Replace the focus verb with its FrameNet frame label (either provided in the Gold data, or tagged via the parser), and train embedding models(word2vec skip-gram Glove, and Fast-Text) on the resulting data. \
This yields joint embedding spaces that contain both common words and FrameNet frame embeddings.

Evaluation metrics: 
1. This lexical metric (lex) evaluates whether the frame embedding is similar to words within its frame and dissimilar to those without.\
$lex(f) = \sum_{w\in f}\frac{cos(E_w, E_f)}{|f|} - \sum_{w\notin f}^{k}\frac{cos(E_w, E_f)}{|k|}$
2. The structural similarity metric str is evaluating whether the frame embeddings are more similar around their neighbors as FrameNet also contains linking relations between frames (eg. used-by, uses), yielding a hierarchy of connected frames.\
$str(f) = \sum_{n\in N}\frac{cos(E_n, E_f)}{|N|} - \sum_{n\notin N}^{k}\frac{cos(E_n, E_f)}{|k|}$

Learning Frame Embeddings

Learn joint embedded representations for domains and lexical items. Then use the linear transformation between two domains as a mapping.

Frames ((Fillmore, 1982)): related systems of concepts, which is exchangeable with the term “domain” used in conceptual metaphor theory

FrameNet (Baker et al., 1998): semantic frames along with the set of lexical units that evoke them

Thus, we consider the transformation from one frame to another as a proxy for a conceptual metaphoric mapping.

University of Michigan - Ann Arbor

Conceptual metaphor theory (CMT): uses conceptual mappings between domains (conceptual structures that group related concepts) to generate linguistic metaphors.

Framework: Given a literal input sentence --> evokes a target domain --> generates metaphoric sentences that evoke desired corresponding source domain.

Eg. Three inputs: 
1. a literal input sentence (They argued against the contract)
2. the target domain evoked by the literal input (ARGUMENT)
3. the desired source domain (WAR) for the metaphorical sentence

The output is a metaphorical sentence that evokes the desired corresponding source domain (They fought against the contract).

How? Using an intended conceptual mapping (ARGUMENT - WAR)

Introduction (Metaphor Generation with Conceptual Mappings)

CM-Lex (Metaphor Generation with Conceptual Mappings)

Creating Parallel Data: 
1. Identify and mask metaphoric verbs(BERT-based metaphor classification model), (i.e “died” in The house where love had died) and replace them with infilling from a language model(pre-trained BERT).
2. Make sure at least 4 of 5 symbolic beams returned by COMET(Bosselut et al., 2019) are the same.
3. Additionally, tag each sentence with FrameNet frames.(Assume that the relation between the frames for the source and target domains reflects a metaphoric mapping.)

CM-BART (Metaphor Generation with Conceptual Mappings)

SBERT: Sentence Embeddings using Siamese BERT-Networks\
Generate sentence embeddings for:
1. the literal input $L$
2. the original gold metaphoric expression $M$
3. the generated output $G$

Distance from Gold Metaphor (dis): $\cos(M, G)$\
Relational distance (rel): $\cos(L, M) - cos(L, G)$\
Percentage of times the model produced the exact gold output.

Learn Before

Related

Learn After