On the Benefits of Enriching the Penn Parsed Corpora of Historical English to Study Change at the Syntax-Semantics Interface: Introducing the Roots & Results (RoRe) Database

1 Context and motivation

English has undergone substantial changes in the way complex events such as resultative and motion expressions are mapped onto the syntactic structure. In terms of Talmy (2000)’s typology, English shows diachronically consistent characteristics of a satellite-framing language in mapping the end state or path onto a grammatical element outside the verb. The inventory of satellites has changed dramatically, however. Old English (OE) and early Middle English (ME) employed an elaborate system of prefixes, which developed out of adverbial preverbs, to mark resultativity and telicity of the predicate (cf. Hiltunen, 1983; Los et al., 2012; Thim, 2012, among others). These prefixes could be inseparable, as in (1) in which the prefix a- moved together with the verb, or separable as in (2) in which the particle ut- is stranded by verb movement. Inseparable prefixes were lost as a productive phenomenon after OE, while the separable prefixes syntactically developed into postverbal particles.

1. 1. (1)

2. 1. 1. Ða

      2. the

      1. þry

      2. three

      1. oðre

      2. other

      1. godspelleras

      2. evangelists

      1. a-writon

      2. up-wrote

      1. heora

      2. their

      1. godspell

      2. gospel,

      1. be

      2. by

      1. Cristes

      2. Christ’s

      1. menniscnysse

      2. humanity

   2. 1. “The three other evangelists wrote down their gospel about Christ’s humanity”

      2. (coaelhom, ÆHom & 1:17.7, adapted from Elenbaas, 2007, 114)

1. 1. (2)

2. 1. 1. þa

      2. then

      1. sticode

      2. stuck

      1. him

      2. him

      1. mon

      2. one

      1. þa

      2. the

      1. eagan

      2. eyes

      1. ut

      2. out

   2. 1. “then his eyes were gouged out”

      2. (coorosiu,Or 4:5.90.13.1822, Elenbaas, 2007, 131)

These developments took place in the context of extensive language contact with French following the Norman Conquest in 1066. The influx of French vocabulary, including a large number of verbs (Durkin, 2014; Percillier et al., 2024), introduced lexical items originating in a language that, as a verb-framing language, differed typologically from English in the expression of motion and result by encoding these meanings in the verb itself (Troberg and Burnett, 2017; Troberg and Leung, 2021). Compare in this regard the PDE phrasal verb move in to its synonym enter, which was directly copied from French entrer in ME.

The changes in complex event expression thus show interaction across several linguistic domains: morphology, syntax and lexical semantics and are at the same time modulated by language contact effects. There is significant research on each of these respective domains, from various analytical perspectives (space prevents a full discussion, but see Elenbaas, 2007; Fanego, 2012; Huber, 2017; Los et al., 2012; Martín Arista, 2012; McFadden, 2015; Rodríguez-Puente, 2018; van Gelderen, 2018 for some influential and recent work). However, how these phenomena relate to each other is still poorly understood. This is also noted by Thim (2012, 157), who, after an extensive review of earlier literature, comes to the conclusion that “what is missing is a treatment of the different types of word formation in a coherent and systematically related way” to understand how morphology, syntax, semantics and etymology interoperate. Arguably, a reason for this is that up to now there is no good resource available which allows studying these domains comprehensively from a diachronic perspective.

Existing resources and methodologies are often optimized for one of these domains. Historical dictionaries such as the Oxford English Dictionary (OED) (Proffitt, 2015), the Middle English Dictionary (MED) (McSparran et al., 2001), and the Bosworth Toller Anglo-Saxon dictionary (BT) (Bosworth et al., 2010) provide detailed information on etymology and morphological structure, but they cannot be used in a way that allows a quantitative analysis of, for instance, prefix-verb combinations, their relation to the syntactic structure or correlations with syntactically relevant verb classes. Conversely, corpus-based and syntactic studies have frequently focused on individual constructions, such as verb-particle combinations, individual lexical items, such as the behaviour of one verb, or only a token-based approach to an individual prefix. Furthermore, the studies often employ annotation practices and datasets tailored to their specific aims.

As a result, findings are not always easily comparable or interoperable across studies, which complicates broader generalizations about the diachronic development of English event structure. The absence of shared annotation schemes and openly available datasets also makes it difficult to combine lexical-semantic, morphological, and syntactic information systematically or to replicate analyses across larger bodies of data. In practice, this means that (substantial) manual curation and (re)annotation are required before quantitative or multifactorial analyses become possible. A fuller understanding of the historical development of motion and resultative constructions therefore benefits from and in fact requires resources that can relate morphology, lexical semantics, syntax, and etymology within a unified and reusable empirical framework.

This paper will introduce and evaluate the Roots & Results (RoRe) dataset, which was specifically developed to fill this empirical and methodological gap. It is the first exhaustive resource available for the full inventory of English verbs which combines the levels of annotation relevant to study change in the expression of resultativity and motion (prefixation, lexical semantics and etymology), fitting well with other current research which aims to quantitatively assess the syntactic and morphological expression of motion and resultativity in early Indo-European languages, such as Rainsford and Piccione (2025) on French and Italian motion expressions and Farina and McGillivray (2025) on preverbs in Ancient Greek and Latin.

The dataset is fully compatible with the Penn Parsed Corpora of English, the gold standard syntactic corpora in the field of historical linguistics. The original versions of the corpora were not lemmatized, making it difficult to enrich them in a time-efficient way, as a consequence of the significant spelling variation in historical texts. Recently, a lemmatized version of the York-Toronto-Helsinki Parsed Corpus of Old English Prose (YCOE) (Taylor et al., 2024) has become available, as well as a lemmatizer (Percillier, 2018; Trips & Percillier, 2020) dedicated to the lemmatization of the collected parsed corpora for ME, the Penn Parsed Corpus of Middle English 2 (PPCME2) (Kroch et al., 2000), the Parsed Linguistic Atlas of Early Middle English (PLAEME) (Truswell et al., 2018) and the Parsed Corpus of Middle English Poetry (PCMEP) (Zimmermann, 2018). This development allows the corpora to be enriched with additional annotation derived from the lexicographic sources, so that it becomes possible to answer questions relating to changes at the syntax-semantics interface from a contact perspective in a uniform and replicable way.

2 Dataset description

The dataset comprises stand-off lemma-based annotation for all uniquely occurring verb lemmas occurring in the Penn Parsed Corpora of English. The dataset comprises annotation for 4562 OE, 4262 ME and 3839 EModE lemmas. The lemmas are matched wherever possible to entries in the BT, MED, and the OED and subsequently annotated for prefixation, etymology of the prefix and verb and lexical semantics in terms of Levin & Rappaport Hovav’s (1995, 2019) Manner/Result complementarity framework. We here provide a brief overview of our annotation scheme. A more detailed annotation manual is included in the repository.

3 Technical implementation

The lemma-based annotation allows semi-automatic enrichment of the Penn Parsed Corpora of English, making it possible to study diachronic changes in expression of argument and event structure in relation to the lexical semantics of verbs. At the time of writing, only the OE and ME corpora have publicly available lemmatization, although the procedure described here has also been used to enrich a non-publicly available lemmatized version of the EModE corpus. The dataset comes with the RoRe Annotator tool, a Python application which can be run via the command line, which allows users to enrich their versions of the corpora themselves via a user-friendly guided prompt. The RoRe Annotator currently supports annotation of corpora in .psd and .psdx (XML) format.

The annotation procedure is described in Figure 1. After the user has selected the time period (OE, ME or EModE) and the format of the corpus files (.psd or .psdx), the program loops through the files to find POS-tags beginning with V and matches the associated lemma with the corresponding lemma in the RoRe data files. Whenever the match is not unique, the lemma ID will additionally be retrieved, if available³. If this results in a unique match, Manner/Result and Result Type attributes are added to the corpus, and the corresponding annotation is added as values. If no match can be made with a lemma at all, Manner/Result and RType attributes will be added, but they will not have a value, ensuring that they will be returned when the corpus is queried for these attributes. If no unique lemma ID match can be established, the Manner/Result and RType annotation for both lemmas will be added, separated by “|”.

Figure 1

The annotation procedure for the .psd files relies on the lemmatization format used in the BASICS lemmatizer (Percillier, 2018), which enriches the .psd corpus format with attributes demarcated by “@”. This is illustrated in (9), where (9a) shows the bare corpus node for the verb setting and (9b) shows the enriched corpus nodes with the attributes l for lemma, m for MEDID and e for etymology.

1. 1. (9)

   1. a.

2. 1. 1. (VAG settyng)

1. 1.  

   1. b.

2. 1. 1. (VAG settyng@l=setten@m=39654@e=nonfrench@)

The YCOE corpus is lemmatized according to a different format (verb form-lemma), as in (10a). The RoRe Annotator has a built-in function which automatically converts the verb form-lemma format to the BASICS verb form@l=lemma@ format, as in (10b), to ensure clear demarcation of the various annotation levels, as well as consistency across the various corpora to facilitate the querying process.

1. 1. (10)

   1. a.

2. 1. 1. (VB^D secene-secan)

1. 1.  

   1. b.

2. 1. 1. (VB^D secene@l=secan@)

Whenever there is a match between the attribute l in the corpus and the RoRe data files, the verb nodes are enriched with the following attributes: pre for prefix, pre_et for prefix etymology, barev for bare verb, barev_et for bare verb etymology, mr for Manner/Result and rtype for Result Type. The fully annotated string for the example in (9) is (11) and the fully annotated string for the example in (10) is (12).

1. 1. (11)

2. 1. 1. (VAG settyng@l=setten@m=39654@e=nonfrench@pre=n.a.@pre_et=n.a.

      2. @barev=setten@barev_et=native@mr=Amb_light@rtype=cos@)

1. 1. (12)

2. 1. 1. (VB^D secene@l=secan@pre=n.a.@pre_et=@barev=secan@barev_et

      2. @=@mr=Manner@rtype=@)

This annotated structure can be queried using CorpusSearch2 (Randall, 2004) and allows the user to look for specific verbs, verbs of one etymology, prefixed verbs or a particular class of lexical semantics or a combination of these attributes by specifying the level of annotation using the exists function, e.g. (*pre=ge-@* exists) to find all verbs that begin with the prefix ge-.

4 The use of prefixes in the history of English

We illustrate the functionality of our dataset with a case study on changes in the English prefixation system. English used to have a verbal prefix system to mark completive aspect and resultativity, such as ge-, be- and a- (although their function may vary, see Elenbaas, 2007; Hiltunen, 1983; Thim, 2012). Most of these prefixes are lost after early ME, although some survive with marginal productivity, such as be- and for-. The substantial number of French and later in the Renaissance era of Latin verbs have been argued to introduce a novel system of prefixes based on Romance preverbs, such as mis-, dis-/des- and contra-.

This raises the question if verbal prefixation as a morphological mechanism was lost from the language and, more specifically, if the Romance prefixes become productive to the same degree and with the same function as native prefixes. Thim (2012) notes that previous work in this domain is usually only concerned with either the loss of the Germanic aspectual prefixes or the rise of Romance prefixes, but that the two changes are rarely discussed in relation to each other. Thim (2012) presents a type-based analysis of the prefix inventory, which reveals a substantial increase from OE to ME. This leads Thim (2012, 156) to posit that “it can be safely concluded that prefixation as a native word formation type remained productive in Middle English.”

However, Thim (2012) is not specific on his measure of productivity, nor does he consider the combinatorial properties of the prefixes and verbs in relation to their etymologies. This is relevant from a contact perspective, because hybrid formations are an important indicator of morphological copying, rather than only lexical copying (e.g. Gardani, 2018). We follow Dalton-Puffer (1996), who studies the influence of French on English morphology in domains other than verbal prefixation, in considering two types of hybrid forms, illustrated in (13). Especially the second hybrid type (13b) is indicative of morphological copying, as it indicates the reanalysis of a morpheme as a meaningful component.

1. 1. (13)

   1. a.

2. 1. 1. Romance base + Germanic affix

1. 1.  

   1. b.

2. 1. 1. Germanic affix + Romance affix

Our dataset allows us to immediately extract all the verbs from the enriched corpora together with their prefix, the etymology of the prefix, the verb stem, and the etymology of the verb stem, which we quantify in terms of types and tokens. We collated the O1 and O2 periods from YCOE (based on manuscript dates) in one period “early OE” (850–950) and the O3 and O4 periods in one period “late OE” (950–1150). Similarly, M1 and M2 from the ME corpora are collated into one period “early ME” (1150–1350) and M3 and M4 in one period “lateME” (1350–1500). The subperiodization for EModE is taken directly from the corpus: E1 (1500–1570), E2 (1570–1640), E3 (1640–1710). We first of all present our findings for prefixation across French and native verbs diachronically, both on type and token level, in Table 1 and Figure 2.

Figure 2

The data show that the frequency of prefixation reduces significantly from OE to early ME. Most interesting for our purposes is the period from early ME onwards, when French verbs establish themselves in the language. Prefixed French and native verbs occur at comparable frequencies on a token level, but prefixed French verbs are less frequent on a type level. The frequency of prefixed types drops with both French and native verbs until the early EModE period, when they stabilize. Crucially, French verbs remain more frequently prefixed on a token level. These findings already allow us to refine Thim’s (2012) confident conclusion that prefixation remained productive. While the type inventory of prefixes increases towards Modern English, the verb types that can function as a base for these verbs reduces significantly, suggesting that these prefixes are not productive.

To further gauge the productivity of the various prefixes we introduce the etymology of the prefix as an additional variable and calculate the relative frequencies of native and French prefixes on the subset of prefixed verbs. The data in Figure 3 show a near categorical separation between prefix etymology and the etymology of the verb base: French prefixes combine with French verbs and native prefixes combine with native verbs. There are two exceptions, which we briefly discuss in turn.

Figure 3

In early ME we find a substantial number of native prefixes attached to a French base, in terms of types as well as tokens. Closer inspection of the data reveals that the majority of these cases, illustrated in (14), are participles prefixed with ge-/i-, which at this stage is typically analysed as an optional inflectional marker (cf. Elenbaas, 2007; Martín Arista, 2012; McFadden, 2015; Thim, 2012).

1. 1. (14)

2. 1. 1. today

      2. today

      1. Þou

      2. you

      1. scalt

      2. shall

      1. ben

      2. be

      1. icrounet.

      2. crowned

      1. biforn

      2. before

      1. Þe

      2. the

      1. king

      2. king

      1. of

      2. of

      1. heuene.

      2. heaven

   2. 1. “today you will be crowned before the king of heaven.” (m2_tr323bt, 288)

We also find some new formations on the basis of the native prefix be-, although this seems to be restricted to a small number of verb types, including begylen “to beguile” on the basis of French guiler, from the noun guile “deceit, trickery”, and betraien “to betray”, on the basis of French traïr “to betray”, potentially stressing the affectedness of the patient associated with the verb. These verbs continue to exist in EModE, although the decompositionality of betraien is lost, as the verb base traien/tray no longer occurs in EModE. At the same time, new formations of a native prefix with a French base emerge, such as misconceive or overcharge. It is clear that the functionality of these prefixes is different: they are not aspectual or resultative markers, but their function is more akin to adverbial modifiers, specifying the way an action is carried out.

The second exception is the combination of French prefixes and native verb bases in EModE. Interestingly, such combinations hardly occur in late ME, despite extensive language contact up until this point. We report one type which occurs only once, suggesting it is based on an analogical formation: enspiren “to inquire” which is derived from the French prefix en- and the native base spiren “to inquire”. The remaining tokens belong to the type reneuen “to renew”, which seems to be the first hybrid form that we can trace in the corpora. The number of hybrid combinations increases in EModE, but the number of types remains relatively low. Most tokens feature the prefix re-, although we also find several new formations based on en-, such as engrave and enrich. These observations corroborate the conclusion in Dalton-Puffer (1996) that the Romance morphology is not productive in ME. At the same time, it shows the relevance of studying contact effects also after language contact has ceased or significantly decreased.

5 Implications/Applications

The present case study allowed us to refine earlier observations about the continued productivity of verbal prefixation in English (cf. Thim, 2012) by distinguishing more systematically between the etymology of both prefix and verb and by tracing their diachronic distribution from both a type and token level, instead of focusing on type frequency of the prefix alone.

This is only one example of the questions that can be addressed on the basis of the RoRe dataset. A further avenue concerns the development of prefixation in interaction with verbs of different ontological classes and in relation to other satellites, such as particles and potentially also preposition phrases. Here, the compatibility with the Penn Parsed corpora is particularly valuable, as this allows the extraction of individual verbs or verb classes in various syntactic structures without the need for extensive manual annotation. In fact, Struik and Kaltenbach (2026) show on the basis of this dataset that especially French Result verbs, which predominantly encode an end state, do not combine with native particles, and our case study suggests that they also do not combine with native prefixes. Together, these observations show that copies from French retain their verb-framing properties.

While the RoRe dataset was developed within the scope of a research project focusing on change in the domains of motion and resultativity in a contact situation, the annotation framework itself was designed to be broadly applicable beyond these domains, for instance, to study the effects of language contact or lexical semantics on other (morpho)syntactic phenomena. Furthermore, the lemma-based annotation pipeline is readily extendible to any lemma-based phenomenon, for instance, suffixation as an immediate counterpart to prefixation or other semantic features of verbs, such as verb classes (as in Levin, 1993, cf. Percillier, 2018). As such, the RoRe dataset illustrates the value of combining parsed corpora with detailed linguistic annotation in a transparent and reusable way to study complex questions regarding diachronic syntactic and semantic change.

Author Contributions

Tara Struik – Conceptualization, methodology, data curation, validation, formal analysis, project administration, software, writing – original draft/review & editing; Lena Kaltenbach – Conceptualization, methodology, writing – original draft/review & editing; Susanne Lang – Data curation, writing - original draft.