Subjective perception of affixation: A test case from Spanish

Highlights

• We examine Spanish verbs with clitic pronouns.

• White noise was overlaid on either the clitic or the stem.

• Participants rated the subjective loudness of the noise, on a 1–5 scale.

• Ratings differed for proclitics vs. enclitics, and for procliticized vs. encliticized stems.

• Morphological structure may act as a cognitive variable affecting perception.

Abstract

Cross-linguistically, prefixes and suffixes differ in both frequency and in phonological behavior. These differences could plausibly have their source in listeners’ subjective perceptual experiences of prefixes and suffixes, an idea that we pursued using a noise-rating task in Spanish. Participants heard minimally-different Spanish words such as me patea ‘s/he kicks me’ versus patéame ‘kick me’, where the clitic pronoun me behaves phonologically like a prefix versus a suffix, and rated the loudness of white noise overlaid on either the pronoun or the verb stem. Results demonstrated that participants assigned significantly different ratings to noise occurring on prefixes versus suffixes, and on prefixed versus suffixed stems, even when the signal-to-noise ratio remained constant across conditions. That is, listeners’ subjective perceptual experience of the noise differed according to what morpheme type the noise occurred on, suggesting that morphological structure can act as a cognitive variable affecting perceptual clarity.

Introduction

Prefixes and suffixes act as morphological equivalents in many ways. Both types of affix can be derivational or inflectional, both can encode similar syntactic and semantic representations, and both come from closed classes that typically contain only a subset of a language's phonological inventory. Nevertheless, the literature has noted many differences between these two types of morphemes. One difference lies in frequency: overall, prefixing morphology is far less frequent than suffixing morphology, and is overwhelmingly restricted to languages with verb-object ordering (Dryer, 2013, Hawkins and Cutler, 1988, Hawkins and Gilligan, 1988). Another difference lies in phonological behavior: cross-linguistically, prefixes tend to participate in fewer alternations than suffixes do (Hyman, 2008), even though morphological boundaries are generally active sites for such alternations.

Given the similarities between prefixes and suffixes, their differences seem rather puzzling, but previous research has offered some provocative explanations from both functional and formal perspectives. For example, Hawkins and Cutler (1988) examine the typological distribution of prefixing and suffixing morphology, and offer an explanation grounded in speech processing (see also Colé et al., 1989, Cutler et al., 1985, and more recently Himmelmann, 2014). They cite evidence that the initial portions of a word drive the process of recognition (Grosjean, 1980, Marslen-Wilson, 1984, Marslen-Wilson, 1987, Marslen-Wilson and Welsh, 1978, Marslen-Wilson and Zwitserlood, 1989, Nooteboom, 1981, and many others), and also observe that listeners prefer to interpret the semantic information encoded in roots before they interpret the syntactic information typically encoded in affixes (as the authors point out, this observation holds mostly for inflectional affixes, but they argue that a similar observation should apply to derivational affixes). According to their logic, then, suffixing morphology should provide a perceptual advantage because it temporally aligns the important information of the root with the perceptually-prominent position of the word onset. Prefixing morphology, on the other hand, should suffer from a disadvantage because the root and the word onset are temporally mis-aligned. Hawkins and Cutler (1988) put forth these proposed perceptual differences to account for the different frequencies of prefixing versus suffixing languages.

Previous research has also tackled differences in phonological behavior. Whereas suffixes generally participate in both regressive and progressive alternations, prefixes do not. Specifically, although prefixes often undergo regressive alternations triggered by roots, they rarely trigger progressive alternations on the following root (Hyman, 2008). To illustrate by way of simple voicing examples from English, we see cases where suffixes trigger regressive assimilation (lea[f], lea[v]-es), where suffixes undergo progressive assimilation (cat-[s], dog-[z]), and where prefixes undergo regressive assimilation (tran[s]-sexual, tran[z]-national), but we do not see cases where prefixes trigger progressive assimilation (*off-[p]eat for off-beat, *sub-[b]ar for sub-par). The nasal assimilation process in Luganda is an example of this otherwise rare occurrence: m-báànj-a → m-máànj-a ‘I demand payment’, Hyman and Katamba (1999:397). The “Alternation Asymmetry”, as we refer to it, holds for a heterogeneous set of processes, including local assimilations, long-distance assimilations such as consonant harmony (Hansson, 2001) and vowel harmony (Walker, 2011), and vowel elision (Casali, 1997), suggesting the need for a very general explanation.

Research in theoretical phonology, particularly in Optimality Theory, has used the concept of positional faithfulness to approach this problem. The idea is that certain positions within a word possess a privileged status, and faithfulness constraints preserve the underlying identity of segments in those positions. When ranked above the relevant markedness constraints, then, positional faithfulness constraints prevent alternations from occurring in certain positions, even if the alternation remains active elsewhere. McCarthy and Prince (1995) initially proposed a positional faithfulness constraint that preferentially preserves segments in roots but not affixes. Beckman (1997) and Casali (1997) took a step further and proposed constraints that preserve segments specifically in the initial portions of roots. These analyses can competently model the Alternation Asymmetry because they preferentially protect just those root segments which are closest to the prefix (namely, the initial segments most likely to undergo any putative progressive alternations triggered by the prefix). For example, the constraint Ident-σ₁(voice) would state that segments in root-initial syllables should have identical voicing values in the input and the output, thereby mitigating specifically against changes such as off-beat → off-[p]eat. Furthermore, because faithfulness constraints can prevent any type of surface alternation, including local and long distance assimilations as well as other changes, this theoretical solution seems to be a satisfyingly general one.

In many ways, these previous proposals – one functional, one formal – represent significant progress in our understanding of the differences between prefixes and suffixes. In other ways, however, these proposals simply do not fit with the existing evidence about how listeners perceive spoken words. For example, Hawkins and Cutler (1988) conclude that suffixing morphology should offer a perceptual advantage over prefixing morphology, but they do not test this idea explicitly. In fact, the few studies that do examine this issue suggest the opposite pattern. In priming experiments, prefixed forms prime their stems (insincere primes sincere) as well as related prefixed forms (unfasten primes refasten). Suffixed forms also prime their stems (punishment primes punish), but they do not prime related suffixed forms (confession does not prime confessor) (Marslen-Wilson et al., 1994). Feldman and Larabee (2001) demonstrated a similar prefixation advantage across several modalities and inter-stimulus intervals. Not only does prefixing morphology appear to offer a processing advantage over suffixing morphology, it also appears to offer an advantage over bare roots. Schriefers et al. (1991) asked listeners to perform gating and phoneme monitoring tasks while listening to spoken Dutch words containing early versus late uniqueness points. To their surprise, however, results showed no effect of uniqueness point. Instead, in gating tasks, listeners needed significantly more sensory information to identify bare roots versus prefixed words, regardless of early versus late uniqueness points. Similarly, in phoneme monitoring tasks, listeners responded more slowly in the bare root condition compared to the prefixed condition, again regardless of uniqueness point. Interestingly, even though their study did not explicitly test the question of affixed words versus bare roots, Marslen-Wilson et al. (1994) report a compatible result: “[p]refixed pairs prime each other as well as, if not better than, pairs made up of a free stem and a prefixed form” (1994:27, with specific reference to their Experiment 4). Taken together, these experimental findings suggest a prefixation advantage that is at odds with Hawkins and Cutler's (1988) processing proposal.

The OT concept of positional faithfulness also raises problems when we attempt to translate it into perceptual terms. Essentially, the theory claims that if segments occupy root-initial positions, those segments should not alternate. The implication – sometimes made explicit, as in Beckman (1997) – is that alternations somehow interfere with accurate segmental perception, and such interference cannot be tolerated in a position that is so important for word recognition. But it is not clear whether this implication really holds. The initial portions of a word drive the process of recognition (works cited above), and therefore possess an important status. But the perceptual consequence of alternations in this position remains an open question. Previous studies examining alternations in word-medial or final positions have found that they do not interfere with word recognition (Coenen et al., 2001, Gaskell and Marslen-Wilson, 1996, Snoeren et al., 2008). In fact, in some circumstances, alternations may even facilitate recognition (Gow, 2001, Gow and Im, 2004). If similar findings hold for alternations in word-initial position, then the perceptual basis for positional faithfulness constraints would appear to be rather tenuous, a point that Becker et al. (2012) also make in their investigation of alternations on monosyllabic versus polysyllabic words.

These problems suggest that a very basic issue remains unresolved: we simply do not know how listeners perceive prefixes versus suffixes. Most of the literature on processing of complex words has focused on a different problem, namely whether listeners perceive prefixes and suffixes at all. There is good evidence that they do. Results from lexical decision experiments have repeatedly shown that morphologically related words – whether prefixed or suffixed – prime one another, strongly suggesting that people decompose spoken words into their constituent roots and affixes (Feldman and Larabee, 2001, Marslen-Wilson et al., 1994, Taft and Forster, 1975, and many others, although see Tyler et al., 1988). Evidence for decomposition exists for both derived and inflected words in English (Fowler et al., 1985, although see Stanners et al., 1979). For derived words, the process is modulated by individual lexical characteristics, such as productivity (Marslen-Wilson et al., 1994) and relative frequency of base versus derived forms (Caramazza et al., 1988, Hay, 2001, Hay, 2003, Hay and Baayen, 2002, Laudanna and Burani, 2013).

Researchers focusing on Spanish report largely similar findings (for an overview, see Domínguez et al., 2000), although it is important to note that most of these studies examine printed, rather than auditory, word recognition. Sánchez-Casas et al. (2003) compared inflectionally related words such as niño - niña (‘boy-girl’), derivationally related words such as rama - ramo (‘branch-bunch’), and non-related but orthographically overlapping words such as foco - foca (‘floodlight-seal’). Results from visually masked priming with short stimulus-onset asynchronies showed that both inflectionally and derivationally related words produce a significant facilitation effect, of an equivalent size, but that this effect was not observed with forms that were merely similar orthographically. Álvarez et al. (2011) used similar stimuli in an event-related potential (ERP) investigation, and showed that inflectionally and derivationally-related words triggered events with a similar time course, but different locations. Domínguez et al. (2006) also used ERP to demonstrate that participants process Spanish words which share a prefix (as in reacción-REFORMA ‘reaction-reform’) differently than those which share a pseudo-prefix (regalo-REFORMA ‘gift-reform’); this result fits with those reported for primed lexical decision of prefixed and pseudo-prefixed words (Domínguez et al., 2010). In a similar vein, Allen and Badecker (2002), Domínguez et al. (2002), and Beyersmann et al. (2013) all report results suggesting that participants decompose morphologically complex Spanish words during processing. As in English, the relative frequency of Spanish base versus Spanish derived forms affects lexical decision times (López-Villaseñor, 2012).

Given that listeners can perceive prefixes and suffixes as distinct from the roots they attach to, we turn in the current study to the question of how they perceive them. Our research goal is to characterize the subjective experience that people have when listening to a prefix or a suffix – put somewhat crudely, do they experience relatively good perception when listening to a particular affix type, or do they experience relatively poor perception? A motivation for this focus on subjective perceptual experience comes from the work of Ohala (1993), Blevins (2004), and others, who argue that diachronic sound changes – and the resulting synchronic phonological alternations – originate in the mind of listeners as they attempt to interpret the variability inherent in speech. For example, when listeners hear a partially nasalized vowel in a VN sequence, they may attribute the nasalization to the adjacent nasal consonant. Alternatively, if they fail to attend to the consonant, they may misinterpret this contextual variation and attribute nasalization to the vowel itself. Such a scenario could give rise to phonemically nasal vowels; in such cases, the nasal consonant typically disappears (cf. French brun [brœ̃] ‘brown’, Ohala, 1993:243). Certain sequences of vowels and consonants encourage misperception more than others, giving rise to alternation types that recur across languages (Blevins, 2004). In the current research, we extend this idea by hypothesizing that certain sequences of morphemes could also encourage misperception more than others, with the aim of taking a step forward in explaining the differences between prefixes and suffixes.

To investigate this hypothesis, we conducted an experiment using a noise-rating task with spoken Spanish stimuli. Native Spanish-speaking participants heard verbal phrases such as me patea [me paˈtea] ‘s/he kicks me’ or patéame [paˈtea me] ‘kick me’, where strikethrough indicates the presence of white noise, and assigned a rating indicating how loud they thought the noise sounded, on a scale from 1 to 5. Comparing the examples, both of which contain the morphemes me ‘1Sg.Obj’and patea ‘kick’, we see that Spanish personal object pronouns change positions depending upon the status of the verb: such pronouns occur before regularly conjugated verbs, but after imperatives, infinitives, and gerunds. Importantly, the pronoun has the same segmental content and meaning in either position, and it can attach in both instances to a segmentally and prosodically identical verb stem. By comparing listeners’ responses to stimuli like me patea versus patéame, then, we can factor out extraneous differences among morphemes and focus more clearly on the morphological status of the affixes themselves. More specifically, given that the objective noise levels in me patea versus patéame are equivalent, we ask whether the subjective noise levels are also equivalent. Do listeners perceive the noise differently depending upon the position of the pronoun? To evaluate the same question with regards to stems, we also presented stimuli in which noise had been overlaid on the verb, e.g. me patea versus patéame.¹

According to some analyses of Romance languages, the Spanish pronouns are technically clitics, not affixes, because they can represent syntactic constituents (e.g. Kayne, 1975, Rizzi, 1986). Thus, procliticized examples like me patea ‘s/he kicks me’ and encliticized examples like patéame ‘kick me’ constitute complete sentences in which me represents the direct object argument to the verb. From a syntactic perspective, then, the Spanish pronoun clitics differ from inflectional affixes, which can agree with other constituents in the sentence but cannot independently represent them (although some researchers have argued that the Spanish clitics do not, in fact, differ from affixes at all; see Franco, 2013). From a phonological perspective, however, the Spanish pronoun clitics are straightforwardly affixes. They do not occur as independent words, do not take primary stress, and do not take phrasal accent even under conditions of contrastive focus (Hualde, 2005:258–9). They do not normally take secondary stress, either, although the enclitics can do so optionally (Hualde, 2012:162). Thus, while me patea and patéame are complete sentences, they are nevertheless single phonological words (for more comprehensive discussion of clitics, see Anderson, 2005). The phonologically-dependent status of Spanish pronouns renders them comparable to prefixes and suffixes and therefore suitable candidates for pursuing our research question.

In addition to minimal pairs for clitic position, Spanish has two more characteristics that make it a good test language for probing the general differences between prefixes and suffixes. First, more than one Spanish pronoun can attach to the same verb root: se me pisa [se me ˈpisa] ‘I am stepped on’, písamelo [ˈpisa me lo] ‘step on it for me’. In these examples, the pronoun me is procliticized or encliticized to the verb stem, but it does not occupy the absolute initial or final position in the word. Crucially, this allows us to treat word-edge proximity and morphological status as separate factors. Thus, if listeners’ responses to stimuli like me patea versus se me pisa (analogously, patéame versus písamelo) are equivalent, then word-edge proximity would appear to be irrelevant, and we could focus our attention only on potential differences between proclitics and enclitics; conversely, if responses to such stimuli differ, then word-edge proximity would appear to play a role.

Second, Spanish provides an interesting test of our general hypotheses for another reason: most varieties exhibit spirantization of voiced stops. The details differ from one variety to another, but the basic pattern is one in which underlying /b, d, g/ alternate to [β, ð, ɣ] after a vowel (Hualde, 2005:138–9). Note that this alternation is structure-changing, because [β, ð, ɣ] do not occur in the underlying inventory. Spirantization is relatively insensitive to positional restrictions, and may occur at the beginning or end of syllables, at the beginning or end of words, and in stressed or unstressed positions. As a consequence, Spanish proclitics can trigger alternations on the initial edge of verb roots, in precisely the position where they are cross-linguistically rare (and in precisely the position where positional faithfulness constraints would mitigate against them): batéame [baˈtea me] ‘hit me (with a bat)’ versus me batea [me βaˈtea] ‘s/he hits me (with a bat)’. We can therefore use Spanish to investigate whether alternations in this position affect listeners’ perceptual experiences. If listeners respond differently to stimuli such as me [p]atea, [p]atéame on the one hand versus me [β]atea, [b]atéame on the other, we could potentially begin to offer perceptually-based explanations for the origins of the Alternation Asymmetry.

Previous authors have noted that many Spanish prefixes exhibit special behavior with regard to syllabification, in a manner potentially relevant to the current experiment (Face, 2002, Hualde, 2005, see also Harris, 1983). In general, within word boundaries, syllabification creates consonant clusters, as long as such clusters are permissible word onsets: for example, broma [bro.ma] ‘joke’, abro [a.bro] ‘I open’ (Hualde, 2005:74). Across word boundaries, however, syllabification does not create such clusters: club romano ‘Roman club’ [kluβ.romano] (Hualde, 2005:88). In this regard, certain prefixes behave as independent words, resisting the formation of clusters in a similar manner: subregión ‘subregion’ /sub.rexion/ (Hualde, 2005:95, following his transcription in phonemic brackets). This pattern suggests that Spanish prefixes constitute a separate phonological domain whose morpheme and syllable edges are aligned (Face, 2002), and furthermore that prefixes occur outside the prosodic word of the following stem (Selkirk, 1996, see also Himmelmann, 2014). If so, prefixed stems without suffixes exhibit perfect alignment with prosodic word boundaries (while suffixed stems do not, since the prosodic word extends to the suffix), potentially enhancing their perception. Such an argument could conceivably apply to procliticized stems without enclitics, of the kind used here.

The current experiment uses a noise-rating task, whose logic is relatively straightforward. Noise interferes with the recognition of the word, and loudness ratings probe the extent to which listeners actually experience this interference. Previous studies have used this task to explore the effect of prior exposure. Jacoby et al. (1988) presented listeners with old and new sentences against a background of white noise at varying intensities, and found that listeners assigned lower ratings to noise on the sentences they had heard previously, even when they were not required to identify those sentences. Goldinger et al. (1999) used the same technique with individual words and found that listeners assigned lower ratings to noise on words they had heard previously, even when they failed to correctly recognize those words as old. In both studies, then, listeners misattributed the relative ease with which they could interpret the old spoken stimuli to a difference in noise level – a perceptual illusion. In the current study, we continue with the assumption that listeners will misattribute relative perceptual ease to differences in noise level, and ask whether listeners experience more (or less) ease for proclitics versus enclitics, and procliticized stems versus encliticized stems.

Although the noise-rating task used in the current experiment differs from the techniques more commonly used to investigate the perception of complex words, such as primed lexical decision and gating, its features are well-suited to the research question at hand. To begin with, as Jacoby et al. (1988) point out, the task does not require participants to report or reflect on the words they have heard, making it unlikely that they will use linguistic knowledge to strategically alter their judgments. This is important for the current experiment because we are investigating a process that listeners are not consciously aware of – specifically, whether they have different perceptual experiences of different morpheme types – but in order to do so, we are obliged to use stimuli that differ in ways which listeners are consciously aware of (most people could probably report, or reflect upon, the fact that the pronoun me occurs in different positions in me patea versus patéame).

Even more importantly, the outcome variable of the noise-rating task represents a subjective judgment on the part of the participant, which distinguishes it from primed lexical decision or gating tasks, where the outcome variables are accuracy and reaction time. As Jacoby et al. (1988) point out, the subjective experience underlying equally accurate or equally fast results may differ. As an example, consider Marslen-Wilson et al.’s finding (1994:27) that prefix–suffix and suffix–prefix pairs facilitate each other equally well in a cross-modal priming task. Specifically, spoken English primes like dis-trust facilitate lexical decision for printed English words like trust-ful by an average of 30 milliseconds, while spoken primes like judg-ment facilitate lexical decision for printed words like mis-judge by an average of 31 milliseconds. What these nearly-equal RT results do not reveal is whether listeners had different subjective experiences of, for example, listening to the prefixed primes dis-trust versus suffixed primes like judg-ment. It is entirely possible that they did; if so, such differences either failed to exert an effect on reaction times, or exerted effects that were masked by additional differences in reaction times later on in the task, such as the time necessary to read the printed targets. Either way, the subjective experience of the participants as they encountered these words – a worthy topic of investigation in its own right – remains largely hidden in the RT data.

In the current experiment, then, the noise-rating task provided an advantage over other methodologies because it allowed us to collect the types of judgments that were most closely related to the question at hand, namely, do listeners experience spoken procliticized forms differently than encliticized forms? In using the noise-rating task to pursue this question, we are extending it in several ways. From a simple methodological perspective, we are extending it to a smaller linguistic unit, and to a different language. Thus, while Jacoby et al. (1988) presented listeners with noisy English sentences and Goldinger et al. (1999) presented noisy English words, the current study presents noisy Spanish morphemes. Extending the task in this way highlights its versatility: we can use noise to “isolate” one morpheme while keeping the whole word intact, thus addressing a significant methodological challenge in the study of spoken complex words. As a consequence, though, the current experiment uses sequences of noisy and clear speech within the same stimulus, a departure from previous work that should be borne in mind when interpreting the results.

We are also extending the noise-rating task to a new predictor variable. Whereas previous work examined the effects of prior exposure (old versus new), the current experiment examines the effects of morphological status (proclitic versus enclitic, and procliticized versus encliticized). As Jacoby et al. (1988) point out, prior exposure is just one of many cognitive variables that affect our subjective perception of a stimulus. If we read the lyrics of a rock song before we listen to it, those lyrics seem clearer. If we talk with a friend at a party, their voice seems louder than that of other people nearby. In these instances, the true difference lies in attention or prior experience, but the subjective difference is one of increased clarity. We can characterize the current experiment, then, as a test of the hypothesis that morphological structure acts as a “cognitive variable” that influences our subjective perception of spoken words.

In light of the broad considerations outlined above, we formulated several hypotheses about this test case with spoken Spanish verbs. First, for the clitics themselves, we hypothesized that listeners would experience relatively poor perception of proclitics compared to enclitics. If poor perception can lead to diachronic loss, then poor perception of proclitics/prefixes would help explain the greater frequency of encliticizing/suffixing languages cross-linguistically. To be consistent with this hypothesis, our results should demonstrate higher noise loudness ratings for stimuli such as me patea compared to patéame. Second, for the verbs stems that clitics attach to, the hypothesis is somewhat more tentative. Previous experimental results demonstrate an overall processing advantage for prefixed forms (e.g., Schriefers et al., 1991), although it is not clear to what extent the results from phoneme-monitoring and gating studies bear on the subjective judgments from a noise-rating task. If they do have some bearing, then these findings would suggest that listeners should experience relative clarity during the perception of procliticized stems compared to encliticized stems, an idea that also finds some support in the distinct pattern of Spanish prefix syllabification, discussed in Section 1.2. Results consistent with this hypothesis would show lower noise loudness ratings for stimuli such as me patea compared to patéame. Third and finally, we hypothesize that listeners’ perceptual experiences of verb phrases will not differ when stem-initial segments alternate, a somewhat counter-intuitive notion that nevertheless fits with previous experimental results demonstrating potentially beneficial effects of alternations (e.g., Gow, 2001) and with the general insensitivity of Spanish spirantization to prominent positions. Results consistent with this hypothesis would show equivalent loudness ratings for stimuli such as me [p]atea, [p]atéame versus me [β]atea, [b]atéame although we must bear in mind that null results do not definitively rule out the presence of an effect.