Vorstellung, lemma & lexicalisation: A cognitive analytic introduction to representational language
This short essay seeks to discuss current psycholinguistic perspectives of and for the underlying processes concerned with the conceptualization/semantic representation of ‘meaning’ in spoken language form. Here the dynamic processes for the conceptualization/semantic representation of 'meaning' (in spoken language) are held as being freighted by the term lexicalization. In concert with the conceptualization/semantic representation of meaning [lexicalization] reference will be made to evidence admitting from constraint or inhibition of the former processes seen through the lens of the object naming paradigm. In seeking to satisfy the stated purpose (attending to inhibition in the object naming paradigm) close attention has been given to the following papers: Damian et al., 2001; Kroll & Stewart, 1994; and, Wheeldon & Monsell, 1994.

Evidence from speech errors

The crucial factor in the problematic reduction of these psycholinguistic processes into constituent steps is by analysing process failures, or put more simply what Dell (1986: p. 284) calls attending to “the way in which the system breaks down.” The type and frequency of spoken errors have long since been thought to provide valuable insights into the inchoate processes of fluent speech. The collection of evidence in this field thus relies heavily upon faulty speech; furthermore, for over a century faults in speech processing have been harvested to supply grist for the psycholinguistic mill (e.g. Dell 1986; Freud 1891b; 1900a; 1901b; 1950a; Fromkin 1971; 1973; Stemberger 1982). Therefore errors of speech found in the laboratory inform the basis from which falsifiable hypotheses may account for dynamic speech production (Dell 1986; Dell et al. 1999; Garrett 1976; 1982).

Typical speech errors can include: Spoonerisms = e.g. a transposition of two initial wetters of lords – here exampled by letters /w/ and /l/> (see Fromkin 1971; 1973; Garrett 1976; MacKay 1970); Anticipation errors = e.g. a word is spoken sequentially earlier than it been to have been - here exampled by ; Morpheme-exchange errors = e.g. where words two are sequentially transposed – here exampled by and (see Smyth et al. 1987). Notwithstanding the situation regarding the practical use of speech errors in approaching fluent speech production, Stemberger (1984) further distinguished speech errors from lexical errors. After Barry (2008) typical lexical errors can include: Substitution/Malapropism errors = displacing one word for another which may be similar (see Hotopf 1980; Fromkin 1980); loss errors = a word is omitted; addition error = an extra word or words is produced (Fromkin 1980); Blend errors = two words are co-activated in coincidence creating a ‘chimeric’ form (see Fromkin 1980; Garrett 1980).

The group of speech errors correctly termed parapraxis, are more commonly known as ‘Freudian slips’. Freud writes of his ‘discovery’ of these anomalies of speech production in 1901 [Ger. Fehlleistung; Eng. ‘faulty function’, also rendered into English as ‘compromise-formations’]. The scope of the faulty function captured by the German term Fehlleistung extends to also ensnare the so-named ‘tip-of-the-tongue’ (TOT) phenomenon [Ger. Vergessen] (see Laplanche and Pontalis 1973). That said, modern laboratory research into ‘TOT’ phenomena (e.g. Brown and McNeill 1966; Schriefers et al. 1990; Vigliocco et al. 1997) can be said to have come to espouse two general hypotheses relevant to the processes of lexicalization via the exploration of the role of inhibitory processes of constraint. The first theory posits a partial activation of a word’s phonological form. The second quite interesting theory posits that competition arises from a semantic similarity effectively constraining the target signifier (Damian et al. 2001; Kroll and Stewart 1994). The suggestion of ‘negative’ priming effect not dissimilar to those demonstrated by the Stroop effect in visual processing may be of not inconsiderable significance to research in the field. Furthermore, interest in the precise lexical nature of processes occurring at the semantic stage (Levelt et al. 1999) or at the semantic and syntactic stages (Dell et al. 1997; Vigliocco et al. 1997) may have some bearing on approaches to psycholinguistics in general.

Lexicalization in semantic representation

For the sire of pragmatics, Peirce (c. 1907), his sociocognitive schema for the production of meaning, semiosis (CP 1.176-179, 6.7-34 in Houser et al. 1998), is construed as a composite cycle of three iterative temporal relations: representamen; object; and, interpretant. And we shall return to this important notion of a psychical-linguistic iterative process presently. That being so, for Harley (2001: p. 359) the term lexicalization can be defined as “the process in speech production whereby we turn the thoughts underlying words into sounds: we translate a semantic representation (the meaning) of a content word into its phonological representation or form (its sound).” Some researchers (Levelt 1989; Levalt et al. 1999) hold that lexicalization occurs when a conceptualization is freighted by a ‘lemma’ and thenceforward translated into a word-form in terms expressed through recourse to morphemic and phonemic encoding and articulation stages. Whilst for others, (e.g. Caramazza 1997), there is simply no evidence for an intermediate [lemma] stage; thus, meaning appears to be intrinsic to words. Such a materialist view seems to imply that words themselves are implicitly meaningful rather than concepts, and, if true, words themselves might represent concretized objects impervious to the drift of diachronic influence. But what then is the lemma? Closer inspection of the term lemma reveals two common psycholinguistic uses: (a) After Crystal (1987), as deriving from morphological and/or lexicographical studies where the base form of the verb is held as different to the word stem for a set of forms. (e.g. Walks, walked and walking are forms of the same lexeme, with walk as the lemma or word-stem form.) (b) Following Eysenck and Keane (2005: p. 560), as deriving from psycholinguistic study where an “abstract word form possessing syntactic features but not phonological ones” has been selected for utterance in the inchoate processes of speech production. The abstract lexical representation is thought to be indexed from the mental lexicon, whereupon the conceptualized abstract form is held to inaugurate the processes of speech production prior to any further encoding or articulation becoming attached to the said mental concept (also see Crystal 1987; Evans and Green 2006).

For Garrett’s (1982) psycholinguistic model for speech production there are five discrete, serial levels of representation [‘r’] processing deemed necessary and sufficient, these are: message-level r; functional-level r; positional-level r; phonetic-level r; and, articulatory-level r. According to Garrett, the speaker is assumed to spend time in a latent syntactic planning [stage 3. positional level] state prior to speaking. This assumption has been tested against the kind, frequency, latency and duration of speech errors (Boomer 1965; Smith 2000). In spite of this, it can be said that Garrett’s (1982) model appears to be restricted to the prediction of errors rather than finding a wider applicability within the realm of error-free speech production. That said, for Harley (2001), the trenchant view that speech production requires the necessity of syntactic preparation in the form of pauses between utterances can be said to be far from clear as yet. Whilst for Ferreirra and Swets (2002: p. 77) the matter at hand – syntactic preparation - is not simply one of for or against but rather the degree or “balance” struck between semantic preparation (“planning”) versus the context of response (“initiating speech”). Notwithstanding these laboratory-driven deliberations, one might well ask whether there exists demonstrable clinical research to the contrary. One notable clinical study of 1,381 children aged between 3 and 16yrs took place. Here researchers Swann and Mittler (1976) looked closely at language abilities from a sample of ‘ESNs’ (eductionally sub-normals – of course, today we would not use such terminology). More than 40% of the 16yr group were unable to use grammatical constructions, and 17.5% had not reached single-word stage (Swann and Mittler 1976; Mittler 2004). Other groups across the sample population presented with clear physical (e.g. deafness) and/or acquired language dysfunctions (e.g. CVA ‘stroke’ aphasia). Never the less, the predominant finding across the sample population was that there was no clearly demonstrable reason for language delay in the bulk of cases.

Dell (1986), Dell and O’Seaghdha (1991) and Dell et al.’s (1997a) proposal for a psycholinguistic model reduces the processes of speech production to four temporal relations. Moreover, for some (Dell et al. 1997; Vigliocco et al. 1997) insights gained from parallel distributed processing (see Rumelhart and Mclellend 1986) steers their thoughts toward a nodal architecture allowing for an iterative, cascade of parallel activations known as the spreading-activation theory: semantic; syntactic; morphological; and, phonological. Here speech errors were successfully predicted (Dell 1986) to be discrete; that is, limited to each categorical architectural layer of processing (e.g. morphological). Furthermore, the prediction of a simple rule-based process set within a rule-based hierarchy is suggestive that such a nodal architecture ought also to constrain competition between activations. In other words, a layer-specific inhibitory function or error-checking function which some (Mclellend and Rumelhaer 1981 in Rumelhart and Mclelland 1986) have termed ‘lateral inhibition’. These lateral inhibitions are not dissimilar to those ‘sentinel’ functions employed within modern information processing network infrastructures to inhibit or guard against parallel cascades of device activations (e.g. SMTP data packets) which can result in an affective flooding or ‘broadcast storm’ across the whole of the data layer of the OSI model. That analogy having been noted, it is interesting that such a ‘flooding’ of speech errors are predicted to occur within this model when a ‘wrong’ word has been ‘tricked’ into increased activation over the ‘right’ word. Wheeldon & Monsell (1994: p. 333) discern a ‘competitive priming’ effect, or co-activation of lemma representation candidates, which predicts a retardation of ‘right’ selection in picture naming. Wheeldon & Monsell’s (1994) experiments yielded unambiguous evidence for a robust competitive priming effect. Yet for some researchers, notably Glaser (1992) and Roelofs (2000), Dell’s (1986) prediction for large numbers of errors was not the borne out by the evidence. Both researchers (Glaser and Roelofs) separately highlight picture naming tasks accompanied by semantically related distractors designed to inhibit ‘right’ lexical selection (e.g. [picture of DOG] and ) as findings show only a modest increase in error rate against much higher predicted (Dell 1986) error results. The suggestion might be made that the findings from Glaser (1992) and Roelofs (2000) are compatible with the hypothesis of an inhibition of activation at the level of semantic representation which, in turn, would appear to logically indicate the existence of a cognitive process of conceptualization [lemma]. Cutting & Ferreira (1999) found that ‘sympathetic’ distractor words related to the homographic homophone (e.g. after Barry (2008), [picture of BALL] and ) were interpretated as leading to a faster priming of the lemma representation and a subsequent activation of the ‘right’ word. These findings can be suggested to also be compatible with a competitive activation theory.

A psycholinguistic/computational model, initially proposed by Levelt (1989) and then augmented through further research by Levelt et al. (1999a), known as WEAVER++, simplifies the number of speech production processes to three serial temporal relations utilizing spread-activation: conceptualization; formulation; and, encoding. Put simply, the process architectures proposed by Garrett (1976; 1982) and Levelt et al. (1999) differ from Dell et al. (1986; 1991; 1997a; Vigliocco et al. 1997) notably in their valuation of in two crucial respects: discrete serial activation processes over parallel cascaded activation processes; and, the explicit consideration of a mediating abstract lexico-semantic representation [lemma].

Reflections in conclusion

Wheeldon & Monsell’s (1994) findings (F1(2, 108) = 5.5, p < 0.05, F2(2, 192) = 3.7, p < 0.05) of slower responses in alternate word definition and picture naming tasks; where the target was a ‘competitor’, are significant in that they favour a robust inhibitory competitive priming effect. Here the data suggests the inclusion of an inhibition at the stage between conceptual/semantic and morpho-phonological [lexeme] processes. That is to say, evidence for the presence of a parallel inhibitory function appearing not incompatible with those theories supporting a two stage (Levelt 1989; Levelt et al 1999) [lemma] processing model. Kroll and Stewart’s (1994) findings (q(4, 142) = 6.36, p < 0.01) also report significant inhibitions in the form of categorical interference within bilingual translation tasks resulting in slower times in categorically similar lists of words than those found within randomized lists. Kroll and Stewart (1994) argue that their findings are in support of an asymmetrical relation between the bi-directional translation function indicating a conceptual reliance on a favoured lexical activation process. Which is to say, the data is compatible with the suggestion that translation from L1 to L2 (where L = language) can be said to be conceptually mediated, whereas translation from L2 to L1 can be said to be lexically mediated. If indeed these finding can be taken as a whole, so to speak, then Kroll and Stewart’s (1994) notion for an asymmetrical relation between conceptual and lexical representation is quite interesting. Here an asymmetry can be said to support a conceptual semantic stage [lemma] model. For Damian et al. (2000) findings show that semantic similarity slows the response times for same-category picture and word naming tasks. Here semantic interference – i.e. lexical competition – can be said to be notionally responsible for the resulting data.In other words, Damian et al. (2000: B85) claim that the primary outcome of their experimentation with German language picture and word naming tasks turns upon a competition said to exist between concepts for words which, can be said to form the basis of an understanding “at the root of this effect.”