formation process in English, and even more so in related languages such as German. Thus, it is quite likely in both these languages that native speakers encounter novel compounds on a daily basis. Indeed, even when compounds are not novel in the language, they are quite likely to be novel to individual language users because of their low overall frequency of occurrence. As an illustration of this, consider the 1,437 noun-noun compounds which are currently listed in the English CELEX database (Baayen et al. 1993). Over half of these (836) have a Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 6 written frequency of less than one in a million and only 35 have a frequency of over ten in a million. By contrast, words such as dog and cat have frequencies of roughly 75 and

25 in a million, respectively.

To the extent that speakers of a language can interpret novel compound structures, they do so through the isolation and interpretation of constituent morphemes. This is not to say that their interpretations are guaranteed, or even likely, to be correct. The vast majority of compound words cannot be understood solely on the basis of their constituent morphemes. Inasmuch as members of a language community share particular meanings for compound words - rather than possible meanings for those words - storage is required. Seen in this perspective, compound words are rather unique structures in that they seem to require a morphological parsing strategy that is always on, as well as whole-word representations that are always available. Understanding how these features interact has been the focus of my research.

1.3. The significance of semantically opaque compounds

Semantically opaque words offer an excellent opportunity to enhance our understanding of the interplay between storage and morphological computation. These are words for which morphological processing will yield results that are incompatible with whole-word meanings. This means that, if retained, the whole-word meanings would generate false comprehensions of the kind that were characteristic of RS's paraphrases discussed above. It is of psycholinguistic interest to note that semantically opaque compounds are not Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 7 nearly as exotic as we might think. While we might be somewhat puzzled by the possible referents for uncommon fruit and vegetable terms such as stonefruit, hackberry, cowpea, and plumapple, the semantic opacity of well-known fruit and vegetable compounds such as grapefruit, strawberry, chickpea, and pineapple scarcely gets noticed. The question of whether such words throw the lexical processing system into some degree of momentary disarray has been the subject of considerable research. As a final example of the phenomena that make compound processing interesting psycholinguistically and reveal the nature of morphological processing, let us return to the uncommon fruit and vegetable terms stonefruit, hackberry, cowpea, and plumapple. While most native speakers would not know what these words mean, very few would have difficulty parsing them into their constituents: stone+fruit, hack+berry, cow+pea, and plum+apple. A good deal of the research that I have been involved in over the past years has been focused on trying to determine how this online morphological parsing in reading is actually accomplished. Compound words present a much greater challenge to online morphological parsing than do affixed words. The difference is that affixes comprise a closed-class set in the language, so that morphological parsing can be accomplished by stripping affixes from their stems, as originally suggested by Taft and

Forster (1975). Compound words, on the othe

r hand, are formed, for the most part, from members of open-class sets. Thus, there are no reliable heuristics that can be employed to ensure that morpheme boundaries are correctly determined for stone+fruit, hack+berry, cow+pea, and plum+apple. Yet, native speakers of English do this easily and Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 8 automatically, while not over-parsing words such as carpet and barking into car+pet or bar+king.

2. THE HUNGRY LEXICON

Together with students and colleagues, I have had the opportunity to conduct experimental investigations on aspects of compound processing in Chinese, Dutch, English, French, German, Hebrew, and Japanese. What I would like to draw attention to is the general conclusion that emerges from this research, namely that the morphological processing system is not designed to get the right answer. It is not organized to generate exactly the correct morphological parse, nor is it organized to generate exactly the correct lexical excitation. Rather, it is designed to make the greatest number of potentially useful representations and analyses available to other components of the cognitive system: in this sense then, the lexicon may be said to be hungry. The robustness of compound processing is perhaps most clearly illustrated by looking at how novel compounds are parsed (section 2.1), which has led us to re-assess our understanding of how semantic transparency facilitates word recognition (sections 2.2 and 2.3).

2.1. Ambiguous novel compounds

The first clue that morphological parsing is opportunistic arose from a set of studies that employed a rather unusual set of stimuli (Libben 1994; Libben et al. 1999). These stimuli can most appropriately be described as ambiguous novel compounds. They are words such as clamprod , seathorn, and cartrifle, which were constructed so that they could be Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 9 parsed in two ways. In the case of clamprod, for example, the two parses would be clam+prod and clamp+rod. The initial purpose for the construction of these stimuli was to understand the properties of the putative morphological parser. I reasoned that if morphological parsing proceeds in a beginning-to-end manner, searching for morphemic representations along the way, ambiguous novel compounds such as clamprod would all be parsed with a morpheme boundary at the first opportunity (i.e., as clam+prod), generating a first possible parse. If on the other hand, the parser proceeded in a beginning-to-end manner until the largest lexical unit could be created, then the structures would most likely be given a last possible parse (i.e., as clamp+rod). In an initial experiment, participants were provided with a list of novel compound stimuli in which these special items were embedded. They were asked to read them aloud as clearly and carefully as possible, so that their speech could be used as materials for learners of English as a second language. This clear and careful speech created identifiable pauses at morpheme boundaries, and the location of these pauses was recorded as the dependent variable in the experiment. The results of the experiment revealed that digraphs such as the th in seathorn substantially constrained parsing choices. In contrast to this, for words without digraphs, results favoured neither the first possible parse nor the last possible parse alternatives. Most importantly, parsing choices were correlated with the semantic plausibility of alternative parses so that a string such as cardriver would be parsed as car+driver, but a string such as cartrifle, would be parsed as cart+rifle. The fact that both these strings Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 10 begin with the possible initial morpheme car suggests that the choices were not driven by initial-constituent frequency, but rather by the semantic fit between constituents. The fact that semantic plausibility turned out to be the driving factor in parsing choices led to an important paradox. How could the relative semantic plausibility of the alternative parses drive the choice of morphological parse, when the calculation of such plausibility would require that the activation of constituents had already occurred? It seemed that the only possible resolution of this paradox would require that at the time of parsing choice - when participants had read the words aloud clearly and carefully - both alternative parses had been completed, and the semantically more plausible one was selected. This conclusion was counter-intuitive given the fact that during debriefing, when queried, only one of the 30 experiment pa rticipants indicated that she had noticed that some of the stimuli could be read in two ways. If ambiguous novel compounds were in fact receiving both parses automatically, they should take longer to process than matched non-ambiguous strings: for example, clamprod (consistent with either clam+prod or clamp+rod) should take longer to process than prodclam (consistent only with prod+clam). This prediction, which was tested in a follow-up experiment, turned out to be correct. In a lexical decision experiment, 30 participants were shown real compounds and novel compounds in the centre of a computer screen and were asked to judge as quickly as possible whether the stimuli were real words of English by pressing either a "yes" or a "no" key. It took participants significantly longer to reject the ambiguous novel compounds than to reject the unambiguous ones. The effect size was large, with Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 11 unambiguous stimuli showing mean rejection times of 538 milliseconds and the ambiguous stimuli showing rejection latencies of 670 milliseconds - over 130 milliseconds slower. Interestingly, the ambiguous novel compounds that contained digraphs, such as seathorn, showed rejection latencies (661 milliseconds) that were only nine milliseconds faster than those for the orthographically unconstrained ambiguous novel compounds such as clamprod. This suggests that although the presence of digraphs affected the choice of competing parses, it did not constrain initial activation of alternative constituents. In a subsequent series of experiments Libben et al. (1999) explored these stimuli further in a primed recall task. It was found that seeing an ambiguous novel compound facilitated recall of semantic associates of all four of the possible constituent morphemes. Taking clamprod again as our example, the presentation of the string facilitated recall of sea (a semantic associate of clam), hold (a semantic associate of clamp), push (a semantic associate of prod), and stick (a semantic associate of rod). It seemed clear from these results that morphological parsing had automatically activated all possible constituents. These findings led us to conclude that "the lexicon is hungry".

2.2. The semantic transparency of compound constituents

What consequences does this view of a hungry lexicon have for the processing of existing compounds, particularly those that are semantically opaque? If the findings from novel word processing generalize to the processing of real words - if those words also undergo automatic and obligatory morphological processing - then semantically opaque Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 12 compounds should be more difficult to process than semantically transparent ones. This is because the morphological parser would activate constituent representations that are incongruent with whole word representations. We began by building on the work of Sandra (1990) who, in his investigation of the effects of semantic transparency in Dutch, had found that semantically transparent compounds facilitate recognition of semantic associates of their constituent morphemes. Thus, Sandra (1990) found that words such as suntan facilitated (or primed) the recognition of moon, which is a semantic associate of the initial compound constituent sun , but not of the whole word. Would the same effect occur for semantically opaque compounds such as Sunday? Sandra (1990) found that it did not, and concluded that in contrast to semantically transparent compounds, semantically opaque compounds do not undergo morphological parsing (also called prelexical morphological decomposition). This conclusion, while perfectly consistent with the data, generated the same paradox we had noted in our study of ambiguous novel compounds. How could the cognitive system know whether a compound was transparent or opaque unless it had processed it in the first place? We embarked upon a series of studies (Libben et al. 2003) in which we explored the relationship between whole words and constituents among four types of compounds, given in (2). (2) RELATIONSHIP BETWEEN WHOLE WORDS AND CONSTITUENTS IN COMPOUNDS

EXAMPLE RELATIONSHIP ABBREVIATION

car-wash transparent-transparent TT strawberry opaque-transparent OT jailbird transparent-opaque TO hogwash opaque-opaque OO Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 13 In one experiment, the relationship between whole words and constituents for such compound words was investigated through a constituent priming paradigm. Eighty-seven participants first saw a compound constituent displayed on a computer screen for 500 milliseconds and then were shown the entire compound. The dependent variable was the lexical decision response time for the target compound words. As show in Figure 1, the general pattern of results suggested that both opaque and transparent compound words receive roughly equivalent facilitation from prior presentation of their constituent morphemes. Prior presentation of wash facilitated hogwash in much the same way that wash facilitated carwash. On average, compounds with transparent heads were responded to more quickly than opaque heads, indicating the latter require extra processing. 600
625
650
675
700
725
750
775
800
825
850
875
900

UnprimedConstituent 1

prime

Constituent 2

prime

Response time (ms)

TT OT TO OO Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 14 Figure 1: Patterns of constituent priming for transparent-transparent (TT), opaque- transparent (OT), transparent-opaque (TO), and opaque-opaque (OO) compounds found by Libben et al. (2003) (Experiment 2). In Libben (1998), I proposed a model that might account for this apparent inconsistency by appealing to a distinction among levels. In this model, both opaque and transparent compounds undergo morphological parsing. They have identical lexical representations (shown in the middle level of Figure 2), but they differ with respect to how they are linked to the representations of their constituent morphemes at the conceptual level. In Figure 2, lines show associations between representations. TT, OT, and TO compounds are distinguished from each other in terms of how the conceptual level maps onto the lexical level. For example, with TT compounds, since both members of the compound are transparent, both the first and second term of the compound map from the conceptual level to the lexical level. This contrasts with OT and TO compounds, where only the transparent term is mapped from the conceptual level to the lexical level. The model also distinguishes between componential and non-componential compounds. The latter may have transparent associations between the meanings of the constituents in the compound and their meanings as free morphemes, but cannot be understood in terms of normal head-modifi er relations. The exocentric compound bighorn would be an example of this type, as it is a type of sheep rather than a type of horn. Square brackets show structured representations, and componential versus noncomponential compounds are distinguished in terms of whether there are structured or Copyedited by RMD/KP, CJL/RCL 2005-088, 13 April 2006 15 not at the conceptual level. For example, the TT-componential compound [blue][berry] has internal structure at the conceptual level, while the TT-non-componential compound bighorn does not. Figure 2: Semantic transparency and componentiality in Libben (1998). Many aspects of the model in Figure 2 appear to be on the right track. Yet, I would blue [blue][berry] berry blue - [blue][berry] - berry blueberry

TT-componential

straw [straw][berry] berry straw - straw][berry] - berry strawberry

OT-componential

shoe [shoe][horn] horn shoe - [shoe][horn] - horn shoehorn

TO-componential

big bighorn horn big - [big][horn] - hornquotesdbs_dbs35.pdfusesText_40

[PDF] jeux poétiques collège

[PDF] fiche d'identité d'une entreprise modèle

[PDF] poésie humour cycle 3

[PDF] resultat bac 2016 maroc

[PDF] cne maroc

[PDF] fiche d'identité entreprise vierge

[PDF] foie douloureux au toucher

[PDF] ostéopathie viscérale foie

[PDF] douleur projetée épaule

[PDF] ostéopathie viscérale estomac

[PDF] extrait sec total du lait

[PDF] comment calculer l'extrait sec du lait

[PDF] mouvement accéléré physique

[PDF] mouvement rectiligne ralenti

[PDF] mouvement ralenti