<<

THE TOLERANCE PRINCIPLE: A CLOSER LOOK AT THE DATIVE SHIFT

A Thesis submitted to the faculty of A* San Francisco State University % In partial fulfillment of the requirements for rs/cn the Degree

Master of Arts

In

Psychology: Mind, Brain, and Behavior

by

Divya Subramanian

San Francisco, California

August 2019 Copyright by Divya Subramanian 2019 CERTIFICATION OF APPROVAL

I certify that I have read The Tolerance Principle: A Closer Look at the Dative Shift by

Divya Subramanian, and that in my opinion this work meets the criteria for approving a thesis submitted in partial fulfillment of the requirement for the degree Master of Arts in

Psychology: Mind, Brain, and Behavior at San Francisco State University. THE TOLERANCE PRINCIPLE: A CLOSER LOOK AT THE DATIVE SHIFT

Divya Subramanian San Francisco, California 2019

Some dative appear in both the double construction (Give me the toy) and the prepositional construction (Give the toy to me), while others can only be used in one construction (Say something to me // *Say me something). Children cannot expect to hear every grammatical possibility, nor can they assume that dative verbs they have not previously heard in a construction can never be used in that construction. This results in a paradox: How do children learn what constructions are ungrammatical in the absence of negative evidence? Yang (2016) proposed a possible explanation to this paradox using the Tolerance Principle, a model that predicts linguistic productivity. In this thesis, we replicate Yang’s analysis of how children learn the dative shift using positive evidence alone. We then conduct a modified analysis to address some of the potential limitations in Yang’s study.

I certify that the Abstract is a correct representation of the content of this thesis.

Date TABLE OF CONTENTS

List of Tables...... vii

List of Figures...... viii

List of Appendices...... ix

1. Introduction...... 1

1.1 Rules and Exceptions...... 1

1.2 Productivity...... 2

1.3 The Past Tense Debate...... 4

2. The Tolerance Principle...... 5

2.1 Evidence in Favor of the Tolerance Principle...... 8

2.2 The English Dative Shift...... 14

2.3 Discussion...... 27

3. Analysis of the Dative Shift: Replication of Yang (2016)...... 28

3.1 M aterials...... 29

3.2 CHILDES Corpus Extraction...... 30

3.3 Analysis 1: Establishing Caused Rules...... 32

3.4 Analysis 2: Developmental Stages of a Caused Possession Rule...... 35

3.5 Analysis 3: Narrow-Range Rules...... 37

3.6 Discussion...... 41

v TABLE OF CONTENTS (cont.)

4. Analysis of the Dative Shift with Modifications...... 45

4.1 M aterials...... 45

4.2 Analysis 1: Establishing Caused Possession Rules...... 46

4.3 Analysis 2: Developmental Stages of a Caused Possession Rule...... 51

4.4 Analysis 3: Narrow-Range Rules...... 54

4.5 Discussion...... 58

References...... 62

Appendices...... 65

vii LIST OF TABLES

Table Page

1. Table 1...... 7 2. Table 2 ...... 10 3. Table 3 ...... 23 4. Table 4 ...... 37 5. Table 5 ...... 53 LIST OF FIGURES

Figures Page

1. Figure 1 ....3 LIST OF APPENDICES

Appendix Page

1. Appendix A ...... 65 2. Appendix B ...... 68 3. Appendix C ...... 69 4. Appendix D ...... 70 1

1. Introduction

1.1 Rules and Exceptions Second language learners are likely familiar with the struggle of painstakingly memorizing a grammatical rule only to be later presented with a list of exceptions. As it stands, nearly all grammatical rules have exceptions. The English language, which will be the of this paper, is rife with irregularities. For example, the regular English past-tense ending is the -ed suffix (1), but there are also many exceptions to this rule, such as sang, ran, or ate (2).

(1) I walk the dog. —► I walked the dog. (2) I eat sandwiches. —> I ate sandwiches.

Typically, second language learners are explicitly taught the exceptions to a rule. But in the case of first language acquisition, children are not always formally instructed on which forms are rule-following and which are exceptions. Children may begin using the past tense forms of verbs as early as 24 months of age. Around this time, adult feedback is inconsistent. Some adults may correct errors, whereas others may repeat the error back to the child in ‘baby talk’. Furthermore, there is evidence that correction, even when provided, makes little difference in language development (Marcus, 1993). Then, how do children learn what is or is not grammatical? A conservative approach might suggest that children simply refrain from uttering a phrase they haven’t previously heard before. Under this assumption, a child who has never heard the past tense form purchased would never use it in a sentence. Child speech data would quickly disprove this, as it is apparent that children do not limit their speech to what they have previously heard. Instead, children extract information about what is grammatically correct from what they have heard and extend it to new phrases. Therefore, a child could 2

extract information from other past tense forms (e.g. hugged, laughed, and talked) and extend the pattern to produce the form purchased.

1.2 Productivity This is due to one of the key principles of human language, productivity. In general terms, productivity means that what people produce in speech extends far beyond what they have previously heard. In relation to grammatical rules, productivity refers to the generalizability of a rule. A rule is said to be productive if it is extended to new or unfamiliar cases. In a famous experiment known as the Wug Test, children were shown to already be using language productively (Berko, 1958). As part of this test, children were introduced to several made up stems, and then asked to produce the past tense form (3). Most preschoolers and first-graders were able to produce the regular past tense form for half of the verbs shown to them. Although their proficiency was not on the same level as adult participants, this study demonstrated that even for young children, the English past tense is typically denoted by the -ed suffix. This suggests that the regular past tense ending is productive, and children as young as preschool-aged can grasp it.

(3) Past tense. Man swinging an object. “This is a man who knows how to rick /rik/. He is ricking. He did the same thing yesterday. What did he do yesterday? Yesterday he ."

However, as was previously mentioned, not all English verbs follow the regular past tense ending. And often, when children begin extending the regular past tense ending, they incorrectly extend it to irregular verbs, producing errors of overregularization such as she eated a sandwich. Overregularization errors have become the focus of many research studies, largely for their nonlinear developmental pattern. One would expect past tense acquisition to have a consistently upwards trajectory, in which children initially make frequent errors, but gradually improve to a 3

native proficiency. Instead, in many cases children display what has been called a U- shaped learning curve. The English past tense is a notable example of this phenomenon. In early stages of language acquisition, children correctly produce past tense forms of irregular verbs, (e.g. eat —* ate). This is followed by a middle stage in which children suddenly begin to overregularize by incorrectly applying the regular past tense ‘-ed’ ending to irregular verbs, leading to errors such as eated. Finally, children once again correctly produce irregular past tense forms.

There is some as to why U-shaped learning occurs. Verbs with higher frequencies will typically be learned before verbs with lower frequencies. As a majority of the most frequent English verbs have irregular past tense forms, they are acquired early on. But as the vocabulary size increases, verbs with regular past tense endings 4

eventually outnumber verbs with irregular past tense endings. This influx of regular verbs leads to the dip in the U-shaped curve, in which children begin overregularizing the irregular verbs they previously said correctly. However, this pattern of behavior has introduced new questions on how language is processed in the brain.

1.3 The Past Tense Debate In the past, there were two main perspectives to understanding this aspect of language acquisition. The distinction between them can be simplified down to whether irregular verbs are processed differently than regular verbs. Under computationalism, regular forms are generated by rules and irregular forms are committed to associative memory. The computationalist approach believed in the use of symbolized rules that are used by language production mechanisms but remain inaccessible to the speaker. Once a rule is encoded, all irregular verbs are accessed first. If a verb stem does not apply to any of the irregular past tense forms, the regular past tense form is generated by the -ed past tense rule. In , the connectionist perspective proposed a learning process of association without reliance on explicit rules. All verbs—regular and irregular—are stored in associative memory and accessed based on their frequency. As proponents of connectionism, Rumelhart and McClelland (1986) sought to simulate the acquisition of the English past tense using a Parallel Distributed Processing (PDP) model. Their model, commonly referred to as the RM model, followed a pattern association approach, using phonetic features to represent the stem (input) and past-tense forms (output) of verbs. By simplifying past-tense formation down to units of sound, the RM model did not need to rely on explicit representations of words or rules. The RM model was able to produce learning patterns similar to those of children; for example, a U-shaped pattern of development characterized by the overregularization of the -ed suffix. However, Pinker and Prince (1988), proponents of the rule-based computationalist approach, critiqued several features of the RM Model. Namely, a model 5

that does not acknowledge words or rules cannot explain the different past tenses of homophones, verbs that are phonetically identical but generate different past-tense forms {ring —> rang, wring —> wrung). And some errors, such as mail —► membled, are unlike any a human would make. The past tense debate did not conclude with either perspective being declared the victor. There is, however, a third perspective that has sparked new interest in recent years. In this third perspective, it is possible for both regular and irregular past tense forms to be generated by rules. Minority rules, such as sing —> sang, ring —> rang, can be productive alongside the majority -ed past tense ending. Children are predicted to maximize efficiency, thereby refining old rules or seeking out new rules when necessary. The storage of irregular items in associative memory must be a last resort. Then, the question becomes how do we know if a rule is productive?

2. The Tolerance Principle

In 2016, Charles Yang proposed the Tolerance Principle as a new solution to understanding language acquisition and processing. The Tolerance Principle (2) is a learning model that quantifies the exact tipping point at which a rule becomes productive. If we imagine a rule R applying to N items, of which some, M, are rule-following, and others, e, are exceptions, we call think of this as:

(1) R: N = M + e

According to the Tolerance Principle, a rule is said to be productive if the number of exceptions, e, does not exceed a threshold, On . This threshold is quantified as the total number of items, N, divided by the natural log of N. When a rule is productive, all exceptions are listed according to frequency and accessed before rule-following items. If a rule is not productive, then all N items are accessed based on their frequency. 6

(2) The Tolerance Principle If R is a productive rule applicable to N candidates, then the following relation holds between N and e, the number of exceptions that could but do not follow/?: N e

The formula for the Tolerance Principle is derived from the expected time it would take to access a rule R with e exceptions opposed to the expected time it would take to list all N items by frequency. This assumes any sample of words will follow Zipfs law (Zipf, 1949), in that the frequency of any word is inversely proportional to its rank. If it is more time efficient to access a rule R, then R is said to be productive. Of course, this requires the number of exceptions, e, to be relatively small. Under R, all exceptions are processed before rule-following items. As e grows, higher frequency rule- following items must wait for all lower frequency exceptions to be processed, and it becomes increasingly less efficient to process all exceptions before rule-following items. At this point, it becomes more efficient to process all N items by frequency, and R is no longer productive. Therefore, the Tolerance Principle can be seen akin to an economic model of language learning. By listing out the threshold for productivity at different vocabulary sizes (Table 1), some key observations can be made. For the sake of familiarity, we’ll apply these values to the English past tense in the following example. Imagine a child who knows 10 past tense forms. For the -ed rule to be productive, no more than 4 of these past tense forms can be irregular. This means that at a vocabulary size of 10, a rule can be productive even when 40 percent of the items are exceptions. Now imagine a child who knows 50 past tense forms. In this case, for the -ed rule to be productive, no more than 13 of these past tense forms, or 26 percent, can be exceptions. This means that as a 7

vocabulary size increases, the percentage of permissible exceptions decreases. According to Yang (2016), this could mean that children, who have smaller vocabularies, may be advantaged in favor of finding productive rules.

N On %

10 4 40.0

20 7 35.0

50 26.0 13 100 23 23.0

200 38 19.0

500 80 16.0

1,000 145 14.5

5,000 587 11.7 Table 1. The threshold for productivity across vocabulary sizes.

This isn’t to say that all rules are productive. If the number of exceptions exceeds the critical threshold, 6 n , then a rule cannot be productive. What happens then? Do children have no choice but to commit all N items to associative memory? Yang (2016) views language learning as a continuous search to maximize productivity. If the number of exceptions to a rule is not sufficiently low to make that rule productive, the Tolerance

Principle is applied recursively to a revised rule (/?’) to obtain a new set of items (N 1 and e'). This allows the productivity of nested rules, which can be thought of as conditionals. The recursive nature of the Tolerance Principle comes into play in later examples, such as the German nouns and the English dative shift. 8

2.1 Evidence in Favor of the Tolerance Principle The simple formula of the Tolerance Principle allows it to be easily tested in relation to any grammatical rule. This method, outlined by Yang (2016), proceeds as follows:

a. Obtain a rule R along with its structural description and structural change. b. Count N, the number of lexical items that meet the structural description of R. c. Count e, the subset of N that are exceptions to R.

d. Compare e and the critical threshold On = N/lnN to determine productivity.

In his book, Yang (2016) provided evidence for the Tolerance Principle across several languages. In the next few sections, we will outline some of his key examples: 1) The English past tense, the center of the debate between connectionist and computationalist perspectives, 2) An exploration of the German plural nouns, an example of the recursive application of the Tolerance Principle, 3) A detailed summary of the English dative shift, which will be the main focus of this paper.

2.1.1 The English Past Tense The English past tense, as we previously mentioned, was the center focus of what became known as the past tense debate. In this case, a productive rule refers to the generalizability of the regular past-tense ending, the -ed suffix. To test whether the regular past-tense ending is productive, Yang (2016) extracted all past-tense verb forms of child-directed North American English from the CHILDES corpus (MacWhinney, 2000). This resulted in 1,022 unique verbs, of which 127 were irregular verbs and therefore exceptions to the rule. The threshold was calculated to be 147. As the number of exceptions (e = 127) fell under the threshold (6*1022 = 147), the Tolerance Principle correctly predicted the productivity of the regular past tense ending. From samples of child speech, we already know that the regular past tense ending is productive even for children. The Tolerance Principle provides quantitative proof that this is the case. 9

(3) The Tolerance Principle states:

e < On where 6 n = N I InN 127 < 0io22 where $1022 = 1022 / lnl022 =147

However, Yang (2016) found that the regular past tense form was not productive at all points of frequency. For example, in the top 100 most frequent verbs the “fly ^ fle w ” rule demonstrated productivity, and between the top 200 to 500 verbs the “sing-^sang' rule demonstrated productivity. Table 2 depicts the productive state of various past-tense rules at different word frequencies. In each cell, N refers to the total number of items that fit the criteria for minority rules, and e refers to the items within N that do not follow a given rule. Productive rules are shaded grey. As an example, in the top 200 past tense verbs, the sing-^sang is productive. Under this rule, present tense verbs with the -ing ending take -ang as their past tense ending. There are a total N = 3 verbs that could apply to this rule: bring, ring, and sing. Only bring does not take the - ang past tense ending, and therefore e = 1. This may explain why the erroneous past- tense form brang is so common. Overall, the productivity of these narrow-range past tense rules represents the first stage of the U-shaped learning curve, where children initially produce correct past tense forms for irregular verbs. When we observe the developmental pattern of the regular past tense ending, we see that it does not become productive until the top 800 most frequent verbs. By this time, all narrow-range past-tense rules are dropped. Under this rule, the default past tense ending is the -ed suffix. The productivity in -ed represents the dip in the U-shaped curve, when children overextend the regular past tense ending to irregular verbs. Essentially, the productivity of different past tense rules corresponds with developmental stages of child speech, providing evidence for the Tolerance Principle as a model of language development and processing. 10

Top iV sing—*sang feed—*fed fly—>flew -d 6n & O 0 li II

100 — — (N= 8,e = 3) " 22 o' II 200 (iV = 3, e = 1) — II (N = 200, e = 76) 37

300 (Ar = 3,e=l) — (.N = 13, e = 8) (N = 300, e = 92) 52 & cT 1! 500 II (N=6,e = 3) (N= 15, e = 10) (N = 500, e = 103) 80 & oo" & 00 II II II II fl II II 1! " 00 O O 800 «> K> 119 II II

1022 o© (N= 13, e = 9) (N = 22,e=l6) (N = 1022, e = 127) 147 Table 2. The productivity of different past tense endings across word frequencies.

2.1.2 German Noun In English, the default plural ending is -s. Aside from a few exceptions, such as children, mice, or sheep, -s is easily the most common. In contrast, German does not have one plural suffix that stands out as the majority. Five suffixes are possible: -e, -(e)n,

-er, -0 (with the possible addition of an umlaut), and -s. Of these five suffixes, even the most frequent, -e(n), is used for less than half of all German nouns (Clahsen, Rothweiler, Woest, & Marcus, 1992). The least frequent suffix, -s, is only used for approximately 4 percent of German nouns. Despite this, it is typically used as the default plural ending for novel or loan words (das Musical —*• die Musicals). All this points to -s being a productive plural suffix. However, under the Tolerance Principle, a rule that is not in the majority (and in this case, the minority) should not be able capable of productivity. Recall the thresholds for productivity from Table 1. At a vocabulary size of just 10 items, a rule can only be productive if less than 40 percent of those items are exceptions, and this percentage will only decrease as vocabulary size increases. Compared against all possible suffixes, there appears to be no chance for -s to be productive—that is, if we assume all other suffixes are exceptions to the default -s rule. If we consider the other suffixes productive rules applying to different 11

subsets of nouns, it is still possible for -s to be productive under the Tolerance Principle despite being a numerical minority. Here, Yang (2016) demonstrates the recursive nature of the Tolerance Principle, which will again come into play in his analysis of the English dative shift. The recursion results in nested rules, which can be thought of as a series of conditionals (4). Instead of ascribing a single suffix rule to all German nouns, a nested rule would suggest that a noun with Property A and B would receive suffix X, a noun with only Property A would receive suffix Y, and the remaining nouns receive suffix Z. This would make R3 the most general or default rule. Of course, this recursive process does not limit itself to 3 rules nor 2 properties. As Yang (2016), suggests, children are expected to maximize productivity, and this recursive process will continue until they are left with the fewest possible exceptions.

(4) Ri: If [+A, +B] THEN X R2: If [+A] THEN Y R3: Z

To begin explaining the productivity of the German plural suffixes, Yang (2016) would have to identify different properties of these nouns that could fall within productive rules. One of the key properties of German nouns is gender, which can be feminine, masculine, or neuter. This property is picked up very early by children, with error rates dropping below 10 percent by age 3, children (Szagun, Stumper, Sondag, & Franik, 2007). Therefore, it is intuitive to begin by separating nouns into gender-based subclasses. Yang (2016) extracted the 500 most frequent plural nouns from the child-directed, German-language section of the CHILDES corpus (MacWhinney, 2000). After eliminating annotation errors and mass nouns, this resulted in a total of 458 nouns (166 feminine, 200 masculine, and 92 neuter). 12

Within the feminine nouns, Yang (2016) found that the -(e)n suffix was productive. There were a total of 166 feminine nouns, of which 146 took the ~(e)n suffix. The threshold for productivity was calculated to be 32. As the number of exceptions (e = 166 - 146 = 20) fell under the threshold, the -(e)n rule was predicted to be the default for feminine nouns.

(5) The Tolerance Principle states:

e<0N where 0n~N / InN

20 < 0i66 where 6 \ee =166 / lnl66 = 32

As a productive rule was found for feminine [+FEM] nouns, the Tolerance Principle must recursively apply a productive rule for the remaining [-FEM] plural nouns. Within the masculine and neuter nouns, Yang (2016) found a productive phonological rule. Nouns that ended in a schwa followed by -llrln tend to take the null suffix 0 . For example, das Zimmer (the room) becomes die Zimmer (the rooms). Of the

83 masculine or neuter nouns that ended in a schwa, only 6 did not take the null suffix 0 as their plural form. These 6 exceptions fall significantly under the threshold of 18.

(6) The Tolerance Principle states:

e < On where On - N / InN 6 < 083 where 083 = 83 / ln83 = 18

Similarly, nouns that end in a final schwa nearly always take the suffix -n regardless of gender. Of the 11 masculine/neuter nouns that end in a schwa, all took the plural suffix -n. 13

(7) The Tolerance Principle states:

e< On where Bn = N / InN 0 < Bn where Bn = 11 / lnl 1 = 4

For the remaining 129 masculine nouns, -e was predicted to be the productive plural suffix. Only 22 were exceptions to this rule, which fell under the threshold 26.

(8) The Tolerance Principle states:

e < B x where Bn ~ N / \t\N

22 < 0i29 where Bn9 = 129 / lnl29 = 26

Within the remaining neuter nouns, monosyllabic neuter nouns that contain back vowels /a, o, and u/ typically take the ending -er. Out of a total of 22, only 6 are exceptions.

(9) The Tolerance Principle states:

c

6 < On where 822 = 22 / ln22 = 7

Finally, Yang (2016) was left with 46 neuter nouns. Although a productive rule could not be found with this final set, it was clear that loanwords seemed to take the suffix -s, latinate nouns with the -um ending took the plural suffix -en, leaving -e to be the productive suffix for the remaining neuter nouns. In essence, the productivity of the 5 plural suffixes proceeds in this order:

(10) a. If [+FEM] then [-(e)n] die Blume (the flower) —> die Blumen (the flowers) 14

b. If [schwa + -1/r/n] then [0] das Zimmer (the room) —► die Zimmer (the rooms) c. If [schwa] then [-n] der Drache (the dragon) —► die Drachen (the dragons) d. If [+NEUT] with monosyllabic back vowels then [-er + umlaut] das Haus (the house) —► die Hauser (the houses) e. -e der Hund (the dog) —► die Hunde (the dogs) f. -s das Handy (the cellphone) —► die Handys (the cellphones)

As seen in (10), rules applying to the German plural noun suffixes are processed in order from most specific to most general. A feminine noun will not be considered for rules beyond (10a), and excluding certain phonetic rules, -e is the default suffix for masculine and neuter nouns. This also explains how -s, the least frequent of all the plural suffixes, can be a default plural noun ending. Unlike the rules preceding it, -s is not associated with any specific gender, and is therefore the most general. As such, it becomes the default suffix when either a) a noun does not apply to any of the preceding rules, or b) a child is unsure of whether a noun applies to any of the preceding rules.

2.2 The English Dative Shift Among the most noteworthy examples Yang provided as evidence for the Tolerance Principle is his analysis of the dative shift. The dative shift refers to an alternation between two constructions in which the sentence structure changes, but the meaning remains approximately the same. Dative verbs, or verbs that participate in either or both constructions, typically possess meanings related to transferring objects or ideas to an individual or entity. While some dative verbs can alternate (be used in both constructions), there are others that can only be used in a double object construction 15

(DOC) or prepositional construction (PPC). As an example, (11) depicts an alternating dative verb, whereas (12) depicts a verb that can only be used in the DOC, and (13) a verb that can only be used in the PPC.

( 11) a. He sent me a package. b. He sent a package to me. (12) a. She asked me a question. b. *She asked a question to me. (13) a. *She said me something funny. b. She said something funny to me.

Children learning the dative shift will observe a pattern in which verbs that appear in one construction are sometimes extended to the other. For example, a child who observes the verb lend in the DOC (I’ll lend you my coat) may feel confident using it in the PPC (Lend your coat to me). However, children will not always know which verbs can or cannot shift. This occasionally leads to errors such as, ‘She said me something’. This bears similarities to previous examples, such as the English past tense. In both these cases, children observe some regularity in how a verb is used but must navigate the exceptions. However, there is one stark difference between the English dative shift and English past tense. In the case of the English past tense, a child who produces an error such as ‘eated’ will eventually hear the correct form ‘ate ’ and can rule out ‘eated’ as ungrammatical. But in the case of the dative shift, a child who has observed a verb in one construction does not have sufficient evidence for or against the other construction. After all, in order for language to be productive, children cannot assume everything they haven’t yet heard is ungrammatical. So how do children learn which dative constructions are ungrammatical? This leamability problem has become known as Baker’s Paradox (Baker, 1979). Children learning the dative shift do not seem to receive negative evidence, yet 16

inexplicably manage to learn which forms are not grammatical. Yang (2016) provided a resolution to Baker’s Paradox, suggesting that positive evidence is sufficient to learn grammatical constraints. This can be tested for using the Sufficiency Principle, a derivation of the Tolerance Principle (14). According to the Sufficiency Principle, a rule is not generalized until the difference between N, the total number of items, and M, the number of items that follow a rule, is less than or equal to the threshold On. The formula of the Sufficiency Principle is quantitatively equivalent to the Tolerance Principle. If all items are either rule-following or exceptions, then N = M+e. Therefore, N-Mof the Sufficiency Principle is equal to e of the Tolerance Principle.

(14) The Sufficiency Principle Let R be a generalization over N items, of which M items are attested to follow R. R can be extended to all N items if and only if: N N - M < On where On = \nN

Although the two are numerically equivalent, the logic of the Sufficiency Principle differs from that of the Tolerance Principle. The Tolerance Principle tests the productivity of a rule based on the number of exceptions to that rule. This relies on the presence of exceptions; for example, children hearing irregular verbs such as ate or cut. In contrast, the Sufficiency Principle tests the amount of positive evidence required before a generalization can be made. If a child has only heard one dative verb used in both the DOC and PPC, that would not be sufficient evidence to generalize other dative verbs from one construction to the other. Therefore, to resolve Baker’s Paradox, Yang (2016) had to explain how children can learn the dative shift using positive evidence alone. This was done using the Sufficiency Principle to test for productive generalizations. 17

Yang’s resolution to Baker’s Paradox centered around the idea that children use semantic features of verbs to determine what constructions they can be used in. There is evidence that children are primed with semantic primitives, and thereby use the meaning of a word to determine what constructions it can be used in (Pinker, 1989). In the case of the dative verbs, Yang (2016) tested the productivity of dative constructions within the semantic subclass of caused possession verbs. Like the name suggests, caused possession verbs are verbs that inherently signify an action that causes someone to be in possession of something. Initially, it was thought that all dative verbs in the DOC carried a caused possession meaning, and all dative verbs in the PPC carried a caused motion meaning. Hovav and Levin (2008) disputed this and proposed that some dative verbs inherently held a caused possession meaning and have the of caused possession in both the DOC and PPC (15-1). These verbs typically signify acts of giving, future having, and communication. Other dative verbs only have the semantics of caused possession in the DOC (15-2). In the PPC, they have the semantics of caused motion. These include verbs of sending, instantaneous causation of ballistic motion, bring and take, and instrument of communication. Although there is no comprehensive list of all caused possession verbs, these general properties are agreed upon.

(15) A classification of caused possession words from Hovav and Levin (2008): 1. Dative verbs having only a caused possession meaning (a) Verbs that inherently signify acts of giving: give, hand, lend, loan, pass, rent, sell,... (b) Verbs of future having: allocate, allow, bequeath, grant, offer, owe, promise,... (c) Verbs of communication: tell, show, ask, teach, read, write, quote, cite,... 18

2. Dative verbs having both caused motion and possession meanings (a) Verbs of sending (send-type verbs): forward, mail, send, ship,

(b) Verbs of instantaneous causation of ballistic motion (throw- type verbs): fling, flip, kick, lob, slap, shoot, throw, toss,... (c) Verbs of causation of accompanied motion in a deictically specified direction: bring, take (d) Verbs of instrument of communication: e-mail, fax, radio, wire, telegraph, telephone,...

This leaves some ambiguity in whether other subclasses of dative verbs should be considered verbs of caused possession. ‘Verbs of manner of speaking’ (e.g. mumble, shout, whisper) are somewhat similar to ‘verbs of communication’ (e.g. tell, show, ask), and Yang (2016) classified both as caused possession in his analysis. Other dative verbs, such as ‘verbs of fulfilling’ (e.g. credit, present, entrust) and ‘verbs of continuous causation of accompanied motion in some manner’ (e.g. carry, pull, push) are sometimes classified as caused possession (Beavers, 2010), although Yang (2016) did not include them in his classification. (We will include the latter two categories in our modified analysis in Section 4.) Yang’s sample of dative verbs is based on an exhaustive list from Beth Levin’s English Verb Classes and Alternations: A Preliminary Investigation (1993). The total list exceeds 300 verbs, many of which are very uncommon and most likely won’t be heard by a child. For that reason, Yang trimmed this list to dative verbs that appeared in CHILDES, a child-based corpus (MacWhinney, 2000). The CHILDES corpus consists of transcripts collected from conversations with children. Yang’s analysis focused solely on child-directed speech, or speech spoken by adults to children. This provides an estimated sample of the dative verbs children are likely to hear. 19

2.2.1 CHILDES Corpus Extraction Yang (2016) extracted all verbs from Levin’s list appearing in either the DOC or PPC from the CHILDES corpus (MacWhinney, 2000). This resulted in a total of 63 verbs in the PPC and 42 in the DOC. Dative verbs that appeared in both constructions were included in both lists.

(16) Dative verbs extracted from CHILDES (Yang, 2016) a. Dative verbs in PPC (N- 63): address, assign, award, bounce, bring, carry, credit, deliver, describe, donate, drag, drop, explain, feed, flip, give, grant, guarantee, hand, haul, hit, hoist, introduce, kick, leave, lend, lift, mail, mention, offer, pass, pay, present, promise, pull, push, quote, raise, read, rent, report, return, roll, say, sell, send, serve, ship, shoot, shout, shove, show, sing, take, teach, tell, throw, toss, tow, trade, transport, whisper, write b. Dative verbs in DOC (N= 42): ask, assign, bet, blast, bring, call, charge, consider, cost, feed, find, give, grant, guarantee, hand, leave, lend, loan, mail, make, name, offer, owe, pass, pay, promise, pronounce, quote, read, rent, sell, send, serve, show, sing, take, teach, tell, throw, toss, trade, write

2.2.2 Analysis 1: Establishing Caused Possession Rules The first part of Yang’s analysis focused on two broad-ranged rules relating to semantic conditions: 1) a verb with the meaning of caused possession can be applied to the PPC, 2) a verb with the meaning of caused possession can be applied to the DOC. Broad-range rules can be seen as parallel to the general rules in Yang’s analysis of the German plural noun suffixes. A productive broad-range rule would be applied if a verb does not have characteristics that apply to any narrow-range rules. If a productive 20

broad-range rule cannot be found, verbs are recursively divided into subclasses applying to increasingly narrow-range rules.

Caused Possession => Prepositional Construction (17) Caused possession verbs that could apply to the PPC (N= 36) a. Caused possession in PPC (M = 35): address, assign, award, bring, deliver, describe, donate, explain, feed, give, grant, guarantee, introduce, leave, lend, mention, offer, pass, pay, promise, quote, read, rent, report, return, say, sell, serve, show, take, teach, tell, trade, transport, write b. Caused possession not in PPC (e = 1): ask

To test this rule, Yang collected all verbs of caused possession from CHILDES (MacWhinney, 2000) that could apply to the PPC. Recall that in the PPC, many verbs of caused possession become verbs of caused motion. Therefore, there are very few exceptions to this rule. Yang only found 1, ‘ask’, which is only grammatical in the DOC. As the number of exceptions is so low, we can clearly expect this rule to be productive. The threshold for N = 36 was calculated to be 10. The productivity of this rule suggests that children can generalize verbs with the semantics of caused possession to the prepositional construction.

(18) The Sufficiency Principle states:

N - M < On where On = N / ln/V 1 < #36 where 036 - 36/ln36 =10 21

Caused Possession => Double Object Construction (19) Caused possession verbs that could apply to the DOC (N = 49) a. Caused possession in DOC (M =38): ask, assign, bet, blast, bring, charge, cost, feed, find, give, grant, guarantee, hand, leave, lend, loan, mail, make, offer, owe, pass, pay, promise, quote, read, rent, sell, send, serve, show, sing, take, teach, tell, throw, toss, trade, write b. Caused possession not in DOC (e = 11): address, deliver, describe, explain, introduce, mention, report, return, say, ship, transport

Similarly, Yang (2016) tested the productivity for caused possession verbs to be generalized to the DOC. In this case the number of exceptions is much higher. With a total N = 49, the threshold for productivity was calculated to be #49 = 12. This is just one unit higher than the number of exceptions (e = 11). Therefore, this rule is also predicted to be productive under the Tolerance Principle, but just barely.

(20) The Sufficiency Principle states:

N - M < On where On - N/\nN

11 <049 where O49 = 49/ln49 =12

Within this sample of verbs, it’s possible for children to learn that caused possession verbs can be generalized to the DOC. However, there will realistically be some variation between children in their acquisition of dative verbs. It is possible that caused possession verbs would not be productive for the DOC in another sample of dative verbs, or that the rule may not even be productive in the long term.

2.2.3 Analysis 2: Developmental Stages of a Caused Possession Rule To test the stability of the rule caused possession => DOC, Yang (2016) looked into its developmental stages. As verbs with higher frequencies are typically learned 22

before verbs with lower frequencies, Yang modeled developmental stages of this rule by ranking caused possession verbs by their frequency. This was done using the SUBTLEX-US corpus (Brysbaert & New, 2009), an adult language corpus that includes word frequency and part-of-speech information. Yang (2016) obtained the frequency of all verbs extracted from CHILDES (MacWhinney, 2000). The verb with the lowest frequency was ‘h o i s t with a frequency of 84 in 51 million. Yang then obtained the frequency of all caused possession verbs from Levin’s list (Levin, 1993) whose primary part-of-speech on SUBTLEX-US was listed as verb. Verbs such as ‘hand* or ‘maiV, which are more frequently used as nouns, were excluded. This resulted in a total of 92 verbs, 52 of which were expected to take the DOC and 40 that were not.

(21) Caused possession verbs from SUBTLEX-US (N= 92) a. Caused possession verbs expected to take DOC (M =52): ask, assign, bet, bring, carry, chuck, drag, extend, feed, flip, give, guarantee, haul, heave, hit, kick, lend, offer, owe, pass, pay, preach, promise, pull, push, quote, read, render, repay, save, sell, send, shoot, shove, show, sign, sing, slam, slap, slip, smuggle, sneak, spare, take, teach, tell, throw, toss, vote, wager, write, yield b. Caused possession not expected to take DOC (e = 40): administer, admit, announce, communicate, confess, contribute, convey, cry, declare, deliver, demonstrate, describe, dictate, dispatch, donate, explain, express, forfeit, holler, introduce, mention, propose, provide, recite, recommend, refer, repeat, restore, return, reveal, say, scream, shout, snap, squeal, submit, surrender, trust, whisper, yell

After ranking caused possession verbs by their SUBTLEX-US frequency, Yang (2016) found that the rule caused possession => DOC loses productivity in the top 50 most frequent caused possession verbs, about halfway through this child-based sample of 23

verbs. (See Table 3). Therefore, this broad-range rule is not sufficient in the long term. Yang hypothesized that the search to maximize productivity is continuous, and if a larger productive rule cannot be found, children will seek out productive narrow-range rules.

Top N On e Productive?

10 4 1 Yes

20 7 3 Yes

30 9 4 Yes

40 11 10 Yes

50 13 16 No

60 15 21 No

70 16 27 No

80 18 34 No

90 20 38 No

92 20 40 No Table 3. Productivity of the rule caused possession => double object construction across verb frequencies from the SUBTLEX-US corpus (Yang, 2016).

2.2.4 Analysis 3: Narrow Range Rules To test for narrow-range rules, key characteristics of caused possession verbs need to be identified. Yang (2016) noted that the phonological properties of words are closely linked to their syntactic categories. Therefore, to test the productivity of narrow- range rules, Yang looked at the characteristics of 1) word syllables and 2) word stress. 24

2.2.4.1 Syllabic Rules A syllable is a unit of sound made up of a vowel bordered by two optional consonants. Monosyllabic words are those with just one syllable, while polysyllabic words have two or more syllables. Yang (2016) tested the productivity of monosyllabic verbs and polysyllabic verbs separately. Of the caused possession verbs from SUBTLEX-US, 50 were monosyllabic and 42 were polysyllabic.

Caused Possession + Monosyllabic => Double Object Construction (22) Caused possession monosyllabic verbs that could apply to the DOC (N- 50) a. Caused possession + monosyllabic in DOC (M = 42): ask, bet, bring, chuck, drag, feed, flip, give, haul, heave, hit, kick, lend, owe, pass, pay, preach, pull, push, quote, read, save, sell, send, shoot, shove, show, sign, sing, slam, slap, slip, sneak, spare, take, teach, tell, throw, toss, vote, write, yield b. Caused possession + monosyllabic not in DOC (e = 8): cry, say, scream, shout, snap, squeal, trust, yell

There was a total of 50 monosyllabic verbs of caused possession, of which only 8 did not appear in the DOC. These 8 exceptions fall under the threshold dso = 12. Therefore, Yang (2016) expects this rule to be productive.

(23) The Sufficiency Principle states:

N - M < 6n where 6n = N/lnN

8 < 050 where 650 = 50/ln50 = 12 25

Caused Possession + Polysyllabic => Double Object Construction (24) Caused possession polysyllabic verbs that could apply to the DOC (N = 42) a. Caused possession + polysyllabic in DOC (M= 10): assign, carry, extend, guarantee, offer, promise, render, repay, smuggle, wager b. Caused possession + polysyllabic not in DOC (e = 32): administer, admit, announce, communicate, confess, contribute, convey, declare, deliver, demonstrate, describe, dictate, dispatch, donate, explain, express, forfeit, holler, introduce, mention, propose, provide, recite, recommend, refer, repeat, restore, return, reveal, submit, surrender, whisper

Out of the 42 polysyllabic verbs of caused possession, only 10 are expected to appear in the DOC. The remaining 32 exceptions exceed the threshold #42 = 11.

(25) The Sufficiency Principle states:

N -M < Qn where On - N/\nN

32 is not < 042 where 642 = 42 / ln42 =11

2.2A.2 Stress Properties As a productive syllabic rule could not be found for the DOC within the polysyllabic caused possession verbs, Yang (2016) further divided them based on their stress properties. Stress refers to the special emphasis given to one of the syllables in a word. Yang (2016) tested the productivity of verbs that had stress on the first syllable and words that did not have stress on the first syllable. Of the 42 polysyllabic caused possession verbs, 12 were stress-initial and 30 were not. 26

Caused Possession + Stress-Initial => Double Object Construction (26) Caused possession stress-initial verbs that could apply to the DOC (N= 12) a. Caused possession + stress-initial in DOC (M= 6): carry, offer, promise, render, smuggle, wager b. Caused possession + stress-initial not in DOC (e = 6): demonstrate, donate, forfeit, holler, mention, whisper

Very few of the polysyllabic verbs were stress-initial. With a total jVof 12 caused possession stress-initial verbs, 6 appeared in the DOC and 6 did not. The threshold was calculated to be 4. As the number of exceptions exceeded the threshold, this rule was not expected to be productive.

(27) The Sufficiency Principle states:

N -M < 6 n where 6 n = N/lnN

6 is not < 0i2 where 8n = 12 / lnl2 = 4

Caused Possession + Non Stress-Initial => Double Object Construction (28) Caused possession non stress-initial verbs that could apply to the DOC (N= 30) a. Caused possession + not stress-initial in DOC (M= 4): assign, extend, guarantee, reply b. Caused possession + not stress-initial not in DOC (e = 26): administer, admit, announce, communicate, confess, contribute, convey, declare, deliver, describe, dictate, dispatch, explain, express, introduce, propose, provide, recite, recommend, refer, repeat, restore, return, reveal, submit, surrender 27

There was a total N of 30 caused possession non stress-initial verbs, of which only

4 appeared in the DOC. The 26 exceptions clearly surpass the threshold 630 = 8. Therefore, this rule is not expected to be productive.

(29) The Sufficiency Principle states:

N -M < 6n where 6n = N/InN

26 is not <030 where 630 = 30/ln30 = 8

2.2.5 Conclusions Of the four narrow range rules, Yang (2016) only determined one to be productive: caused possession + monosyllabic => DOC. Therefore, even after children can no longer generalize all caused possession verbs to the double object construction, they can generalize those with only one syllable to the DOC. For the remaining polysyllabic verbs, negative generalizations can be formed. Children can learn that, overall, polysyllabic caused possession verbs are unlikely to be found in the DOC, and those that are not stress-initial are the most unlikely to be found in the DOC. As such, this explains how children may retreat from errors in the dative shift through the use of positive evidence alone.

2.3 Discussion The Tolerance Principle provides a new approach to our understanding of language acquisition and how language is represented in the brain. In the past, there was contention between connectionist approaches, which proposed that all items—both rule- following and exceptions—were stored in associative memory (Rumelhart & McClelland, 1986), and computationalist approaches, which proposed that rule-following items were generated by rules, whereas exceptions were stored in associative memory (Pinker & Prince, 1988). In contrast, the Tolerance Principle suggests that multiple rules 28

can be productive simultaneously, and even exceptions may be generated by productive minority rules. The Tolerance Principle is a relatively new model of language acquisition and processing. With our future analyses, we hope to test whether there is sufficient evidence in favor of the Tolerance Principle as an adequate model of language acquisition. In Section 3 we will “replicate” Yang’s analysis of the dative shift. We will follow his methods as closely as possible, building assumptions when necessary to reach the same conclusions. In Section 4, we will conduct a modified analysis of the dative shift to address some of the limitations in Yang’s methods. 29

3. Analysis of the Dative Shift: Replication of Yang (2016)

In his book, The Price o f Linguistic Productivity. How Children Learn to Break the Rules o f Language, Yang provided a resolution to Baker’s Paradox using the Tolerance Principle. Baker’s Paradox is a leamability problem centered around how children learn negative constraints through positive evidence alone. One famous example is the dative shift. The dative shift is an alternation between two constructions, a double object construction (DOC) and a prepositional construction (PPC), also known as the to- dative form. In (1-3) the first sentence (a) corresponds with the DOC and the second sentence (b) corresponds with the PPC.

(1) a. I gave her the mail. b. I gave the mail to her. (2) a. I asked him a question. b. *1 asked a question to him. (3) a. *1 donated the museum a painting. b. I donated a painting to the museum.

Children learning the dative shift will see verbs such as in (1) be used in both the DOC and PPC and may form the generalization that all dative verbs can be extended from one construction to the other. There are, however, dative verbs that can only be used in the DOC (2), and others that can only be used in the PPC (3). Unlike children learning the English past tense, children learning the English dative shift do not receive evidence of what can’t be said. While a child who says eated will eventually hear ate, allowing them to rule out eated as ungrammatical, a child who hears a verb used in one construction does not have evidence for or against the other construction. 30

We outlined Yang’s analysis of the English dative shift in Section 2.2. In essence, Yang used the Sufficiency Principle, a derivation of the Tolerance Principle, to illustrate how children form productive rules about the English dative shift using positive evidence alone. This allows them to both extend their use of the dative shift as well as retreat from errors. In this section, we tested the English dative shift for productive rules following Yang’s methods (Yang, 2016) as closely as possible. In some cases, information was ambiguous, and we built certain assumptions to best match his results.

3.1 Materials English Verb Classes and Alternations: A Preliminary Investigation by Beth Levin (1993) is a classification of English verbs based on their syntactic and semantic properties. Levin lists all verbs that participate in dative constructions. Dative verbs are sorted based on their meaning (e.g. ‘give verbs’, ‘send verbs’, ‘drive verbs’, etc.) and their possible constructions (DOC and/or PPC). As some verbs have multiple meanings, there were 24 verbs that appeared more than once in Levin’s list. We eliminated duplicate verbs, resulting in a total count of 338 dative verbs. For the original, unmodified list, see Appendix A. The CHILDES corpus (MacWhinney, 2000) is an online database of transcripts collected from conversations with children (ages ranging from 6 months to 12 years). This corpus can be downloaded directly from the site in either a CHA or XML file format. We chose to use the CHA file format, as several subcorpera were not available in XML format. Our analysis of this corpus was limited to adult speech data, as it provides a representation of what children will hear. The SUBTLEX-US corpus (Brysbaert & New, 2009) is an adult language corpus based on movie subtitles, which includes word frequency and part-of-speech information. Word frequency information was used to model developmental stages of dative verb acquisition, as verbs with higher word frequencies are likely to be learned 31

before verbs with lower word frequencies. We only included words that had with a primary part-of-speech of verb in accordance with the method used by Yang (2016).

3.2 CHILDES Corpus Extraction First, we collected all dative verbs identified by Levin (1993) and collapsed them into a single list. Using a python script, we extracted all instances of dative verbs from the CHILDES corpus (MacWhinney, 2000). Then, we looked over the extracted data and determined whether each verb was used at least once in the DOC or PPC. Dative verbs that were not used in a were not included. Therefore, a construction like (4a) would be excluded. Furthermore, the constructions in (4b) and (4c) would be counted as two separate instances, as we tallied verbs in the DOC and PPC into two separate lists. Like Yang (2016), we did not rely on Levin’s judgment on which constructions a dative verb may appear in. For example, a verb that appeared in the double object construction in the CHILDES corpus was added to our list of double object verbs, even if it was recorded by Levin (1993) as only allowing the prepositional construction.

(4) a. Jane threw the ball. b. Jane threw me the ball. c. Jane threw the ball to me.

Our corpus search resulted in a total of 64 verbs that appeared in the prepositional construction and 45 verbs that appeared in the double object construction. Some alternating verbs, such as give, appear in both lists.

(5) Dative verbs extracted from CHILDES a. Dative Verbs in PPC (N = 64): address, assign, award, bounce, bring, carry, contribute, deliver, describe, drag, drive, drop, explain, feed, 32

give, hand, haul, hit, hoist, introduce, kick, leave, lend, lift, loan, mail, mention, offer, owe, pass, pay, pitch, present, promise, prove, provide, pull, push, raise, read, recommend, rent, report, return, roll, say, sell, send, serve, ship, shoot, shout, show, sing, take, teach, tell, throw, toss, tow, trade, whisper, write, yell b. Dative Verbs in DOC (iV = 44): ask, assign, bet, bring, call, charge, consider, cost, drop, feed, find, give, grant, guarantee, hand, leave, lend, loan, mail, make, name, offer, owe, pass, pay, promise, pronounce, rate*, read, roll, save, sell, send, serve, show, sing, take, teach, tell, throw, toss, trade, wish, write

At this point, our results diverge somewhat from Yang’s (2016), as there were 15 verbs (10 in the prepositional construction and 5 in the double object construction) identified in our corpus search that were not reported by Yang (2016), and a separate 11 verbs (8 in the prepositional construction and 3 in the double object construction) that were reported by Yang (2016) but did not appear in our corpus search. Therefore, our verb counts in the subsequent analyses will not be identical to those in Yang (2016).

(6) Verbs that appeared in Yang (2016) that we were not able to extract (n = 11) a. Verbs in PPC (n = 8): credit, donate, drag, flip, grant, guarantee, hoist, lift, quote, shove, tow, trade, transport b. Verbs in DOC (n = 3): blast, quote, rent

(7) Verbs that we were able to extract that did not appear in Yang (2016) (n= 15) a. Verbs in PPC (n = 10): contribute, drive, loan, owe, pitch, prove, provide, recommend, snap, yell b. Verbs in DOC (n = 5): drop, rate, roll, save, wish 33

3.3 Analysis 1: Establishing Caused Possession Rules As mentioned earlier, there is evidence that children use the semantic features of verbs to determine what constructions they can be used in. In the case of the dative verbs, we will be analyzing the semantic subclass caused possession. Like the name suggests, caused possession verbs describe an action that causes someone to be in possession of something. Our classification of caused possession verbs is based on the judgment of Hovav and Levin (2008) and Yang (2016). Hovav and Levin (2008) provide broad classes of verbs that have the semantics of caused possession, but do not list every applicable verb (See Appendix B). For this reason, we also relied on the assessment of Yang (2016), although there were some inconsistencies in his classification that will be expanded upon in the discussion. In addition to all broader caused possession categories listed in Hovav and Levin (2008), we also included ‘manner of speech verbs’ (e.g. mumble, whisper, yell) as they were classified as caused possession verbs by Yang (2016). In our analyses, we will use the Sufficiency Principle, a derivation of the Tolerance Principle, to test each construction of the dative shift for productivity within the caused possession semantic subclass. In other words, we are testing whether there is sufficient evidence for a child to learn that a verb with the semantics of caused possession verb can be used in either of the dative constructions.

(8) The Sufficiency Principle Let if be a generalization over N items, of which M items are attested to follow R. R can be extended to all N items if and only if: N N - M < 6 n where 6 n = 7~rT In N 34

3.3.1 Caused Possession => Prepositional Construction (9) Caused possession verbs that could apply to the PPC (N= 36) a. Caused possession in PPC (M = 34): address, assign, award, bring, contribute, deliver, describe, explain, feed, give, introduce, leave, lend, loan, mention, offer, owe, pass, pay, promise, read, recommend, rent, report, return, say, sell, serve, show, take, teach, tell, trade, write b. Caused possession not in PPC (e = 2): ask, save

For this rule, N would be the total number of caused possession verbs that could apply to the prepositional construction. Using the criteria specified above, we find that N = 36. Of these 36 verbs of caused possession, 34 appeared in the prepositional construction in the CHILDES corpus (MacWhinney, 2000), and 2 did not. The 2 verbs that did not appear in the prepositional construction are considered exceptions to this rule, and therefore e = 2. Now, we calculate 6n, the threshold for productivity. With a total of

36 verbs of caused possession, 6n = 36 / ln(36) = 10. When we insert these values into the formula for the Sufficiency Principle we find:

(10) The Sufficiency Principle states:

N - M < 6n where 6n = N / \ n N 2 < 036 where 036 = 36/ln36 =10

This holds true, as the number of exceptions falls well under the threshold. Therefore, we can predict that this rule is productive. Our outcome here is near identical to Yang’s, with the addition of one exception. Still, there is little contention that children can generalize caused possession verbs to the prepositional construction. 35

3.3.2 Caused Possession => Double Object Construction (11) Caused possession verbs that could apply to the DOC (N = 52) a. Caused possession in DOC (M = 36): ask, assign, bet, bring, charge, cost, feed, find, give, grant, guarantee, hand, leave, lend, loan, mail, make, offer, owe, pass, pay, promise, read, save, sell, send, serve, show, sing, take, teach, tell, throw, toss, trade, write b. Caused possession not in DOC (e = 16): address, contribute, deliver, describe, explain, introduce, mention, pitch, provide, recommend, rent, report, return, say, ship, yell

Our sample of caused possession verbs that apply to the DOC differs somewhat from Yang’s. There are 2 fewer caused possession verbs that apply to the DOC, and 5 more exceptions. Based on our current sample, this rule cannot be expected to be productive. The threshold for a vocabulary size of 52 items was calculated to be 13, which is not sufficient for the 16 exceptions we found. In contrast, this rule met the threshold for productivity in Yang (2016), but just barely.

(12) The Sufficiency Principle states:

N - M < On where On = N/]nN

16 is not < 652 where 652 = 52/ln52 =13

3.3.3 Conclusions Even when using the same criteria to classify verbs as caused possession, our results differed from Yang (2016) for the rule Caused Possession => Double Object Construction. It’s possible this may be due to some variability in the sample of dative verbs we extracted from CHILDES (MacWhinney, 2000). However, even Yang acknowledged the productivity of this rule was tenuous. In Analysis 2 we will 36

investigate the developmental stages of this rule, and test for productive narrow-range rules.

3.4 Analysis 2: Developmental Stages of a Caused Possession Rule 3.4.1 Procedure In this analysis, we created a bootstrapped list of the caused possession verbs we expect children to hear. Words with higher frequencies are typically learned before words with lower frequencies. Therefore, we used the SUBTLEX-US corpus (Brysbaert & New, 2009) to identify the frequencies of caused possession verbs and used this information to model developmental stages of learning. Like Yang (2016), we identified the CHILDES verb with the lowest frequency. This can be thought of as the most uncommon verb (or the lowest frequency of a verb) that children will hear, and therefore we can assume that children will hear all verbs with higher frequencies. The lowest frequency verb ‘hoist’, with a frequency of 84 in 51 million. We then identified the frequency of all dative verbs on Levin’s list (Levin, 1993) that had a frequency higher than 84 in 51 million. Additionally, we excluded verbs that did not have ‘verb’ listed as their primary part of speech. For example, ‘hand", which has a primary part-of-speech as ‘noun’, would not be included, as its frequency on SUBTLEX-US would primarily refer to its use as a noun. Then, we removed all verbs that would not be classified as caused possession under Yang (2016).

3.4.2 Results Although we followed the same procedure outlined by Yang (2016), we identified an additional 5 verbs of caused possession in the SUBTLEX-US corpus: bash, leave, pose, serve, and will. This led us to a total of 97 verbs of caused possession, 57 expected to take the DOC and 40 not expected to take the DOC. With this many exceptions, it’s impossible for the caused possession => double object construction rule to be productive. 37

(13) Caused possession verbs from SUBTLEX-US (N= 97) a. Caused possession verbs expected to take DOC (M = 57): ask, assign, bash, bet, bring, carry, chuck, drag, extend, feed, flip, give, guarantee, haul, heave, hit, kick, leave, lend, offer, owe, pass, pay, pose, preach, promise, pull, push, quote, read, render, repay, save, sell, send, serve, shoot, shove, show, sign, sing, slam, slap, slip, smuggle, sneak, spare, take, teach, tell, throw, toss, vote, wager, will, write, yield b. Caused possession not expected to take DOC (e = 40): administer, admit, announce, communicate, confess, contribute, convey, cry, declare, deliver, demonstrate, describe, dictate, dispatch, donate, explain, express, forfeit, holler, introduce, mention, propose, provide, recite, recommend, refer, repeat, restore, return, reveal, say, scream, shout, snap, squeal, submit, surrender, trust, whisper, yell

Table 4 lists the productivity of the rule across word frequencies. Within the top 40 most frequent caused possession verbs, the number of exceptions falls below the threshold and the rule maintains productivity. However, beyond the top 40 most frequent caused possession verbs, the number of exceptions grows too large, and this rule fails to stay productive. As a consequence, children can no longer generalize that verbs with the meaning of caused possession can be used in the double object construction. In Yang (2016) we find the same outcome, in which the rule caused possession => DOC loses productivity after the top 40 most frequent verbs. 38

T o p # On e Productive? e Productive? (Yang, 2016)

10 4 1 Yes 1 Yes

20 7 2 Yes 3 Yes

30 9 4 Yes 4 Yes

40 11 9 Yes 10 Yes

50 13 14 No 16 No

60 15 20 No 21 No

70 16 26 No 27 No

80 18 32 No 34 No

90 20 37 No 38 No

97 21 40 No -- — Table 4. Productivity of the rule caused possession => double object construction across verb frequencies from the SUBTLEX-US corpus (Brysbaert & New, 2009). A between Yang (2016) and this study.

3.5 Analysis 3: Narrow-Range Rules At this point, children are predicted to resort to narrow-range rules to learn the exceptions. The productivity of the rules caused possession => PPC and caused possession => DOC would be described as broad-range rules. After acquiring broad- range rules, children extend their usage of the dative shift from one construction to the other. Then, to learn the more specific restrictions, narrow-range rules are used. In this case, caused possession verbs are no longer productive in the DOC after the top 40 verbs from SUBTLEX-US (Brysbaert & New, 2009). Children are predicted to use the Tolerance Principle recursively to find productive narrow-range rules. Narrow-range rules can be thought of as subsetted rules. For example, if a verb has the semantics of caused possession and a secondary property, then it can be used in a certain construction. 39

Following the procedure used by Yang (2016), we will be analyzing the phonological properties of caused possession verbs, as the phonological properties of words have been linked to their syntactic properties.

3.5.1 Syllabic Rules A syllable is a unit of sound made up of a vowel bordered by two optional consonants. While some verbs have one syllable (ask, lend, owe), others have multiple {deliver, donate, return). Of the 97 caused possession verbs we extracted from SUBTLEX-US, 55 were monosyllabic and 42 were polysyllabic. We tested the productivity of monosyllabic verbs and polysyllabic verbs separately.

Caused Possession + Monosyllabic => Double Object Construction (14) Caused possession monosyllabic verbs that could apply to the DOC (N= 55) a. Caused possession + monosyllabic in DOC (M= 47): ask, bash, bet, bring, chuck, drag, feed, flip, give, haul, heave, hit, kick, leave, lend, owe, pass, pay, pose, preach, pull, push, quote, read, save, sell, send, serve, shoot, shove, show, sign, sing, slam, slap, slip, sneak, spare, take, teach, tell, throw, toss, vote, will, write, yield b. Caused possession + monosyllabic not in DOC (e = 8): cry, say, scream, shout, snap, squeal, trust, yell

The five additional verbs we extracted from SUBTLEX-US were all monosyllabic verbs. Therefore, our total N is 5 verbs larger than that of Yang (2016). These five verbs are: bash, leave, pose, serve, and will. All five are expected to appear in the DOC, also increasing the value of M. With a total of 55 monosyllabic verbs of caused possession, 47 were expected to take the DOC and 8 were not. The threshold was calculated to be 13. As the number of exceptions fell within the threshold, this rule was predicted to be productive. 40

(15) The Sufficiency Principle states:

N - M < On where On - N / \nN

8 < O55 where O55 = 55/ln55 = 13

Caused Possession + Polysyllabic => Double Object Construction (16) Caused possession polysyllabic verbs that could apply to the DOC (N= 42) a. Caused possession + polysyllabic in DOC (M = 10): assign, carry, extend, guarantee, offer, promise, render, repay, smuggle, wager b. Caused possession + polysyllabic not in DOC (e = 32): administer, admit, announce, communicate, confess, contribute, convey, declare, deliver, demonstrate, describe, dictate, dispatch, donate, explain, express, forfeit, holler, introduce, mention, propose, provide, recite, recommend, refer, repeat, restore, return, reveal, submit, surrender, whisper

The values of our polysyllabic caused possession verbs are identical to those in Yang (2016). Out of a total N- 42 polysyllabic caused possession verbs, only 10 are expected in the DOC. With 32 exceptions, we cannot expect there to be sufficient evidence for children to generalize this rule to other polysyllabic caused possession verbs.

(17) The Sufficiency Principle states:

N - M < On where On = N / lnN

32 is not < 042 where 8 4 2 = 42/ ln42 = 1 1

3.5.2 Stress Properties The 42 polysyllabic verbs are now broken down into more narrow properties in search of a productive rule. In the following analyses we will test verbs for productivity 41

based on their stress properties. Stress refers to the special emphasis given to one of the syllables in a word. We compared the likelihood of verbs with primary stress and verbs without primary stress to be used in the DOC.

Caused Possession + Stress-Initial => Double Object Construction

(18) Caused possession stress-initial verbs that could apply to the DOC (jV = 13) a. Caused possession in DOC (M = 6): carry, offer, promise, render, smuggle, wager b. Caused possession not in DOC (e = 6): demonstrate, donate, forfeit, holler, mention, whisper

Our values here are identical to those in Yang (2016). Of our 12 stress-initial verbs of caused possession, only 6 are expected to appear in the DOC. This is not predicted to be sufficient evidence for children to generalize this rule, as the 6 exceptions exceed the threshold 9 n = 4.

(19) The Sufficiency Principle states:

N - M < On where On - N/\nN 6 is not < 0i2 where 0i2 - 1 2 /ln l2 = 4

Caused Possession + Non Stress-Initial => Double Object Construction (20) Caused possession non stress-initial verbs that could apply to the DOC (N= 30) a. Caused possession in DOC (M= 4): assign, extend, guarantee, repay b. Caused possession not in DOC (e = 26): administer, admit, announce, communicate, confess, contribute, convey, declare, deliver, describe, dictate, dispatch, explain, express, introduce, propose, 42

provide, recite, recommend, refer, repeat, restore, return, reveal, submit, surrender

Again, our values are equal to those in Yang (2016). With a total of 30 caused possession verbs that are not stress-initial, only 4 are expected in the DOC. This is clearly not sufficient evidence for generalization, as the 26 exceptions significantly outweigh the threshold.

(21) The Sufficiency Principle states:

N - M < On where On = N / ln/V

26 is not < 030 where O30 - 30/ln30 = 8

3.5.3 Conclusions As with Yang (2016), we found there is sufficient evidence for children to form productive narrow-range rules once the generalization caused possession => DOC is no longer productive. The only verbs that easily lend themselves to the DOC are the monosyllabic caused possession verbs. Of the remaining polysyllabic caused possession verbs, neither the stress-initial nor the non stress-initial verbs could productively be generalized to the DOC. However, about half of the stress initial verbs were expected to appear in the DOC, whereas nearly all the non stress-initial verbs were not expected to appear in the DOC. Using this information, children can form the negative generalization that non stress-initial verbs are extremely unlikely to appear in the DOC. Finally, the polysyllabic verbs that are expected to appear in the DOC are committed to associative memory.

3.6 Discussion The main purpose of our study was to test whether the Tolerance Principle can explain acquisition of the English dative shift. We wanted to test whether there is 43

sufficient evidence for children to both use the dative shift productively, but also retreat from errors of overregularization. In this study, we attempt an analysis of the English dative shift using the same materials and procedure in Yang (2016). Overall, we were able to reach similar conclusions. We found that children can productively generalize verbs of caused possession to the PPC. When it comes to the DOC, a productive caused possession rule is more tenuous. After looking at word frequencies, we found that caused possession words can productively be extended to the DOC during the early stages of dative acquisition.

3.6.1 Limitations We aimed to produce results similar to Yang (2016) following his methods as closely as possible. In some instances, Yang’s methodology was not clear, and assumptions were required. After inspecting Yang’s classification system in closer detail, a few inconsistencies were noted. For example, Yang classified both ‘explain’ and iaward' as caused possession verbs in his PPC analysis. Therefore, both these verbs were added to the total N that could apply to the rule caused possession => PPC. However, when testing the DOC, Yang included ‘explain’ but not ‘award? in his total N of verbs that could apply to the rule caused possession -> DOC. Neither of these verbs appeared in the DOC, and therefore would both be listed as exceptions. Understating the number of exceptions may have caused rules to be marked as productive when in reality, they aren’t. There were three additional verbs that should have been added to the DOC analysis, and 5 that should have been applied to the PPC. Furthermore, Yang included two verbs that were unlikely to have appeared in a dative construction: ‘fin d ’, a ‘declare verb’ and ‘make’, a ‘dub verb’. Used datively, these verbs would form sentences similar to those in (22). However, these verbs are more commonly used in the context of the benefactive alternation (23). On the surface, the benefactive alternation is very similar to the dative shift; the alternation occurs between a DOC and PPC. However, the benefactive alternation typically relates to verbs 44

with the meaning of ‘creation’ or ‘obtaining’ and can be grammatical even while omitting the benefactive (Levin, 1993), as seen in (24). Additionally, the PPC is formed with the preposition Tor’ rather than ‘to’.

(22) a. I found you guilty. b. I made you a king. (23) a. I found you a jacket. / / 1 found a jacket for you. b. I made you a cake. / / 1 made a cake for you. (24) a. *1 made a king. b. I made a cake.

Lastly, we return to the fact that there is not a comprehensive list of all caused possession verbs. This is not necessarily a limitation in Yang’s work; rather, we wanted to test whether similar conclusions could be reached using a different interpretation of caused possession verbs. Yang (2016) did not include ‘verbs of fulfilling’ and ‘carry verbs’ in his classification of caused possession, though there is reason to believe they should be included (Beavers, 2010). ‘Verbs of fulfilling’ may fit the classification of ‘verbs that inherently signify acts of giving’ based on Hovav and Levin (2008). And in accordance with other movement based verbs, it’s likely the ‘carry verbs’ have the semantics of caused possession in the DOC and caused motion in the PPC. ‘Manner of speaking verbs’, which Yang did include in his classification, are not typically associated with the semantics of caused possession (Beavers, 2010). A construction such as (25) does not confirm that Amy received the warning that was yelled. This may be the case of caused possession verbs becoming caused motion in the PPC. However, Levin (1993) does not predict ‘manner of speaking verbs’ to be grammatical in the DOC, and they were rarely used as such in the CHILDES corpus (MacWhinney, 2000). Therefore, we may remove this subclass in our modified analysis (Section 4). 45

(25) a. I yelled a warning to Amy. b. *1 yelled Amy a warning.

The previous “limitations” in Yang’s methods have to do with which verbs should be classified as caused possession. However, we also noted some problems with the SUBTLEX-US corpus (Brysbaert & New, 2009) as an identifier of verb frequencies. 1) A large number of verbs were excluded from our analyses because their primary part-of- speech wasn’t verb. SUBTLEX-US does report the frequency of each part of speech a word can take, so there is realistically no reason to exclude verbs that aren’t primarily used as verbs. 2) The frequency for verb tenses is listed separately. For example, there is a separate word frequency for ‘hoist' , ‘hoisted’, ‘hoisting’, and ‘hoists’, though Yang only included the frequency of the present tense.

3.6.2 Future Directions In the next section, we will try to address some of the limitations listed above. Instead of SUBTLEX-US (Brysbaert & New, 2009), we will be using the Corpus of Contemporary American English (COCA; Davies, 2008). COCA lists the frequencies of words by part of speech and lemma. Unlike our SUBTLEX-US analysis, we won’t need to exclude words that are not primarily used as verbs. Additionally, we will take into account the frequencies of all tenses of a verb. This means that we can include a larger and more informative sample of caused possession verbs when modeling developmental stages of the DOC. We will also be changing our classification of caused possession verbs to include ‘verbs of fulfilling’ and ‘carry verbs’, and removing ‘verbs of manner of speaking’. A few smaller changes will be made, which will be detailed in Section 4. 46

4. Analysis of the Dative Shift with Modifications

4.1 Materials Our sample of dative verbs consists of those extracted from the CHILDES corpus (MacWhinney, 2000) in Section 3.2. One small change was made. The verb ‘fin d 1 was removed, as the definition of the verb extracted did not match that on Levin’s list. In a dative context, ‘fin d is a ‘declare’ verb (e.g. ‘Ifin d you guilty'). The only instances of *fin d in the CHILDES corpus was with a benefactive definition (e.g. ‘Ifound you a shoe ’ / 'Ifound a shoe for you’) in which the alternation uses the preposition Tor’ rather than ‘to’. The dative verbs used in our modified analysis are listed below.

(1) Dative verbs extracted from CHILDES a. Dative Verbs in PPC (N = 64): address, assign, award, bounce, bring, carry, contribute, deliver, describe, drag, drive, drop, explain, feed, give, hand, haul, hit, hoist, introduce, kick, leave, lend, lift, loan, mail, mention, offer, owe, pass, pay, pitch, present, promise, prove, provide, pull, push, raise, read, recommend, rent, report, return, roll, say, sell, send, serve, ship, shoot, shout, show, sing, take, teach, tell, throw, toss, tow, trade, whisper, write, yell b. Dative Verbs in DOC (N= 43): ask, assign, bet, bring, call, charge, consider, cost, drop, feed, give, grant, guarantee, hand, leave, lend, loan, mail, make, name, offer, owe, pass, pay, promise, pronounce, rate*, read, roll, save, sell, send, serve, show, sing, take, teach, tell, throw, toss, trade, wish, write

We used the Corpus of Contemporary American English (COCA; Davies, 2008) in place of the SUBTLEX Corpus (Brysbaert & New, 2009) for our frequency analysis. COCA is a large, genre-balanced corpus of American English that lists word 47

frequencies by lemma and part of speech. This allowed us to include frequencies for all verb tenses (‘gave’ as well as ‘give’) and ensure that no dative verbs were excluded from our analysis because their primary part of speech wasn’t verb.

4.2 Analysis 1: Establishing Caused Possession Rules In the following sections we tested the productivity of two broad-range rules: 1) verbs with the semantics of caused possession can be generalized to the PPC, and 2) verbs with the semantics of caused possession can be generalized to the DOC. We did, however, change our classification of caused possession verbs from that made by Yang (2016). In summary, these changes included: 1) The inclusion of verbs that were inconsistently excluded from the N of all caused possession verbs. For example, based on Levin (1993), the verbs ‘explain’ and ‘award1 are only expected to be used in the PPC. However, when testing the rule caused possession => DOC, Yang (2016) included the verb ‘explain’ in his total N of all caused possession verbs that could apply to the DOC, but not ‘award*. We systematically went through each verb extracted by Yang (2016) and included all verbs that were inconsistently removed from previous analyses. Caused possession verbs that would actually be caused motion verbs in the PPC were not included in the N for our PPC analysis. 2) Reclassifying subclasses of caused possession verbs. As there is not a comprehensive list of all caused possession verbs, there is some variability in which subclasses of dative verbs have the semantics of caused possession. We removed ‘manner of speaking verbs’, as other analyses on caused possession verbs do not typically include them as such (Beavers, 2010). We also included all ‘verbs of fulfilling’ and ‘carry verbs’ as listed in Beavers (2010). In accordance with other movement based verbs, we classify the ‘carry verbs’ as caused possession only in the DOC and caused motion in the PPC. 48

3) Verbs that are not actually used in a dative context or do not match the definition listed in Levin (1993). This includes the verb ‘‘fin d ', a ‘declare verb’, mentioned in Section 4.1, as well as ‘make’, a ‘dub verb’. In a dative context, these verbs would form constructions similar to those in (2). The more common definition for these verbs can be found in (3), in which the DOC is actually a construction of the benefactive alternation. Like the dative shift, the benefactive alternation features a shift between two constructions—one being a double object construction and the other a prepositional construction. Though in this case, the prepositional form uses the preposition ‘fo r’ rather than ‘to’. Although similar to the dative shift, the benefactive alternation typically relates to verbs with the meaning of ‘creation’ or ‘obtaining’, and can be grammatical even while omitting the benefactive argument (Levin, 1993), as seen in (4). After removing ‘you’, only the second sentence is grammatical. The specific changes to caused possession verbs added or removed will be detailed in each of the following analyses.

(2) a. I found you guilty. b. I made you a king.

(3) a. I found you a jacket. / / 1 found a jacket for you. b. I made you a cake. / / 1 made a cake for you.

(4) a. *1 made a king. b. I made a cake.

4.2.1 Caused Possession => Prepositional Construction In summary, 3 verbs were removed from our classification of caused possession, and 4 verbs were added. The 3 verbs removed were: ‘address' , ‘bring and ‘take ‘Address' was removed as its context in the corpus (‘ Who do you wanna address this to?') does not indicate caused possession. ‘Bring ’ and ‘take should actually be classified as caused motion in the PPC according to Hovav and Levin (2008). 49

The 4 verbs that were added were: ‘b et\ ‘hand’, ‘p r e s e n tand ‘provide’. ‘Bet’ was added as a caused possession verb that is grammatically only expected to appear in the DOC, and was inexplicably excluded from the PPC analysis. tHand>, although categorized as a ‘send’ verb in Levin (1993), is specifically listed as a caused possession verb in Hovav and Levin (2008). ‘Present’ and ‘provide’ are both ‘verbs of fulfilling’, which were included after incorporating the subcategory into our classification of caused possession verbs. All verbs used in this analysis are listed in (5).

(5) Caused possession verbs that could apply to the PPC (N= 37) a. Caused possession in PPC (M = 34): assign, award, contribute, deliver, describe, explain, feed, give, hand, introduce, leave, lend, loan, mention, offer, owe, pass, pay, present, promise, provide, read, recommend, rent, report, return, say, sell, serve, show, teach, tell, trade, write b. Caused possession not in PPC (e = 3): ask, bet, save

As in our replication analysis of caused possession => PPC, the number of exceptions to this rule is very small. Of the 37 caused possession verbs in the CHILDES corpus (MacWhinney, 2000) that could apply to the PPC, only 3 do not appear in the PPC. The threshold for N = 37 was calculated to be 10. As e = 3 falls under 10, we again predict that it is productive for children to generalize caused possession verbs to the prepositional construction.

(6) The Sufficiency Principle states:

N - M < 6 n where 6 n = N / InN 3 < 037 where 037 = 37/ln37 =10 50

4.2.2 Caused Possession => Double Object Construction A total of 7 verbs that were classified as caused possession in our replication study were removed, and another 12 were added. The 7 verbs removed were: ‘address', ‘charge', ‘cost', ‘find', ‘make', ‘sing', ‘yell' . ‘Address', as mentioned in the previous section, did not appear in CHILDES (MacWhinney, 2000) in the context of caused possession. ‘Charge’ and ‘cost' similarly don’t imply caused possession, and instead seem to imply the loss of possession. ‘Find’ and ‘make ’ were removed for only appearing in the CHILDES corpus in the context of the benefactive alternation. ‘Sing' and ‘yell ’ were removed for being ‘manner of speaking verbs’, which we did not classify as caused possession verbs in our modified analysis (See Section 4.2). The 12 verbs added were: ‘award, ‘hit', ‘kick', ‘shoot' (listed as caused possession by Yang (2016) for the PPC analysis, but should reasonably also be included in the DOC analysis), ‘carry', ‘drag', ‘haul', ‘hoist', ‘present', ‘puli', ‘push,', ‘tow' (verbs that belong to either ‘verbs of fulfilling’ or ‘carry verbs’, which were both included in our classification of caused possession verbs in this analysis).

(7) Caused possession verbs that could apply to the DOC (N- 57) a. Caused possession in DOC (M = 31): ask, assign, bet, bring, feed, give, grant, guarantee, hand, leave, lend, loan, mail, offer, owe, pass, pay, promise, read, save, sell, send, serve, show, take, teach, tell, throw, toss, trade, write b. Caused possession not in DOC (e = 26): award, carry, contribute, deliver, describe, drag, explain, haul, hit, hoist, introduce, kick, mention, pitch, present, provide, pull, push, recommend, rent, report, return, say, ship, shoot, tow

As in our replication analysis, this rule does not meet the threshold for productivity. However, the sample from our replication analysis was only 3 exceptions 51

away from meeting the threshold for productivity, whereas in the present sample, e is nowhere close to the threshold. With a total N = 57 caused possession verbs, only 31 appeared in the DOC. This means that nearly half these items are exceptions. This is likely due to our inclusion of the ‘carry verbs’ as verbs of caused possession. Although Levin (1993) listed them as alternating, none of them appeared in the DOC, whereas 8 appeared in the PPC. Under our current classification of caused possession verbs, the DOC has no chance to be productive.

(8) The Sufficiency Principle states:

N - M < On where On = N / InN 26 is not < #57 where 057 = 57/ln57 = 14

4.2.3 Conclusions The productivity of caused possession verbs in the PPC was consistent across Yang (2016), our replication analysis, and the present analysis. In contrast, the productivity of caused possession verbs in the DOC has been more inconsistent. In Yang (2016), this rule just barely met the threshold for productivity. In our replication study, we found the rule to be just a few exceptions away from being productive. In our current analysis, the rule is nowhere close to being productive, as nearly half of all N items are exceptions. However, we don’t believe our outcome in 4.2.2 necessarily contradicts what Yang (2016) reported. After using word frequency information to model developmental stages of dative acquisition, Yang (2016) found that the rule caused possession => DOC was only productive for the top 40 most frequent caused possession verbs. Based on our results, it’s possible that this rule loses productivity earlier than previously thought. 52

4.3 Analysis 2: Developmental Stages of a Caused Possession Rule To test this, we modeled developmental stages using a different word frequency corpus, the Corpus of Contemporary American English (COCA; Davies, 2008). COCA has a much larger database than SUBTLEX-US (Brysbaert & New, 2009), which allowed us to include a greater number of caused possession verbs in our frequency analysis. Additionally, COCA lists word frequencies by lemma, which usefully includes the frequencies of all tenses of a verb (walk, walks, walked, walking).

4.3.1 Procedure First, we identified the word frequencies of all verbs extracted from the CHILDES corpus (MacWhinney, 2000). The verb with the lowest frequency was '‘loan’, which had a frequency of 923 in the COCA corpus (Davies, 2008). We then identified the word frequency of every dative verb in Levin’s list (Levin, 1993). Only eight verbs from Levin’s list did not appear at all (as a verb) in COCA: ‘modem’, ‘mulct’, ‘netmail', ‘satellite’, ‘semaphore’, ltisk’, "undercharge’, and ‘wireless’. Presumably, these eight are very rarely—if ever—used as verbs, let alone in a dative construction. Next, we collected a list of all dative verbs that had a frequency higher than the lowest CHILDES verb, ‘loan’, at 923. This resulted in a total of 249 dative verbs. We then excluded all dative verbs that did not have the semantics of caused possession, resulting in a final list of 148 caused possession verbs ranked by frequency (9).

(9) Caused possession verbs from COCA (N = 148) a. Caused possession verbs expected to take the DOC (M= 93): accord, advance, allocate, ask, assign, award, bash, bat, bet, bill, bring, carry, cede, chuck, cite, concede, drag, e-mail, extend, fax, feed, fine, flick, fling, flip, give, grant, guarantee, hand, haul, heave, hit, hoist, hurl, issue, kick, lease, leave, lend, loan, lug, mail, offer, owe, pass, pay, peddle, phone, pitch, pose, post, preach, promise, pull, push, quote, 53

read, relay, render, rent, repay, save, sell, send, serve, ship, shoot, shove, show, sign, signal, slam, slap, sling, slip, smuggle, sneak, spare, take, teach, telephone, tell, throw, tip, toss, tow, trade, tug, vote, will, wire, write, yield b. Caused possession verbs not expected to take the DOC (e - 55): administer, admit, allege, announce, articulate, assert, broadcast, communicate, confess, contribute, convey, credit, declare, delegate, deliver, demonstrate, denounce, describe, dictate, dispatch, display, distribute, donate, entrust, exhibit, explain, express, forfeit, furnish, illustrate, introduce, mention, narrate, portray, present, propose, provide, recite, recommend, recount, refer, reimburse, repeat, report, restore, return, reveal, sacrifice, say, state, submit, supply, surrender, transfer, trust

4.3.2 Results As seen in Table 5, the rule caused possession => DOC was initially productive within the top 30 most frequent caused possession verbs. The vast majority of caused possession verbs in this stage could be used in the DOC. There were only 6 exceptions, which fell under the threshold of 9. Immediately after the top 30 most frequent verbs, the number of exceptions exceeds the threshold and the rule fails to maintain productivity. This differs somewhat from both Yang (2016) and our replication analysis, in which the caused possession => DOC rule did not lose productivity until after the top 40 most frequent caused possession verbs. Though overall, a similar pattern is seen. Initially, children can generalize caused possession verbs to the DOC. Eventually, this is no longer possible, and children resort to narrow range rules to learn which caused possession verbs can be used in the DOC. Our frequency analysis from COCA (Davies, 2008) suggests that this may occur earlier on in the dative learning process than Yang (2016) estimated. In our frequency 54

table, this occurs in the top 40 of our 148-verb list, meaning that the rule was productive for less than a third of our total sample. In contrast, the rule only loses productivity about halfway through in Yang (2016), in the top 50 verbs of his 92-verb list.

T op # 6n e Productive? e Productive? e Productive? Dative Yang Replication (2016)

10 4 1 Yes 1 Yes 1 Yes

20 7 3 Yes 2 Yes 3 Yes

30 9 6 Yes 4 Yes 4 Yes

40 11 14 No 9 Yes 10 Yes

50 13 21 No 14 No 16 No

60 15 24 No 20 No 21 No

70 16 28 No 26 No 27 No

80 18 34 No 32 No 34 No

90 20 36 No 37 No 38 No

100 22 39 No - — - -

110 23 46 No — — - --

120 25 48 No -- — — -

130 27 50 No - — -—

140 28 51 No — — — -

148 30 55 No - ~ -- -- Table 5. The productivity of the rule caused possession => double object construction across word frequency stages. A comparison between this analysis, the replication analysis (Section 3.4) and Yang (2016). 55

4.4 Analysis 3: Narrow-Range Rules The extremely short-lived productivity of the caused possession -> DOC rule does not negate Yang’s second hypothesis. Once the semantics of caused possession are no longer sufficient to generalize verbs to the DOC, we expect children to search for productive narrow-range rules. Like we did in our replication study, we will explore the characteristics of word syllables and stress.

4.4.1 Syllabic Rules Caused Possession + Monosyllabic => Double Object Construction (10) Caused possession monosyllabic verbs that could apply to the DOC (N= 76) a. Caused possession + monosyllabic in DOC (M= 74): ask, bash, bat, bet, bill, bring, cede, chuck, cite, drag, fax, feed, fine, flick, fling, flip, give, grant, hand, haul, heave, hit, hoist, hurl, kick, lease, leave, lend, loan, lug, mail, owe, pass, pay, phone, pitch, pose, post, preach, pull, push, quote, read, rent, save, say, sell, send, serve, ship, shoot, shove, show, sign, slam, slap, sling, slip, sneak, spare, take, teach, tell, throw, tip, toss, tow, trade, tug, vote, will, wire, write, yield b. Caused possession + monosyllabic not in DOC (e = 2): state, trust

Our sample of caused possession verbs from COCA has far more monosyllabic verbs than SUBTLEX-US, and much fewer that are not expected to take the DOC. Out of a total N - 76 monosyllabic verbs of caused possession, only 2 are exceptions. As in Yang (2016) and our replication analysis, this rule easily passes the threshold for productivity under the Sufficiency Principle.

(11) The Sufficiency Principle states:

N-M < On where On - N / lmV 2 is <076 where 076 = 76/ln76 1 17 56

Caused Possession + Polysyllabic => Double Object Construction (12) Caused possession polysyllabic verbs that could apply to the DOC (N= 72) a. Caused possession + polysyllabic in DOC (M = 20): accord, advance, allocate, assign, award, carry, concede, e-mail, extend, guarantee, issue, offer, peddle, promise, relay, render, repay, signal, smuggle, telephone b. Caused possession + polysyllabic not in DOC (e = 52): administer, admit, allege, announce, articulate, assert, broadcast, communicate, confess, contribute, convey, credit, declare, delegate, deliver, demonstrate, denounce, describe, dictate, dispatch, display, distribute, donate, entrust, exhibit, explain, express, forfeit, furnish, illustrate, introduce, mention, narrate, portray, present, propose, provide, recite, recommend, recount, refer, reimburse, repeat, report, restore, return, reveal, sacrifice, submit, supply, surrender, transfer

Although our total sample (N= 72) of polysyllabic caused possession verbs is larger than our replication analysis (N = 42), we noticed the percentage of exceptions was approximately 75 percent in both samples (e = 52 and e = 32 respectively). The 72 verbs not expected to take the DOC far exceed the threshold On =16. This rule has consistently missed the threshold for productivity across Yang (2016), our replication analysis, and the present analysis. We will now take these 72 polysyllabic caused possession verbs and test for productive rules based on their stress properties.

(13) The Sufficiency Principle states:

N - M < On where On = N / \nN 52 is not < 072 where On = 72/ln72 = 16 57

4.4.2 Stress Properties A word’s stress refers to the special emphasis given to a syllable. Verbs with initial stress have that emphasis placed on the primary syllable. We will be testing two stress characteristics for productivity: stress-initial and not stress-initial. Verbs that are not stress-initial may have their emphasis in any other syllable of a word. This includes mid or final stress. The characteristic of non-initial stress is associated with Latinate verbs, which are averse to the DOC.

Caused Possession + Stress-Initial => Double Object Construction (14) Caused possession stress-initial verbs that could apply to the DOC (N = 23) a. Caused possession stress-initial in DOC (M= 11): allocate, carry, e- mail, issue, offer, peddle, promise, render, signal, smuggle, telephone b. Caused possession stress-initial not in DOC (e = 12): broadcast, credit, delegate, demonstrate, donate, forfeit, furnish, illustrate, mention, narrate, sacrifice, transfer

In our COCA sample, there are slightly fewer stress-initial caused possession that are expected in the DOC (M= 11) than there are those that are not expected in the DOC (e = 13). As the exceptions outnumber the rule-following items, we cannot expect this rule to be productive.

(15) The Sufficiency Principle states:

N -M < On where On = N / InN 13 is not < 024 where 024 = 24/ln24 = 7 58

Caused Possession + Non Stress-Initial => Double Object Construction (16) Caused possession non stress-initial verbs that could apply to the DOC (N= 49) a. Caused possession in DOC (M= 9): accord, advance, assign, award, concede, extend, guarantee, relay, repay b. Caused possession not in DOC (e = 40): administer, admit, allege, announce, articulate, assert, communicate, confess, contribute, convey, declare, deliver, denounce, describe, dictate, dispatch, display, distribute, entrust, exhibit, explain, express, introduce, portray, present, propose, provide, recite, recommend, recount, refer, reimburse, repeat, report, restore, return, reveal, submit, supply, surrender

When we observe the non stress-initial caused possession verbs, we see that the verbs not expected in the DOC (e = 39) significantly outweigh those that are expected in the DOC (M= 9). Again, we find that we cannot expect this rule to be productive.

(17) The Sufficiency Principle states:

N - M < On where On - N / l n N 39 is not < 048 where 048 = 48 /ln48 = 12

4.4.3 Conclusions Our analysis of narrow-range rules shows there is sufficient evidence for children to form productive generalizations about the DOC. As Yang (2016) proposed, these include negative generalizations. Out of the four narrow-range rules we tested, the only productive rule was Caused Possession + Monosyllabic => DOC. After this rule is applied, all remaining caused possession verbs are polysyllabic. From here, children can turn to the stress properties of verbs. Neither stress-initial nor non stress-initial caused 59

possession verbs were productive for the DOC. However, children can generalize that non-stress initial verbs are even more unlikely to be used in the DOC than stress-initial verbs. The use of these negative generalizations explains how children learn what not to say through positive evidence alone and provides a possible solution to Baker’s Paradox.

4.5 Discussion The changes we made to our classification of caused possession verbs in Section 5 may reflect a more conservative view of caused possession verbs that may partake in the DOC. According to Levin (1993), ‘verbs of fulfillment’, such as credit, provide, and supply, are only grammatical in the PPC. Our inclusion of these verbs means M, the number of caused possession verbs expected in the DOC, is much smaller relative to N, the total caused possession verbs. Therefore, the productivity of the rule caused possession => DOC is less likely. This explains our outcome in 4.2.2, where nearly half of the caused possession verbs did not appear in the DOC. Even with this interpretation, we found a similar developmental pattern to Yang (2016). Higher frequencies of caused possession verbs were productive in the DOC, suggesting that this generalization is possible in early stages of dative acquisition. Moreover, the productivity of monosyllabic caused possession verbs held true in our exploration of narrow-range rules, supporting the idea that children can use the Tolerance Principle recursively when broad-range rules are no longer sufficient.

4.5.1 Limitations In Section 4, we modified Yang’s design with the aim of improving the accuracy of our analyses. However, there are still some limitations in our methods. Yang (2016) reported that he extracted child-directed speech (speech spoken by adults to children) from CHILDES (MacWhinney, 2000). We presume this means he extracted all speech that was tagged with an adult speaker. However, the adult speech from CHILDES is not 60

necessarily spoken to children. In many cases, we extracted instances of dative verbs from the context of adults speaking to other adults. For example, a parent speaking to a researcher before a play session, or parents speaking to each other at the dinner table. Therefore, the dative verbs extracted from CHILDES are not limited to those an adult would use while speaking to a child. Nevertheless, in all these instances children are present, so it can be argued they still absorb information from adult to adult conversations. Additionally, the CHA format of the CHILDES corpus comes with its own host of problems. Although it has the largest number of subcorpora, CHA files do not have a standard format. This makes it difficult to extract information, as a program written to run on the entire corpus will not be able to read some files. Therefore, it’s likely that we were not able to extract several dative verbs that appeared in CHILDES. We chose to replace the SUBTLEX-US coipus (Brysbaert & New, 2009) with COCA (Davies, 2008) as our primary database of word frequencies. COCA provides lemmatized verb frequencies, such that the individual frequencies of tenses answer, answers, answering, and answered are coalesced into a single frequency for the verb answer. Furthermore, COCA has a more expansive database that allowed us to gather a larger collection of caused possession verbs for our analyses. However, while SUBTLEX-US is entirely a spoken word corpus, COCA is composed of several types of media: spoken, fiction, magazine, newspaper, and academic. This means COCA may not have the most representative frequencies of actual speech. And like SUBTLEX-US, COCA does not account for contextual frequencies for verbs with multiple meanings. For example, will is one of the highest frequency dative verbs in both SUBTLEX-US and COCA due to its use as an auxiliary verb (18a) rather than a dative

‘verb of future having’ (18b).

(18) a. She will go to the park. b. She willed the mansion to her grandchildren. 61

4.5.2 Future Directions The Tolerance Principle provides a new approach to our understanding of how language is represented in the brain. In the past, there was contention between connectionist approaches, which proposed that both regular and irregular forms were stored in associative memory (Rumelhart & McClelland, 1986), and computationalist approaches, which proposed that rule-following items were generated by rules, whereas exceptions were stored in associative memory (Pinker & Prince, 1988). In contrast, the Tolerance Principle both provides evidence for a rule-generated approach in which nested, or narrow-range rules, can be productive, and quantifies this procedure. The Tolerance Principle provides a new solution to our understanding of language acquisition and processing. With future analyses, we hope to see whether more evidence arises in favor of the Tolerance Principle as a model of language acquisition. It would be useful for others to test whether the Tolerance Principle can explain the acquisition of other argument structure alternations, such as the locative alternation. Locative verbs are used to describe the action of moving substances to or from surfaces and containers. Like the dative shift, the locative alternation features a shift in the syntactic structure of a sentence. Some locative verbs can be used in both constructions (19), whereas others can only be used in one (20-21).

(19) a. Jack sprayed paint onto the wall. b. Jack sprayed the wall with paint. (20) a. Amy poured water into the bowl. b. *Amy poured the bowl with water.

(21) a. * June covered a blanket over the baby. b. June covered the baby with a blanket.

As with the dative shift, children learning the locative alternation do not receive negative evidence, and therefore do not know which construction(s) a locative verb can 62

be used in. Perhaps the Sufficiency Principle, a derivation of the Tolerance Principle, can also be used to model how children use semantic information to learn the locative alternation. In the case of the dative verbs, we studied the semantic subclass caused possession. An analysis of the locative alternation would follow the semantic subclass of content-oriented verbs identified by Pinker (1989), which are more likely to appear in both constructions than container-oriented verbs. 63

References

Baker, C. L. (1979). Syntactic theory and the projection problem. Linguistic Inquiry, 10(4), 533-581.

Beavers, J. (2010). An aspectual analysis of ditransitive verbs of caused possession in english. Journal o f Semantics, 28(1), 1-54.

Bowerman, M. (1982). Reorganizational processes in lexical and syntactic development. In E. Wanner and L. R. Gleitman, eds., Language acquisition: The state of the art. New York: Cambridge University Press.

Brysbaert, M. & New, B. (2009). Moving beyond Kucera and Francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behavior Research Methods, 41 (4), 977-990.

Clahsen, H., Rothweiler, M., Woest, A., & Marcus, G. (1992). Regular and irregular in the acquisition of german noun plurals. Cognition, 45(3), 225-255.

Davies, Mark. (2008). The corpus of contemporary American English (COCA): 560 million words, 1990-present. Available online at https://www.english- corpora.org/coca/.

Gropen, J. H. (1989). Learning locative verbs: How universal linking rules constrain productivity. Massachusetts Institute of Technology, Cambridge, MA. Retrieved from http://hdl.handle.net/172Ll/14467

Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Affectedness and direct objects: The role of lexical semantics in the acquisition of verb argument structure. Cognition, 41,153-195. 64

Hovav, M. R. & Levin, B. (2008). The English dative alternation: The case for verb sensitivity. Journal o f Linguistics, 44(1), 129-167.

Levin, B. (1993). English verb classes and alternations: A preliminary investigation. Chicago, IL: The University of Chicago Press.

Levin, B. (2006). English object alternations: A unified account. Ms., Stanford University, Stanford, CA. Retrieved from http://www.stanford.edu/~bclevin/pubs.html

MacWhinney, B. (2000). The CHILDES Project: Tools for analyzing talk. Third Edition. Mahwah, NJ: Lawrence Erlbaum Associates.

Marcus, G. F. (1993). Negative evidence in language acquisition. Cognition, 46, 53-85.

Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition, 28(1), 73-193.

Pinker, S. (1989). Learnability and Cognition: The acquisition o f argument structure. MIT Press, Cambridge, MA.

Rumelhart, D. E., & McClelland, J. L. (1986). On learning the past tenses of 64nglish verbs. In McClelland, J. L., Rumelhart, D. E., and the PDP Research Group, editors, Parallel distributed processing: Explorations into the microstructure o f cognition. Volume 2: Psychological and biological models, pages 216-271. MIT Press, Cambridge, MA.

Szagun, G., Stumper, B., Sondag, N., & Franik, M. (2007). The acquisition of gender marking by young german-speaking children: Evidence for learning guided by phonological regularities. Journal o f Child Language, 34(3), 445-471. 65

Yang, C. (2016). The price o f linguistic productivity: How children learn to break the rules o f language. Cambridge, MA: The MIT Press.

i 66

Appendix A

Classification of Dative Verbs (Levin, 1993)

1. Alternating Verbs a) GIVE VERBS (“verbs that inherently signify acts of giving”) : feed, give, lease, lend, loan, pass, pay, peddle, refund, render, rent, repay, sell, serve, trade b) VERBS OF FUTURE HAVING (“commitments that a person will have something at some later point”): advance, allocate, allot, assign, award, bequeath, cede, concede, extend, grant, guarantee, issue, leave, offer, owe, promise, vote, will, yield c) BRING AND TAKE (“verbs of continuous causation of accompanied motion in a deictically specified direction”): bring, take (only) d. SEND VERBS (“verbs of sending”): forward, hand, mail, post, send, ship, slip, smuggle, sneak e. SLIDE VERBS: bounce, float, roll, slide f. CARRY VERBS (“verbs of continuous causation of accompanied motion in some manner”): carry, drag, haul, heave, heft, hoist, kick, lug, pull, push, schlep, shove, tote, tow, tug g. ? DRIVE VERBS: barge, bus, cart, drive, ferry, fly, row, shuttle, truck, wheel, wire (money) h. VERBS OF THROWING (“instantaneously causing ballistic motion”; most): bash, bat, bunt, catapult, chuck, flick, fling, flip, hit, hurl, kick, lob, pass, pitch, punt, shoot, shove, slam, slap, sling, throw, tip, toss i. VERBS OF TRANSFER OF A MESSAGE (“verbs of type of communicated message [differentiated by something like ‘illocutionary force’]”): ask, cite, ?pose, preach, quote, read, relay, show, teach, tell, write i. VERBS OF INSTRUMENT OF COMMUNICATION: cable, e-mail, fax, modem, netmail, phone, radio, relay, satellite, semaphore, sign, signal, telephone, telecast, telegraph, telex, wire, wireless 67

Appendix A (cont.)

2. Non-alternating to-dativc only a. * Primarily Latinate verbs belonging to some of the semantically plausible classes listed above: address, administer, broadcast, convey, contribute, delegate, deliver, denounce, demonstrate, describe, dictate, dispatch, display, distribute, donate, elucidate, exhibit, express, explain, explicate, forfeit, illustrate, introduce, narrate, portray, proffer, recite, recommend, refer, reimburse, remit, restore, return, sacrifice, submit, surrender, transfer, transport b. * SAY VERBS (“verbs of communication of propositions and prepositional attitudes”): admit, allege, announce, articulate, assert, communicate, confess, convey, declare, mention, propose, recount, repeat, report, reveal, say, state c. * VERBS OF MANNER OF SPEAKING: babble, bark, bawl, bellow, bleat, boom, bray, burble, cackle, call, carol, chant, chatter, chirp, cluck, coo, croak, croon, crow, cry, drawl, drone, gabble, gibber, groan, growl, grumble, grunt, hiss, holler, hoot, howl, jabber, lilt, lisp, moan, mumble, murmur, mutter, purr, rage, rasp, roar, rumble, scream, screech, shout, shriek, sing, snap, snarl, snuffle, splutter, squall, squawk, squeak, squeal, stammer, stutter, thunder, tisk, trill, trumpet, twitter, wail, warble, wheeze, whimper, whine, whisper, whistle, whoop, yammer, yap, yell, yelp, yodel d. * VERBS OF PUTTING WITH A SPECIFIED DIRECTION: drop, hoist, lift, lower, raise e. * VERBS OF FULFILLING (“X gives something to Y that Y deserves, needs, or is worthy o f’): credit, entrust, furnish, issue, leave, present, provide, serve, supply, trust 68

Appendix A (cont.)

3. Non-alternating DO Only a) Non-Alternating Double Object Only: accord, ask, bear, begrudge, bode, cost, deny, envy, flash (a glance), forbid, forgive, guarantee, issue (ticket, passport), refuse, save, spare, strike (a blow), vouchsafe, wish, write (check) b. * BILL VERBS: bet, bill, charge, fine, mulct, overcharge, save, spare, tax, tip, . undercharge, wager c. * APPOINT VERBS: acknowledge, adopt, appoint, consider, crown, deem, designate, elect, esteem, imagine, mark, nominate, ordain, proclaim, rate, reckon, report, want d. * DUB VERBS: anoint, baptize, brand, call, cliristen, consecrate, crown, decree, dub, label, make, name, nickname, pronounce, rule, stamp, style, term, vote e. * DECLARE VERBS: adjudge, adjudicate, assume, avow, believe, confess, declare, fancy, find, judge, presume, profess, prove, suppose, think, warrant 69

Appendix B

Classification of Caused Possession Verbs (Hovav & Levin, 2008)

(1) Dative verbs having only a caused possession meaning (a) Verbs that inherently signify acts of giving: give, hand, lend, loan, pass, rent, sell,... (b) Verbs of future having: allocate, allow, bequeath, grant, offer, owe, promise,

(c) Verbs of communication: tell, show, ask, teach, read, write, quote, cite,...

(2) Dative verbs having both caused motion and possession meanings (a) Verbs of sending (send-type verbs): forward, mail, send, ship,... (b) Verbs of instantaneous causation of ballistic motion (throw-type verbs): fling, flip, kick, lob, slap, shoot, throw, toss,... (c) Verbs of causation of accompanied motion in a deictically specified direction: bring, take (d) Verbs of instrument of communication: e-mail, fax, radio, wire, telegraph, telephone,...

Additional Caused Possession Verbs (Beavers, 2010)

(1) Dative verbs having only a caused possession meaning (a) Verbs of fulfilling: credit, present, entrust, supply, trust

(2) Dative verbs having both caused motion and possession meanings (a) Verbs of continuous causation of accompanied motion in some manner: carry, pull, push, schlep, lift, lower, haul 70

Appendix C

Dative Verbs Extracted from CHILDES

Dative Verbs in the DOC (N= 44): ask, assign, bet, bring, call, charge, consider, cost, drop, feed, find, give, grant, guarantee, hand, leave, lend, loan, mail, make, name, offer, owe, pass, pay, promise, pronounce, rate, read, roll, save, sell, send, serve, show, sing, take, teach, tell, throw, toss, trade, wish, write

Dative Verbs in the PPC (N= 64): address, assign, award, bounce, bring, carry, contribute, deliver, describe, drag, drive, drop, explain, feed, give, hand, haul, hit, hoist, introduce, kick, leave, lend, lift, loan, mail, mention, offer, owe, pass, pay, pitch, present, promise, prove, provide, pull, push, raise, read, recommend, rent, report, return, roll, say, sell, send, serve, ship, shoot, shout, show, sing, take, teach, tell, throw, toss, tow, trade, whisper, write, yell 71

Appendix D*

Caused Possession Verbs Ranked by Frequency

Yang Freq Dative Freq Dative Freq (2016) Replication Modification N=92 tf= 9 7 N= 148 e = 40 e = 40 e = 55 Top take 96012 will 103247 say 1915138 10 tell 87756 take 96012 will 824568 say 83066 tell 87756 take 670745 give 59457 say 83066 tell 388155 ask 24605 give 59457 give 384503 bring 16677 leave 27608 ask 284632 show 15690 ask 24605 leave 240482 hit 12636 bring 16677 show 208037 read 12234 show 15690 bring 174366 pay 11845 hit 12636 write 161824 N= 10 e = 1 N= 10 e = 1 N= 10 e = 1 Top send 9164 read 12234 provide 150879 20 shoot 8119 pay 11845 pay 133133 trust 8032 send 9164 read 114094 save 7946 shoot 8119 offer 106473 bet 7465 trust 8032 serve 99660 pull 7257 save 7946 send 96613 write 6467 bet 7465 report 88138 promise 6445 pull 7257 sell 87865 throw 6267 write 6467 pull 87243 explain 5663 promise 6445 pass 86184

Appendix D (cont.)

Caused Possession Verbs Ranked by Frequency

Yang Freq Dative Freq Dative Freq (2016) Replication Modification N= 92 N=91 N= 148 e = 40 e = 40 e = 55 Top adm it 3015 push 3313 present 49659 40 mention 2857 kick 3088 state 47977 cry 2596 admit 3015 reveal 42605 offer 2089 mention 2857 mention 39815 introduce 1772 cry 2596 sign 39418 feed 1708 offer 2089 vote 35224 repeat 1615 serve 1897 adm it 35143 deliver 1439 introduce 1772 refer 34317 drag 1314 feed 1708 announce 33305 spare 1021 repeat 1615 introduce 31345 O II il ll il o £ e = 10 o e = 9 Top scream 987 deliver 1439 express 30947 50 vote 909 drag 1314 contribute 30090 slip 896 spare 1021 demonstrate 29170 provide 887 scream 987 deliver 29169 describe 866 vote 909 feed 28494 guarantee 847 slip 896 extend 25969 yell 824 provide 887 promise 25720 confess 810 describe 866 repeat 22622 sneak 802 guarantee 847 recomm end 21707 snap 742 veil *24 propose 21691 O n i O II II II II

e = 16 O e = 21 Top recommend 714 confess 810 slip 21233 60 propose 664 sneak 802 declare 20550 shove 638 snap 742 kick 20092 surrender 602 recommend 714 trust 19482 lend 592 propose 664 issue 19300 shout 574 shove 638 cite 19011 flip 566 surrender 602 grant 18598 chuck 547 lend 592 hand 18231 express 537 shout 574 display 16806 toss 521 flip 566 drag 16685 N=60 e = 21 N=60 e = 20 N=60 e = 24 73

Appendix D (cont.)

Caused Possession Verbs Ranked by Frequency

Yang (2016) Freq Dative Freq Dative Freq N= 92 Replication Modification e = 40 N=91 N= 148 e = 40 e = 55 Top announce 498 chuck 547 quote 15415 70 reveal All express 537 pose 15230 communicatc 434 toss 521 illustrate 15091 slap 431 announce 498 transfer 14686 declarc 386 reveal All trade 14624 refer 374 communicate 434 toss 14565 extend 327 slap 431 supply 13399 quote 318 declare 386 assign 12836 subm it 282 refer 374 restore 12814 haul 275 extend 327 yield 12634 Top repay 265 quote 318 communicate 12179 80 whisper 262 submit 282 bet 12079 restore 253 haul 275 advance 12056 demonstrate 238 repay 265 guarantee 11430 holler 228 w hisper 262 distribute 10747 donate 203 restore 253 owe 10141 contribute 194 demonstrate 238 assert 10125 heave 179 pose 229 subm it 9507 dispatch 176 holler 228 exhibit 9459 yield 175 donate 203 portray 9202 m OO O oo o oo o m II II II II II e = 32 Top slam 158 contribute 194 slam 9076 90 recite 152 heave 179 pitch 8628 assign 138 dispatch 176 rent 8446 convey 136 yield 175 tip 8300 preach 122 slam 158 render 8288 forfeit 110 recite 152 adm inister 8135 render 103 assign 138 lend 7961 wager 103 convey 136 award 7666 smuggle 102 preach 122 convey 7529 squeal 94 forfeit 110 flip 7274 OO m N= 90 II N= 90 e = 37 JV= 90 e = 36 Appendix D (cont.)

Caused Possession Verbs Ranked by Frequency

Yang Freq Dative Freq Dative Freq (2016) Replication Modification N=92 N=91 N= 148 e = 40 e = 40 e = 55 Top dictate 89 render 103 ship 7264 100 adm inister 85 wager 103 signal 7171

—— smuggle 102 slap 7015

—- squeal 94 confess 6815

—— dictate 89 credit 6611 ---- adm inister 85 donate 6606 — — bash 84 post 6558 ——— — haul 6401 — --—— shove 6327 —— — — concede 6303 o O II N= 92 II N=91 N= 100 e = 39

Top — — — — sneak 5690 110 --—— — sacrifice 5641

—— —— spare 5492

— — — — articulate 5350

——— — broadcast 5138

— ——— preach 5063 ———— dictate 4935

— --— — surrender 4735

—— — ~ allege 4452

— —— ~ recount 4388 N= 110 e = 46

Top —— -- fling 3724 120 ——— — tug 3692 ——— — mail 3600

—— — denounce 3572

— .. —— phone 3243 ______recite 3210 .. —— bat 3132 — —--— flick 3097

___ _ -- — e-mail 3087

--—— — allocate 3038 OO II ll to o 75

Appendix D (cont.)

Caused Possession Verbs Ranked by Frequency

Yang Freq Dative Freq Dative Freq (2016) Replication Modification N=92 N=91 N= 148 e = 40 e = 40 e = 55

Top ——— — fax 2916

130 ——— — hurl 2631

—— — — wire 2582

— — — — furnish 2515

— — —— telephone 2489

—— — — dispatch 2479

—— — — lease 2452 — — —— accord 2422 — — —— chuck 2244

— — — — bill 2198 II

N= 130 O

Top —— — — repay 2196

140 — — ~ — smuggle 1884

——— — heave 1867 —— — — fine 1804 — — —— hoist 1676

—— — — relay 1650

— — ~ — sling 1614

— — —— bash 1578

— ——— tow 1530

—— —— narrate 1380 N= 140

148 — ... — — peddle 1122 — —— — lug 1118 —--—— reimburse 1083 — — — — cede 1023

— — —— forfeit 940

— — — — delegate 932

—— — — loan 923 OO II e = 55

* A comparison of caused possession verb frequencies from 1) Yang (2016) using the SUBTLEX-US corpus (Brysbaert & New, 2009), 2) our dative replication study using SUBTLEX-US, and 3) our dative modification study using COCA (Davies, 2008). Highlighted/bolded items indicate verbs that cannot be used in the double object construction (DOC). TV (the total number of verbs) and e (the number of verbs not expected in the DOC) are cumulative values.