Statistical Language Model

TextAnalysis provide following different Language Models

• MLE - Base Ngram model.
• Lidstone - Base Ngram model with Lidstone smoothing.
• Laplace - Base Ngram language model with Laplace smoothing.
• WittenBellInterpolated - Interpolated Version of witten-Bell algorithm.
• KneserNeyInterpolated - Interpolated version of Kneser -Ney smoothing.

APIs

To use the API, we first Instantiate desired model and then load it with train set

MLE(word::Vector{T}, unk_cutoff=1, unk_label="<unk>") where { T <: AbstractString}
        
Lidstone(word::Vector{T}, gamma:: Float64, unk_cutoff=1, unk_label="<unk>") where { T <: AbstractString}
        
Laplace(word::Vector{T}, unk_cutoff=1, unk_label="<unk>") where { T <: AbstractString}
        
WittenBellInterpolated(word::Vector{T}, unk_cutoff=1, unk_label="<unk>") where { T <: AbstractString}
        
KneserNeyInterpolated(word::Vector{T}, discount:: Float64=0.1, unk_cutoff=1, unk_label="<unk>") where { T <: AbstractString}
        
(lm::<Languagemodel>)(text, min::Integer, max::Integer)

Arguments:

• word : Array of strings to store vocabulary.

• unk_cutoff: Tokens with counts greater than or equal to the cutoff value will be considered part of the vocabulary.

• unk_label: token for unknown labels

• gamma: smoothing argument gamma

• discount: discounting factor for KneserNeyInterpolated

  for more information see docstrings of vocabulary

julia> voc = ["my","name","is","salman","khan","and","he","is","shahrukh","Khan"]

julia> train = ["khan","is","my","good", "friend","and","He","is","my","brother"]
# voc and train are used to train vocabulary and model respectively

julia> model = MLE(voc)
MLE(Vocabulary(Dict("khan"=>1,"name"=>1,"<unk>"=>1,"salman"=>1,"is"=>2,"Khan"=>1,"my"=>1,"he"=>1,"shahrukh"=>1,"and"=>1…), 1, "<unk>", ["my", "name", "is", "salman", "khan", "and", "he", "is", "shahrukh", "Khan", "<unk>"]))

julia> print(voc)
11-element Array{String,1}:
 "my"
 "name"
 "is"
 "salman"
 "khan" 
 "and" 
 "he" 
 "is"
 "shahrukh"
 "Khan"
 "<unk>"

# you can see "<unk>" token is added to voc 
julia> fit = model(train,2,2) #considering only bigrams

julia> unmaskedscore = score(model, fit, "is" ,"<unk>") #score output P(word | context) without replacing context word with "<unk>"
0.3333333333333333

julia> masked_score = maskedscore(model,fit,"is","alien")
0.3333333333333333
#as expected maskedscore is equivalent to unmaskedscore with context replaced with "<unk>"

Evaluation Method

score

used to evaluate the probability of word given context (P(word | context))

score(m::gammamodel, temp_lm::DefaultDict, word::AbstractString, context::AbstractString)

score(m::MLE, temp_lm::DefaultDict, word::AbstractString, context::AbstractString)

score(m::InterpolatedLanguageModel, temp_lm::DefaultDict, word::AbstractString, context::AbstractString)

Arguments:

1. m : Instance of Langmodel struct.
2. temp_lm: output of function call of instance of Langmodel.
3. word: string of word
4. context: context of given word

• In case of Lidstone and Laplace it apply smoothing and,

• In Interpolated language model, provide Kneserney and WittenBell smoothing

maskedscore

maskedscore(
    m::TextAnalysis.Langmodel,
    temp_lm::DataStructures.DefaultDict,
    word,
    context
) -> Float64

logscore

logscore(
    m::TextAnalysis.Langmodel,
    temp_lm::DataStructures.DefaultDict,
    word,
    context
) -> Float64

entropy

entropy(
    m::TextAnalysis.Langmodel,
    lm::DataStructures.DefaultDict,
    text_ngram::AbstractVector
) -> Float64

perplexity

perplexity(
    m::TextAnalysis.Langmodel,
    lm::DataStructures.DefaultDict,
    text_ngram::AbstractVector
) -> Float64

Preprocessing

For Preprocessing following functions:

everygram(seq::Vector{T}; min_len::Int=1, max_len::Int=-1)where { T <: AbstractString}

julia> seq = ["To","be","or","not"]
julia> a = everygram(seq,min_len=1, max_len=-1)
 10-element Array{Any,1}:
  "or"          
  "not"         
  "To"          
  "be"                  
  "or not" 
  "be or"       
  "be or not"   
  "To be or"    
  "To be or not"

padding_ngram(word::Vector{T}, n=1; pad_left=false, pad_right=false, left_pad_symbol="<s>", right_pad_symbol ="</s>") where { T <: AbstractString}

julia> example = ["1","2","3","4","5"]

julia> padding_ngram(example,2,pad_left=true,pad_right=true)
 6-element Array{Any,1}:
  "<s> 1" 
  "1 2"   
  "2 3"   
  "3 4"   
  "4 5"   
  "5 </s>"

Vocabulary

Struct to store Language models vocabulary

checking membership and filters items by comparing their counts to a cutoff value

It also Adds a special "unknown" tokens which unseen words are mapped to

julia> using TextAnalysis
julia> words = ["a", "c", "-", "d", "c", "a", "b", "r", "a", "c", "d"]11-element Vector{String}:
 "a"
 "c"
 "-"
 "d"
 "c"
 "a"
 "b"
 "r"
 "a"
 "c"
 "d"
julia> vocabulary = Vocabulary(words, 2)Vocabulary(Dict("<unk>" => 1, "c" => 3, "a" => 3, "d" => 2), 2, "<unk>", ["a", "c", "-", "d", "c", "a", "b", "r", "a", "c", "d", "<unk>"])
julia> # lookup a sequence or words in the vocabulary
       
       word = ["a", "-", "d", "c", "a"]5-element Vector{String}:
 "a"
 "-"
 "d"
 "c"
 "a"
julia> lookup(vocabulary ,word)5-element Vector{String}:
 "a"
 "<unk>"
 "d"
 "c"
 "a"