2.2.2 Resolution

Resolution is an instance of agreement of the verb with the conjunction phrase subject. It stands in contrast with the single-conjunct agreement in that it is inclusive and symmetrical as the gender values on both conjuncts are (not) simultaneous active goals for agreement. Two procedures present themselves as valid contenders for resolution – resolution rules and the default value insertion rule. These two mechanisms of resolution can co-exist in a single language (and in a single speaker’s grammar, see Figure 1) as illustrated in (6) with the two gender congruent conjuncts (feminine and feminine, neuter and neuter). The shared gender (feminine in (6a) and neuter in (6b)) inflectional morphology on the agreeing verbs indicates application of the resolution rules; while the alternative (masculine in (6a) and (6b)) inflectional morphology on the agreeing verbs indicates application of the default value insertion rule. In Croatian, as in other Slavic languages, masculine is a default gender in plural (with the probe already valued for the plural number feature).²

3.2.1 Goal Resolution accounts

The first two theoretical accounts of Resolution considered under Goal resolution approaches, postulate that the ConjP is valued for a full set of phi-features – the gender feature included. Accordingly, under these accounts, in subject-verb agreement, the ConjP acts as a single active goal for the agreeing probe.

Franks and Willer-Gold (2014) account of resolution argue that gender feature on ConjP is computed ConjP internally and feed by percolation to ConjP; RES rules proceed via unification – in the case of congruent features, and, via the DVI rule – in the case of feature conflict or in absence in the feature value on a conjunct (e.g., neuter singular conjuncts in example (8)). The account further proposes a structure where the conjuncts are ConjPs, namely, the NP is merged with its own ConjP projecting Conj-head – [ConjP2 [ConjP1 Conj1 NP1] [Conj’2 Conj2 NP2]]. Consequently, as resolution applies in parallel to structure building, the maximal projection is inevitably valued for gender feature. The valued gender feature on ConjP (or a single conjunct NP) contributes to the maximal projection targeted as a single goal by a single (non-split) probe. The example in (9) supports the single goal-single probe approach, demonstrating that agreement with two singular conjuncts with incongruent gender features is restricted to masculine gender agreement on the verbal probe (cf., Consistency Principle, Marušič et al. 2015). The contrast in grammaticality with the neuter and feminine plural agreement on the verb indicates that a probe cannot independently target gender feature of a (single) conjunct, and number feature on the ConjP (for detailed discussion on the precise derivation of the masculine value in (9) see Franks and Willer-Gold 2014).

Murphy and Puškar’s (2018) rule ordering account of conjunct agreement equally argues for the gender value(s) on the ConjP to be internally sourced. Resolved agreement is derived by the c-selecting Conj-head consecutively merging with its two conjuncts; followed by the Conj-head probing each conjunct to copy its gender feature. The two accumulated gender features project up to the maximal projection ConjP from which they are visible to a higher phi-agreeing probe. To account for resolution of two incongruent gender features copied on the probe, the account introduces a further assumption – a Conj-head prespecified for the default masculine gender. Once this head is (optionally) selected from the lexicon, the ConjP inherits three gender values later copied onto the probe. Notably, to derive a single gender value for the agreeing verb, resolution operates post-syntactically by deleting all but the least marked value of the probe’s gender features resulting in masculine agreement (for detailed discussion on the precise mechanics of constraint-driven impoverishment rules deriving masculine value see Murphy and Puškar 2018). Consequently, Murphy and Puškar’s distributed account of resolution posits that although the (default) masculine value can be already present on the goal, its resolved status is attributed after copy on the probe.

In contrast to the two presented accounts, the final account of resolution considered in this section argues that the Conj-head cannot compute its own gender value; and, as a consequence, ConjP cannot be inherently specified for gender feature. According to Marušič and colleagues (2015), in No-Peeking grammar, a probe first targets the ConjP goal to agree with its number feature and copy its plural value; and, only consequently targets the ConjP goal to agree with its gender feature. As no value for gender feature is found on ConjP by the probe, in order for the probe to successfully agree in gender with ConjP, the default masculine gender value is inserted on the ConjP ([–singular → +masc]), and copied onto the probe.

In sum, the three presented accounts of Resolution in South Slavic argue for a resolved/default masculine value to be present on the ConjP prior to copy, and, consequently, gender feature valuation on the probe. Note, though, with Murphy and Puškar’s (2018) account distributing the status of masculine across the goal and the probe, respectively; Franks and Willer-Gold (2014) and Marušič and colleagues (2015) accounts present a clear example of Goal resolution approaches. This distinction will be relevant for the precise formulation of the experimental hypothesis for the novel experimental study presented in Section 4.2.

3.2.2 Probe Resolution approach

The two accounts considered for Probe resolution approaches assume that the Conj-head cannot compute the value for the gender feature (only for the number feature), and, hence, no gender feature (value) is present on the ConjP. Crucially, the sole bearers of the (grammatical) gender feature, and, hence, the (inherently) active goals for the probing verb within the conjunction phrase, are the two nouns.

Citko’s (2018) account of resolved agreement argues for a gender probe to agree with the two conjuncts. This instance of Multiple Agree simultaneously copies the gender features of the two conjoined nouns onto the probe. The feature resolution of the two (conflicting) values operates on the probe in line with the language specific gender feature RES rules. Notably, with the two conjoined nouns being (inherently) active goals, no requirement is in place for operation of the DVI rules. Citko’s account then represents a clear example of the Probe resolution approach.

In his tentative account of resolution, Bošković (2009) assumes that a single (split-)probe can target two independent goals. Assuming that no value for gender feature is present on ConjP, the probe is forced to agree with two local goals – the ConjP to value its number feature and the higher conjunct to value its gender feature. This configuration, inevitably, leads to a lethal ambiguity as the (subject) ConjP is blocked from being successfully pied-piped to SpecTP. To resolve this conflict and avoid agreement failure, Bošković tentatively suggests that the valued gender on the probe is deleted and replaced by a (default) masculine value.

4.1.1 Production rate

The elicited production rates indicate that the masculine was produced on the agreeing verbal target in all nine conditions in preverbal order, see Table 3. These production rates varied across conditions due to interaction of two factors: (i) the additive contribution of two dominant agreement strategies, ConjP-agreement and single-closest conjunct agreement; and, (ii) the gender value of non-masculine conjunct(s), neuter and feminine.⁷ The production rate of masculine plural in the masculine congruent condition (the combination of the two masculine conjuncts, MM), was 100%, with ConjP-agreement and single-(closest) conjunct agreement contributing the same value, masculine, to this result. In non-masculine congruent conditions, the combination of two neuter or feminine conjuncts, masculine plural was produced at the lower rate compared to the shared gender, (NN = M:12%, FF = M:15%), (6), with the unique contribution of ConjP-agreement to the production of the masculine on the agreeing past participle. The production rate of masculine plural in the masculine incongruent conditions varied relative to the order and gender values of the two conjuncts. This interaction is observed, for example, in the combination of neuter and masculine conjuncts with a higher rate of masculine produced in the condition where masculine was the second (closest) conjunct (NM = M:92%, MN = M:55%); with additive contribution of ConjP-agreement and closest-conjunct agreement. In the non-masculine incongruent conditions, the combination of neuter and feminine conjuncts, where ConjP and single (closest-)conjunct agreement are argued to transparently compete. In these conditions, the production rate of masculine strongly competes with that of the closest conjunct particularly when the closest conjunct is neuter, with a lower production rates observed for masculine with the closest neuter conjunct (FN = M:38% or NF = M:52%), (5), (see section 2.2.2.1 and Willer-Gold et al. 2016 for neuter derived by resolution rules additionally contributing to this low % of masculine plural). In contrast, when the ConjP subject was placed post-verbally, there was a significant reduction in production rates of masculine (M = 2–5%) in conditions other than MM, MN, and MF (compared to single(closest-)conjunct agreement M/F/M = 90–95%), suggesting unavailability of ConjP-agreement with postverbal subjects (cf., Smith 2017). In sum, the elicited production data direct the research domain of Resolution to pre-verbal contexts.

4.1.2 Production time

The production time was the second measure collected in the elicited production study reported in Willer-Gold and colleagues (2018). Note that this measure was only calculated for the agreement with the preverbal subjects. The production time is a temporal measure, quantifying the time it takes the speaker to produce the prompted response. In this study, the production time was a measure of a time window recorded from the onset of a stimulus presentation to the end of the prompted sentence production when the participant pressed the button to continue to the next model sentence.

The results presented in Figure 2 show that the shortest production times were recorded for the masculine congruent condition (MM) and the longest for the non-masculine incongruent conditions (FN and NF). As argued by the authors, these production times are predicted by the uniformity of productions as summarised in Table 4 listing the individual exponents per condition (nine gender combinations) and per agreement strategy (Resolution, closest-conjunct agreement and distant conjunct agreement). In masculine congruent condition, ConjP agreement and single-conjunct agreement yield uniform feature values (masculine, masculine, masculine) matched in uniformity with their (syncretic) morphological exponents (-i, -i, -i).⁸ In stark contrast, in the non-masculine incongruent conditions, each agreement strategy yields a different gender value (masculine, neuter/feminine, neuter/feminine) mapped onto non-uniform morphological exponents (-i; -a, -e; -e, -a). The masculine incongruent conditions (NM and FM), where the two dominate strategies, ConjP-agreement and single closest-conjunct agreement, yield uniform feature values (masculine, masculine) mapped onto syncretic morphological exponents (-i, -i) result in comparatively faster reading times to the previous two conditions, (see also MN and MF); with comparable reading times measured for the non-masculine congruent conditions, matched in the degree of uniformity (NN and FF, (neuter/feminine, neuter/feminine) (-a, -e; -a, -e)). To summarise the findings, the results indicate the production times for the nine gender combinations in preverbal order strongly correlate with the uniformity of gender value transparently mapped onto the morphological exponent produced on the agreeing verb, with the equal degree of uniformity of productions computed via dominate strategies resulting in faster production time.

Figure 2

Elicited production study: Conjunct agreement. Production times in ms (BCS and Slovenian, Willer-Gold et al. 2018).

Contending uniformity of productions is estimated at the late stage of sentence production and is a measure of the processing cost of uniformity of the productions for the agreeing verb, it is valid to question if this production data should be considered as empirical evidence for the Probe or Goal Resolution accounts. Notably, in contrast to Goal Resolution, Probe Resolution account predicts late valuation of the gender feature on the verbal probe (only) after copying and Resolution of the (conflicting) gender feature values from the two nominal conjuncts.

However, a few words of caution are in order. Firstly, the presented times are a gross measure capturing production at the sentence level rather than the agreeing verb.

Secondly, it is important to dissociate the conflict in feature values indicated by low uniformity resulting from (multiple) alternative agreement procedures operating in parallel and yielding conflicting gender values for a single (or multiple) probe(s) (e.g., -i; -a,-e; -e,-a); from the conflict resulting from incongruent gender values on the two conjuncts (e.g., neuter and feminine) which once resolved contribute a single value for the probe. Arguably, with the latter subsumed under the former, only the former has an impact on the production times (see section 2.3 and 2.4) (note though that the same mechanism could be recruited to resolve the two conflicts, e.g., impoverishment rules).⁹¹⁰ Crucially, then, if uniformity is an estimation of conflict resolution over (multiple/in parallel) valued probe(s) – the (resolved) masculine value included, the production times provide at best only indirect evidence for the locus of Resolution to be on the Probe.

And, thirdly, with the production times capturing production at the sentence level, the observed effect could reflect speakers taking longer to read the second conjunct when conflicting in gender value with the first conjunct. In this case, longer production times could be an indication not of the variance of morphological exponents on the verb but rather of cumulative effects from reading the coordinated subject. This potential of competition being resolved at the earlier stage of sentential processing is explored in the novel self-paced reading study presented in the next section. Confirmation of Resolution taking place at an earlier stage would offer experimental evidence in support of the Goal Resolution accounts.

4.2.1 Predicting (grammatical) gender

Prediction refers to a cognitive process by which present contextual and linguistic information can be used to access new input in advance during rapidly evolving discourse (Kutas and Hillyard 1984; Kamide 2008). Psycholinguistic studies of word-level prediction have consistently shown that grammatical gender feature is a valid predictor of an upcoming target noun (see Sekerina 2012; Akhutina et al. 1999 for Russian; see Badecker and Kuminiak 2007 for Slovak; see Wicha, Moreno, and Kutas 2003; 2004 for Spanish; see van Berkum et al. 2005 for Dutch; Fuchs 2022 for heritage speakers). The early detection of a violation of this prediction is signalled by a significant increase in processing cost replicated across languages and methods suggesting that (grammatical) gender prediction arises routinely and effortlessly.

In their study of anticipating words in discourse, Van Berkum and colleagues (2005) have shown that discourse information not only allows readers to predict the semantic features of the upcoming target noun but to instantaneously activate predictions about its grammatical gender. This study used context to constrain the prediction about the upcoming target noun down to a single lexical item (the bookcase) and, crucially, its gender feature (common), seen in example (10). To validate the prediction, the experimental design manipulated the gender feature on the adjective preceding the target noun with prediction-consistent (neuter) and prediction-inconsistent (common) inflectional agreement. The strength of the gender prediction was reflected in the strength of the disruption in sentence processing by the unexpected gender value on the agreeing adjective (neuter). The small but reliable disruption to the evolving sentential interpretation was quickly picked up by the readers at the adjective, i.e., before the target noun was presented. This result was taken as evidence for the presence of grammatical gender prediction available to the speakers before they read the adjective. In their ERP experiment, this disruption was signalled by positive deflection in the ERP waveform and in the self-paced reading experiment by slowdown in reading times.

In the eye-tracking study on spoken word-recognition, Sekerina (2012) has shown that the grammatical gender inflectionally marked on the adjective acts as a predictor of the upcoming target noun. In this study, speakers were presented with a scene containing four pictures of concrete objects (mis)matched in colour and/or gender (masculine and feminine) with the target noun featured in the auditorily presented sentence, in (11). The study investigated the eye-movements to the target object focusing on two critical regions lined up with the auditory presentation of the adjective, the verb, and the noun. To investigate the earliness of this anticipatory effect, the experimental design manipulated the distance between the colour adjective and the noun, with the noun immediately following the adjective as in (11a) and being split from the adjective by an intervening verb as in (11b). The study captured an early effect of gender with the fixations to the target object observed ahead of the auditory presentation of the target. Additionally, at the critical verb region in the split-constituent condition, the listeners looked significantly more to the target object in the scene that contained the colour competitor mismatched compared to the one matched in gender. This was taken as evidence that the prediction is retained across an intervening verb.

4.2.2 Predicting resolved masculine gender

The current study builds on the research into (grammatical) gender prediction by extending the classic gender prediction paradigms to subject-verb agreement and in particular to the conjunction phrase in subject position. The study is concerned with the conjunction of two (inanimate) plural nouns with congruent and incongruent (grammatical) gender in the pre-verbal subject position and agreement in gender formally marked on the (upcoming) past-participle verb.¹¹ The study sets out to experimentally validate the assumption posited by the Goal Resolution approaches – the presence of valued resolved gender feature on ConjP_[Gend:masc], (1a) repeated below as (12).

In order to achieve this, we start by positing a simple linking assumption that the representation of gender feature value on maximal projection ConjP in subject position activates a single masculine gender, henceforth <masc>, used by the parser to predict the value of the gender feature of the upcoming agreeing verbal target, masculine, (13); this assumption does not hold for the Probe Resolution approach. It follows that when reading the two nouns conjoined by a formal cue of coordination, the conjunction i, the parser incrementally access the grammatical genders of the two nouns and, at that point, can either (i) derive a single representation of gender to predict the gender on the upcoming target verb, (14a); or, (ii) withhold derivation of such a representation until reading the verbal target, (14b).

To validate activation of <masc> by the representation of gender on ConjP, ConjP_{[Gend: masc]}, the postulated activation of <masc> should hold under two conditions.

Firstly, with <masc> activated by the [Gend: masc] on ConjP goal, in line with the Goal Resolution approaches, resolved masculine prediction is hypothesised to be activated early (even) ahead of reading the verbal target, rather than at the verbal target. This latter parse is predicted by Probe Resolution account with the resolved masculine prediction activated by the representation of the valued (grammatical) gender on the probe. From these two conditions, it follows that any disruption caused by encountering a <masc>-inconsistent gender (e.g., neuter) on the verbal target should be quickly picked up; and, its effects should be measured early – at the first encounter of the verbal target, (15a).

Secondly, in line with the observations that resolved masculine is an omnipresent agreement alternative, and, hence, present under any combination of (grammatical) genders on the conjoined nouns, <masc> should hold in the absence of formal cues of masculine gender (masculine nouns) in the subject domain, (15c) and (15d).

In line with the early <masc> hypothesis, at the first encounter of masculine agreeing verb consistent with the <masc> prediction, no increase in reading times is expected for any gender combination of conjuncts in (15), but crucially, not even for conjunction of two non-masculine nouns, (15c) and (15d).¹² If confirmed, this prediction would support the assumption that [Gend: masc] can be represented on ConjP in the total absence of formal cues of masculine gender in the subject domain. Furthermore, in contrast, an increase in reading time is predicted at the first encounter of agreeing verb that is inconsistent with <masc> (e.g., neuter). Confirmation of these two predictions for the paradigm in (15) would support the representational status of [Gend: masc] on Goal and, hence, provide experimental evidence in support of the Goal Resolution approaches.

In the paradigm presented in (15), <masc>-inconsistent neuter gender on the verbal target performs two functions. Firstly, it provides an ungrammatical contrast to, postulated, grammatical masculine agreement on the verbal target, in (15a). Secondly, it introduces a comparatively strong grammatical alternative – single-closest conjunct agreement strategy (preferred agreement alternatives in South Slavic, see section 4.1.1); which is inevitably expected to be processed in parallel (or even compete) with ConjP-agreement (see section 2, Figure 1), in (15b)–(15d). Presumably, with the closest-conjunct agreement parse, the grammatical gender of the second noun, supported by the formal cue on the neuter gender noun, receives an additional boost in activation, <neut>, predicting the presence of neuter gender on upcoming agreeing verbal target, (15b)–(15d).¹³ It follows that due to the parallel activation of the two parses – ConjP (<masc>) and closest-conjunct agreement (<neut>), no difference in reading time should be measured at the verbal target when the verb is morphologically inflected for masculine compared to neuter for the gender combinations with second conjunct neuter noun, (15b)–(15d); to the exclusion of conjunction of two masculine nouns as the (ConjP and) closest-conjunct agreement strategy only predicts masculine gender on the agreeing verbal target, (15a).

Notably, in testing for the representation of ConjP_{[Gend: masc]}, the issue of multiple parses due to availability of multiple agreement strategies in the language presents a challenge to any attempt to constrain the paradigm to a standard 2*2 factorial design with a single (un)grammatical agreement strategy. One strategy to achieve this aim would be to block or suppress the parse of closest-conjunct agreement by priming ConjP-agreement; potentially a challenge in itself with the high level of intra- and inter-individual variation in speakers of South Slavic languages. Second strategy would be to exclude the alternative closest-conjunct agreement parse from the paradigm. This could be achieved in two ways. First, by reducing the paradigm in (15) to a smaller set of gender combinations excluding feminine/masculine and neuter combinations, (15b) and (15c); and, hence, freeing feminine gender to induce the ungrammatical contrast on the verbal target for the remaining two gender combinations, (15a) and (15d). This paradigm would allow for direct contrast between <masc> congruent (grammatical) – masculine verbal target and <masc> incongruent (ungrammatical) – feminine verbal target, with the latter causing disruption in processing measured by the increase in reading times. However, although the alternative agreement would not be directly encoded in the paradigm, the paradigm does not exclude activation of additional <neut> prediction in (15d).¹⁴ And, second, to attempt to fully exclude the possibility of activating additional <neut> prediction. This latter option is being pursued below.

To control for the alternative single-closest-conjunct agreement parse, we posit a second paradigm in (16) which capitalises on the predictions of Consistency Principle which states that single-conjunct agreement strategy is only available for conjuncts matched in plural number with the probe (Marušič et al. 2015). Paradigm in (16) iterates gender combinations from paradigm in (15), with the conjoined nouns being in singular number in order to block closest-conjunct agreement parse in (16b)–(16d). Consistency Principle, consequently, predicts a disruption measured by the increase in reading times at the encounter of target verb morphologically inflected for neuter gender and plural number – not only for the masculine congruent combination of the two nouns, (16a), but, equally, for the combinations of the singular nouns where the second noun is in neuter gender, and in particular where both nouns are in neuter gender, (16b)–(16d) (see section 2.2.2.3). In contrast, assuming that the activation of <masc> is triggered by ConjP subject, and as such is uniformly available for any given combination of gender and number conjoint nouns, verbal target morphologically inflected for masculine plural is expected to be processed without any additional cost, (16).¹⁵ The reading time signatures for the singular conjuncts conditions, (16), are, hence, predicted to parallel those for the combination of plural masculine conjoined nouns, (15a).

The two number-congruent conjuncts (mis)matched in number feature with the target verb are directly compared in (17). Moreover, if the representation of ConjP_{[Gend: masc]} is uniformly available, no difference in reading times is predicted for the singular and plural number congruent conjuncts at the masculine plural verbal target. Alternatively, assuming that the mismatch in formal cues on the conjuncts and the verb induces processing cost, the two singular conjuncts mismatched in number feature with the masculine plural verb should induce longer reading times at the verbal target compared to two conjuncts matched in number values, (17) (see also (16) and (15), respectively).

Finally, the elicited production study on conjunct agreement in South Slavic indicated that the production rate of masculine agreement with the gender combinations featuring in the paradigm in (15) and (16) incrementally decreases from (a) to (d) (masculine in MM:100%> MN:55%>FN:36%>NN:12%); and, that the two sub-sets of (in)congruent (NN, MM/MN, FN) gender combinations cluster together in their production times, (a) and (d) and (b) and (c) (Willer-Gold et al. 2016; 2018). Assuming processing symmetry in production and comprehension of agreement dependencies (cf., Villata and Franck 2019; Slioussar and Malko 2016), we could expect to observe variation in facilitation at the masculine agreement target (in terms of timing or size of the effect, see discussion in section 4.2.4), relative to the gender combination of the conjuncts with, for example, NN combination either eliciting longer or shorter reading times at the verb, respectively, compared to the other three conditions.

To conclude the discussion on the (grammatical) resolved masculine gender prediction, the activation of <masc> at the early stage of sentence processing, in the ConjP subject domain, is predicted by the Goal Resolution accounts. To test that the (early) effect of gender persists at the (linear) distance, the auxiliary, su, and the adverb, uredno, intervened between the ConjP and the agreeing verbal target, (18).

Conversely, the Probe Resolution account predicts that the verb as a locus of Resolution is the locus of <masc>. Therefore, the violation of <masc> is expected to be observed not at the verb but at the next agreeing target (if not before). To accommodate this expectation, the study capitalises on the double past-participle in the verb domain: the be-participle and the lexical-participle. This solution allows to track the timing of Resolution, with Goal Resolution accounts predicting the ungrammaticality effect to be observed on the be-participle, bila, and the Probe Resolution account for it to be measured only later on the lexical-participle, počiščena, (18).

In sum, the paradigms in (15)–(17) verify the postulated representation of ConjP_{[Gend: masc]} activating <masc>-prediction after reading the two nouns in the subject conjunction phrase. Paradigm in (15) included an agreement alternative to Resolution with disruption in processing expected with <masc> inconsistent agreement on the target verb (neuter) only for conjunction of two masculine nouns, as well as no disruption in processing of <masc> consistent agreement on the target verb with conjunction of non-masculine nouns. Paradigm in (16) established a clear bidirectional grammaticality contrast with disruption in processing expected only for <masc> inconsistent agreement on the target verb (neuter). Finally, paradigm in (17) contrasts singular and plural congruent conjuncts and predicts to observe no increase in processing of the <masc> consistent verbal target with the singular congruent conjuncts. In the experimental study detailed in the following sections, these three paradigms were split across three self-paced reading experiments as summarised in Table 5.

4.2.4 General discussion

In the self-paced reading experimental study, two results stand out: (1) the early, robust and sustained detection of the prediction-inconsistent agreement (N-verb) observed at the first encounter of the verbal target (the be-participle) (Experiment 2); this result stands in contrast to (2) the late but robust detection of the prediction-inconsistent agreement (N-verb) observed at the second encounter of the verbal target (the lexical-participle) for masculine congruent condition (Experiment 1). Therefore, the first result is considered as baseline against which to interpret the second result.

This first result confirms that the parser is responsive to prediction violations with the significantly longer reading time measured for the prediction-inconsistent than prediction-consistent verbal target. This contrast in reading time signatures maps directly onto the ungrammatical and grammatical status of the eight experimental conditions, respectively; validating the methodology used in this study and the manipulation controlled for by the design. More to the point, this first result is predicted by Goal Resolution accounts arguing that the gender feature on the maximal projection ConjP is valued by the resolved masculine gender (irrespective of the gender combination of the two (singular) conjoined nouns); and, as such ConjP presents an active goal for copying of the feature values to the agreeing probe. In comparison, in Experiment 1, in the masculine incongruent conditions, no difference in reading times was observed for <masc>-consistent and -inconsistent agreement on the verbal target. The absence of disruption points to the grammatical status of the two alternative agreement strategies – the ConjP agreement-Resolution (<masc>) and the single-closest conjunct agreement (<neut>, suggesting the two strategies are (equally) available to the parser (with no indication of competition between the two strategies) (see also Experiment 3 results Section 4.2.3.).¹⁶

The second result is of special interest as it indicates the peculiar behaviour of the MM plural condition – the delay in capturing the disruption induced by <masc> prediction inconsistent agreement on the verbal target. Namely, in the masculine congruent condition, the violation expected to be observed at the first encounter of the verbal target, the be-participle, is delayed till the second encounter, the lexical-participle. This inconsistency of the result across the two participles is not directly predicted by the Goal Resolution accounts (see discussion in section 4.2.4.1); but falls in line with the predictions of the Probe Resolution account arguing that Resolution operates on the verbal probe after the probe has copied the gender features from the two conjuncts. In accordance with the Probe Resolution account, occurrence of the violation is expected to be observed at the second encounter of the verbal target with the <masc>-prediction activated later, when reading the second past-participle. However, this pattern of delayed effect of <masc> (whether disruptive or facilitatory) does not generalised to other conditions where the activation of <masc>-prediction is expected to be activated in the same manner, i.e., no such delay is observed for the ungrammatical conditions, in Experiment 2; and, more to the point, no delay is observed for the masculine incongruent conditions (FN and NN) crossed with M-verb, in Experiment 1. Finally, the (cumulative) facilitation of reading M-verb at the second encounter is predicted by both accounts.

Crucially, neither Goal nor Probe Resolution accounts can offer a comprehensive and straightforward explanation as to why this late effect is uniquely observed in the ungrammatical condition in the Experiment 1 experimental paradigm. The next section presents three findings from the experimental studies on agreement that, when consolidated, offer a formulation of a processing proposal that captures the late gender effects under the Goal Resolution approaches.